From c250bc560cdde89ed7459f8c5ce54dc42f05794d Mon Sep 17 00:00:00 2001 From: tolk-vm Date: Tue, 11 Feb 2025 22:49:26 +0400 Subject: [PATCH] nullable, with tensors, no smart casts --- crypto/smartcont/tolk-stdlib/common.tolk | 25 +- crypto/smartcont/tolk-stdlib/lisp-lists.tolk | 9 +- crypto/smartcont/tolk-stdlib/tvm-dicts.tolk | 34 +- tolk-tester/tests/a10.tolk | 2 +- tolk-tester/tests/assignment-tests.tolk | 4 +- tolk-tester/tests/c2.tolk | 2 +- tolk-tester/tests/cells-slices.tolk | 11 +- tolk-tester/tests/generics-1.tolk | 8 +- tolk-tester/tests/indexed-access.tolk | 27 +- tolk-tester/tests/inference-tests.tolk | 6 +- tolk-tester/tests/invalid-generics-1.tolk | 2 +- tolk-tester/tests/invalid-generics-13.tolk | 11 + tolk-tester/tests/invalid-mutate-18.tolk | 10 + tolk-tester/tests/invalid-mutate-19.tolk | 10 + tolk-tester/tests/invalid-mutate-20.tolk | 13 + tolk-tester/tests/invalid-typing-14.tolk | 14 + tolk-tester/tests/invalid-typing-15.tolk | 13 + tolk-tester/tests/invalid-typing-16.tolk | 10 + tolk-tester/tests/invalid-typing-17.tolk | 10 + tolk-tester/tests/invalid-typing-18.tolk | 16 + tolk-tester/tests/logical-operators.tolk | 9 +- tolk-tester/tests/null-keyword.tolk | 51 ++- tolk-tester/tests/nullable-tensors.tolk | 404 ++++++++++++++++ tolk-tester/tests/nullable-types.tolk | 109 +++++ tolk-tester/tests/use-before-declare.tolk | 4 +- tolk-tester/tests/var-apply.tolk | 41 ++ tolk/abscode.cpp | 6 +- tolk/ast-from-tokens.cpp | 44 +- tolk/ast-replacer.h | 4 + tolk/ast-replicator.h | 8 + tolk/ast-stringifier.h | 4 + tolk/ast-visitor.h | 4 + tolk/ast.cpp | 4 + tolk/ast.h | 28 ++ tolk/codegen.cpp | 4 +- tolk/generics-helpers.cpp | 156 +++---- tolk/generics-helpers.h | 40 +- tolk/pipe-ast-to-legacy.cpp | 456 +++++++++++++++---- tolk/pipe-calc-rvalue-lvalue.cpp | 12 + tolk/pipe-check-rvalue-lvalue.cpp | 12 + tolk/pipe-constant-folding.cpp | 11 + tolk/pipe-infer-types-and-calls.cpp | 210 ++++++--- tolk/pipe-optimize-boolean-expr.cpp | 5 + tolk/pipe-refine-lvalue-for-mutate.cpp | 2 + tolk/tolk.h | 2 +- tolk/type-system.cpp | 159 ++++--- tolk/type-system.h | 71 +-- 47 files changed, 1649 insertions(+), 448 deletions(-) create mode 100644 tolk-tester/tests/invalid-generics-13.tolk create mode 100644 tolk-tester/tests/invalid-mutate-18.tolk create mode 100644 tolk-tester/tests/invalid-mutate-19.tolk create mode 100644 tolk-tester/tests/invalid-mutate-20.tolk create mode 100644 tolk-tester/tests/invalid-typing-14.tolk create mode 100644 tolk-tester/tests/invalid-typing-15.tolk create mode 100644 tolk-tester/tests/invalid-typing-16.tolk create mode 100644 tolk-tester/tests/invalid-typing-17.tolk create mode 100644 tolk-tester/tests/invalid-typing-18.tolk create mode 100644 tolk-tester/tests/nullable-tensors.tolk create mode 100644 tolk-tester/tests/nullable-types.tolk diff --git a/crypto/smartcont/tolk-stdlib/common.tolk b/crypto/smartcont/tolk-stdlib/common.tolk index 5311ec2f..826c431e 100644 --- a/crypto/smartcont/tolk-stdlib/common.tolk +++ b/crypto/smartcont/tolk-stdlib/common.tolk @@ -154,16 +154,16 @@ fun getCurrentBlockLogicalTime(): int /// Returns the value of the global configuration parameter with integer index `i` as a `cell` or `null` value. @pure -fun getBlockchainConfigParam(x: int): cell +fun getBlockchainConfigParam(x: int): cell? asm "CONFIGOPTPARAM"; /// Returns the persistent contract storage cell. It can be parsed or modified with slice and builder primitives later. @pure -fun getContractData(): cell +fun getContractData(): cell // todo null? asm "c4 PUSH"; /// Sets `cell` [c] as persistent contract data. You can update persistent contract storage with this primitive. -fun setContractData(c: cell): void +fun setContractData(c: cell): void // todo null? asm "c4 POP"; /// Retrieves code of smart-contract from c7 @@ -291,7 +291,7 @@ fun calculateSliceSizeStrict(s: slice, maxCells: int): (int, int, int) /// otherwise the returned value is one plus the maximum of depths of cells referred to from [c]. /// If [c] is a `null` instead of a cell, returns zero. @pure -fun getCellDepth(c: cell): int +fun getCellDepth(c: cell): int // todo null? (see comment) asm "CDEPTH"; /// Returns the depth of `slice` [s]. @@ -346,7 +346,7 @@ fun assertEndOfSlice(self: slice): void /// Loads the next reference from the slice. @pure -fun loadRef(mutate self: slice): cell +fun loadRef(mutate self: slice): cell // todo null? asm( -> 1 0) "LDREF"; /// Preloads the next reference from the slice. @@ -417,12 +417,12 @@ fun getLastBits(self: slice, len: int): slice /// Loads a dictionary (TL HashMapE structure, represented as TVM cell) from a slice. /// Returns `null` if `nothing` constructor is used. @pure -fun loadDict(mutate self: slice): cell +fun loadDict(mutate self: slice): cell? // todo cell? asm( -> 1 0) "LDDICT"; /// Preloads a dictionary (cell) from a slice. @pure -fun preloadDict(self: slice): cell +fun preloadDict(self: slice): cell // todo cell? asm "PLDDICT"; /// Loads a dictionary as [loadDict], but returns only the remainder of the slice. @@ -433,12 +433,12 @@ fun skipDict(mutate self: slice): self /// Loads (Maybe ^Cell) from a slice. /// In other words, loads 1 bit: if it's true, loads the first ref, otherwise returns `null`. @pure -fun loadMaybeRef(mutate self: slice): cell +fun loadMaybeRef(mutate self: slice): cell? asm( -> 1 0) "LDOPTREF"; /// Preloads (Maybe ^Cell) from a slice. @pure -fun preloadMaybeRef(self: slice): cell +fun preloadMaybeRef(self: slice): cell? asm "PLDOPTREF"; /// Loads (Maybe ^Cell), but returns only the remainder of the slice. @@ -497,13 +497,13 @@ fun storeBool(mutate self: builder, x: bool): self /// Stores dictionary (represented by TVM `cell` or `null`) into a builder. /// In other words, stores a `1`-bit and a reference to [c] if [c] is not `null` and `0`-bit otherwise. @pure -fun storeDict(mutate self: builder, c: cell): self +fun storeDict(mutate self: builder, c: cell?): self asm(c self) "STDICT"; /// Stores (Maybe ^Cell) into a builder. /// In other words, if cell is `null`, store '0' bit; otherwise, store '1' and a ref to [c]. @pure -fun storeMaybeRef(mutate self: builder, c: cell): self +fun storeMaybeRef(mutate self: builder, c: cell?): self asm(c self) "STOPTREF"; /// Concatenates two builders. @@ -687,7 +687,8 @@ fun loadMessageFlags(mutate self: slice): int /// Effectively, it's `msgFlags & 1` (the lowest bit present). @pure fun isMessageBounced(msgFlags: int): bool - asm "2 PUSHINT" "MODR"; + asm "1 PUSHINT" "AND"; + // asm "2 PUSHINT" "MODR"; // todo uncomment after no diff in verified /// Skip 0xFFFFFFFF prefix (when a message is bounced). @pure diff --git a/crypto/smartcont/tolk-stdlib/lisp-lists.tolk b/crypto/smartcont/tolk-stdlib/lisp-lists.tolk index 0cb17841..af8b6bd7 100644 --- a/crypto/smartcont/tolk-stdlib/lisp-lists.tolk +++ b/crypto/smartcont/tolk-stdlib/lisp-lists.tolk @@ -14,17 +14,18 @@ fun createEmptyList(): tuple /// Adds an element to the beginning of lisp-style list. /// Note, that it does not mutate the list: instead, it returns a new one (it's a lisp pattern). @pure -fun listPrepend(head: X, tail: tuple): tuple +fun listPrepend(head: X, tail: tuple?): tuple asm "CONS"; /// Extracts the head and the tail of lisp-style list. @pure -fun listSplit(list: tuple): (X, tuple) +fun listSplit(list: tuple): (X, tuple?) asm "UNCONS"; /// Extracts the tail and the head of lisp-style list. +/// After extracting the last element, tuple is assigned to null. @pure -fun listNext(mutate self: tuple): X +fun listNext(mutate self: tuple?): X asm( -> 1 0) "UNCONS"; /// Returns the head of lisp-style list. @@ -34,5 +35,5 @@ fun listGetHead(list: tuple): X /// Returns the tail of lisp-style list. @pure -fun listGetTail(list: tuple): tuple +fun listGetTail(list: tuple): tuple? asm "CDR"; diff --git a/crypto/smartcont/tolk-stdlib/tvm-dicts.tolk b/crypto/smartcont/tolk-stdlib/tvm-dicts.tolk index 5c436239..41320b85 100644 --- a/crypto/smartcont/tolk-stdlib/tvm-dicts.tolk +++ b/crypto/smartcont/tolk-stdlib/tvm-dicts.tolk @@ -24,7 +24,7 @@ fun dictIsEmpty(self: cell): bool @pure -fun iDictGet(self: cell, keyLen: int, key: int): (slice, bool) +fun iDictGet(self: cell, keyLen: int, key: int): (slice, bool) // todo keep not null? or make another interface? asm(key self keyLen) "DICTIGET" "NULLSWAPIFNOT"; @pure @@ -50,7 +50,7 @@ fun sDictSet(mutate self: cell, keyLen: int, key: slice, value: slice): void @pure -fun iDictSetRef(mutate self: cell, keyLen: int, key: int, value: cell): void +fun iDictSetRef(mutate self: cell, keyLen: int, key: int, value: cell): void // todo null? asm(value key self keyLen) "DICTISETREF"; @pure @@ -81,7 +81,7 @@ fun uDictSetIfExists(mutate self: cell, keyLen: int, key: int, value: slice): bo @pure -fun iDictGetRef(self: cell, keyLen: int, key: int): (cell, bool) +fun iDictGetRef(self: cell, keyLen: int, key: int): (cell, bool) // todo keep not null? asm(key self keyLen) "DICTIGETREF" "NULLSWAPIFNOT"; @pure @@ -94,15 +94,15 @@ fun sDictGetRef(self: cell, keyLen: int, key: slice): (cell, bool) @pure -fun iDictGetRefOrNull(self: cell, keyLen: int, key: int): cell +fun iDictGetRefOrNull(self: cell, keyLen: int, key: int): cell? asm(key self keyLen) "DICTIGETOPTREF"; @pure -fun uDictGetRefOrNull(self: cell, keyLen: int, key: int): cell +fun uDictGetRefOrNull(self: cell, keyLen: int, key: int): cell? asm(key self keyLen) "DICTUGETOPTREF"; @pure -fun sDictGetRefOrNull(self: cell, keyLen: int, key: slice): cell +fun sDictGetRefOrNull(self: cell, keyLen: int, key: slice): cell? asm(key self keyLen) "DICTGETOPTREF"; @@ -120,7 +120,7 @@ fun sDictDelete(mutate self: cell, keyLen: int, key: slice): bool @pure -fun iDictSetAndGet(mutate self: cell, keyLen: int, key: int, value: slice): (slice, bool) +fun iDictSetAndGet(mutate self: cell, keyLen: int, key: int, value: slice): (slice, bool) // todo keep not null? or another interface? asm(value key self keyLen) "DICTISETGET" "NULLSWAPIFNOT"; @pure @@ -133,16 +133,16 @@ fun sDictSetAndGet(mutate self: cell, keyLen: int, key: slice, value: slice): (s @pure -fun iDictSetAndGetRefOrNull(mutate self: cell, keyLen: int, key: int, value: cell): cell +fun iDictSetAndGetRefOrNull(mutate self: cell, keyLen: int, key: int, value: cell): cell? asm(value key self keyLen) "DICTISETGETOPTREF"; @pure -fun uDictSetAndGetRefOrNull(mutate self: cell, keyLen: int, key: int, value: cell): cell +fun uDictSetAndGetRefOrNull(mutate self: cell, keyLen: int, key: int, value: cell): cell? asm(value key self keyLen) "DICTUSETGETOPTREF"; @pure -fun iDictDeleteAndGet(mutate self: cell, keyLen: int, key: int): (slice, bool) +fun iDictDeleteAndGet(mutate self: cell, keyLen: int, key: int): (slice, bool) // todo keep not null? asm(key self keyLen) "DICTIDELGET" "NULLSWAPIFNOT"; @pure @@ -185,7 +185,7 @@ fun uDictSetBuilderIfExists(mutate self: cell, keyLen: int, key: int, value: bui @pure -fun iDictDeleteFirstAndGet(mutate self: cell, keyLen: int): (int, slice, bool) +fun iDictDeleteFirstAndGet(mutate self: cell, keyLen: int): (int, slice, bool) // todo keep not null? asm(-> 0 2 1 3) "DICTIREMMIN" "NULLSWAPIFNOT2"; @pure @@ -198,7 +198,7 @@ fun sDictDeleteFirstAndGet(mutate self: cell, keyLen: int): (slice, slice, bool) @pure -fun iDictDeleteLastAndGet(mutate self: cell, keyLen: int): (int, slice, bool) +fun iDictDeleteLastAndGet(mutate self: cell, keyLen: int): (int, slice, bool) // todo keep not null? asm(-> 0 2 1 3) "DICTIREMMAX" "NULLSWAPIFNOT2"; @pure @@ -211,7 +211,7 @@ fun sDictDeleteLastAndGet(mutate self: cell, keyLen: int): (slice, slice, bool) @pure -fun iDictGetFirst(self: cell, keyLen: int): (int, slice, bool) +fun iDictGetFirst(self: cell, keyLen: int): (int, slice, bool) // todo keep not null? asm (-> 1 0 2) "DICTIMIN" "NULLSWAPIFNOT2"; @pure @@ -236,7 +236,7 @@ fun sDictGetFirstAsRef(self: cell, keyLen: int): (slice, cell, bool) @pure -fun iDictGetLast(self: cell, keyLen: int): (int, slice, bool) +fun iDictGetLast(self: cell, keyLen: int): (int, slice, bool) // todo keep not null? asm (-> 1 0 2) "DICTIMAX" "NULLSWAPIFNOT2"; @pure @@ -261,7 +261,7 @@ fun sDictGetLastAsRef(self: cell, keyLen: int): (slice, cell, bool) @pure -fun iDictGetNext(self: cell, keyLen: int, pivot: int): (int, slice, bool) +fun iDictGetNext(self: cell, keyLen: int, pivot: int): (int, slice, bool) // todo keep not null? asm(pivot self keyLen -> 1 0 2) "DICTIGETNEXT" "NULLSWAPIFNOT2"; @pure @@ -278,7 +278,7 @@ fun uDictGetNextOrEqual(self: cell, keyLen: int, pivot: int): (int, slice, bool) @pure -fun iDictGetPrev(self: cell, keyLen: int, pivot: int): (int, slice, bool) +fun iDictGetPrev(self: cell, keyLen: int, pivot: int): (int, slice, bool) // todo keep not null? asm(pivot self keyLen -> 1 0 2) "DICTIGETPREV" "NULLSWAPIFNOT2"; @pure @@ -299,7 +299,7 @@ fun uDictGetPrevOrEqual(self: cell, keyLen: int, pivot: int): (int, slice, bool) */ @pure -fun prefixDictGet(self: cell, keyLen: int, key: slice): (slice, slice, slice, bool) +fun prefixDictGet(self: cell, keyLen: int, key: slice): (slice, slice, slice, bool) // todo keep not null? asm(key self keyLen) "PFXDICTGETQ" "NULLSWAPIFNOT2"; @pure diff --git a/tolk-tester/tests/a10.tolk b/tolk-tester/tests/a10.tolk index 031e29c9..9d24f38d 100644 --- a/tolk-tester/tests/a10.tolk +++ b/tolk-tester/tests/a10.tolk @@ -2,7 +2,7 @@ import "@stdlib/tvm-lowlevel" fun pair_first(p: [X, Y]): X asm "FIRST"; -fun one(dummy: tuple) { +fun one(dummy: tuple?) { return 1; } diff --git a/tolk-tester/tests/assignment-tests.tolk b/tolk-tester/tests/assignment-tests.tolk index 2d9e8bd8..a9a32fe9 100644 --- a/tolk-tester/tests/assignment-tests.tolk +++ b/tolk-tester/tests/assignment-tests.tolk @@ -197,9 +197,9 @@ fun test115() { fun main(value: int) { - var (x: int, y) = (autoInferIntNull(value), autoInferIntNull(value * 2)); + var (x: int?, y) = (autoInferIntNull(value), autoInferIntNull(value * 2)); if (x == null && y == null) { return null; } - return x == null || y == null ? -1 : x + y; + return x == null || y == null ? -1 : x! + y!; } /** diff --git a/tolk-tester/tests/c2.tolk b/tolk-tester/tests/c2.tolk index 257aba5b..bcbc6c93 100644 --- a/tolk-tester/tests/c2.tolk +++ b/tolk-tester/tests/c2.tolk @@ -8,7 +8,7 @@ fun unnamed_args(_: int, _: slice, _: int) { return true; } -fun main(x: int, y: int, z: int): bool { +fun main(x: int, y: int, z: int): bool? { op = `_+_`; if (0) { return null; } return check_assoc(x, y, z); diff --git a/tolk-tester/tests/cells-slices.tolk b/tolk-tester/tests/cells-slices.tolk index 19e2e215..772812eb 100644 --- a/tolk-tester/tests/cells-slices.tolk +++ b/tolk-tester/tests/cells-slices.tolk @@ -32,7 +32,8 @@ fun test1(): [int,int,int,int,int] { fun test2(): [int,int,int] { var b: builder = beginCell().myStoreInt(1, 32); b = b.myStoreInt(2, 32); - b.myStoreInt(3, 32); + // operator ! here and below is used just for testing purposes, it doesn't affect the result + b!.myStoreInt(3, 32); var cs: slice = b.endCell().beginParse(); var one: int = cs.myLoadInt(32); @@ -43,14 +44,14 @@ fun test2(): [int,int,int] { @method_id(103) fun test3(ret: int): int { - val same: int = beginCell().storeUint(ret,32).endCell().beginParse().loadUint(32); + val same: int = beginCell()!.storeUint(ret,32).endCell().beginParse().loadUint(32); return same; } @method_id(104) fun test4(): [int,int] { - var b: builder = beginCell().myStoreInt(1, 32); - b = b.storeInt(2, 32).storeInt(3, 32); + var b: builder = (beginCell() as builder).myStoreInt(1, 32); + b = b!.storeInt(2, 32)!.storeInt(3, 32); var cs: slice = b.endCell().beginParse(); var (one, _, three) = (cs.getFirstBits(32).loadUint(32), cs.skipBits(64), cs.load_u32()); @@ -116,7 +117,7 @@ fun test10() { fun test11() { var s: slice = beginCell().storeInt(1, 32).storeInt(2, 32).storeInt(3, 32).storeInt(4, 32).storeInt(5, 32).storeInt(6, 32).storeInt(7, 32).endCell().beginParse(); var size1 = getRemainingBitsCount(s); - s.skipBits(32); + s!.skipBits(32); var s1: slice = s.getFirstBits(64); var n1 = s1.loadInt(32); var size2 = getRemainingBitsCount(s); diff --git a/tolk-tester/tests/generics-1.tolk b/tolk-tester/tests/generics-1.tolk index 453ec282..ca310927 100644 --- a/tolk-tester/tests/generics-1.tolk +++ b/tolk-tester/tests/generics-1.tolk @@ -49,17 +49,17 @@ fun manyEq(a: T1, b: T2, c: T3): [T1, T2, T3] { fun test104(f: int) { var result = ( manyEq(1 ? 1 : 1, f ? 0 : null, !f ? getTwo() as int : null), - manyEq(f ? null as int : eq2(2), beginCell().storeBool(true).endCell().beginParse().loadBool(), eq4(f)) + manyEq(f ? null as int? : eq2(2), beginCell().storeBool(true).endCell().beginParse().loadBool(), eq4(f)) ); - __expect_type(result, "([int, int, int], [int, bool, int])"); + __expect_type(result, "([int, int?, int?], [int?, bool, int])"); return result; } -fun calcSum(x: X, y: X) { return x + y; } +fun calcSum(x: X, y: X) { return x! + y!; } @method_id(105) fun test105() { - if (0) { calcSum(((0)), null); } + if (0) { calcSum(((0 as int?)), null); } return (calcSum(1, 2)); } diff --git a/tolk-tester/tests/indexed-access.tolk b/tolk-tester/tests/indexed-access.tolk index ab7995cf..7915536e 100644 --- a/tolk-tester/tests/indexed-access.tolk +++ b/tolk-tester/tests/indexed-access.tolk @@ -86,8 +86,8 @@ fun test104() { } @method_id(105) -fun test105(x: int, y: int): (tuple, int, (int, int), int, int) { - var ab = (createEmptyTuple(), (x, y), tupleSize); +fun test105(x: int, y: int): (tuple, int, (int?, int), int, int) { + var ab = (createEmptyTuple(), (x as int?, y), tupleSize); ab.0.tuplePush(1); tuplePush(mutate ab.0, 2); ab.1.0 = null; @@ -98,7 +98,7 @@ fun test105(x: int, y: int): (tuple, int, (int, int), int, int) { @method_id(106) fun test106(x: int, y: int) { - var ab = [createEmptyTuple(), [x, y], tupleSize]; + var ab = [createEmptyTuple(), [x as int?, y], tupleSize]; ab.0.tuplePush(1); tuplePush(mutate ab.0, 2); ab.1.0 = null; @@ -233,6 +233,25 @@ fun test121(zero: int) { return t; } +fun isFirstComponentGt0(t: (T1, T2)): bool { + return t.0 > 0; +} + +@method_id(122) +fun test122(x: (int, int)) { + return ( + isFirstComponentGt0(x), isFirstComponentGt0((2, beginCell())), isFirstComponentGt0((0, null)), + x.isFirstComponentGt0(), (2, beginCell()).isFirstComponentGt0(), (0, null).isFirstComponentGt0() + ); +} + +@method_id(123) +fun test123() { + var t = [[10, 20]] as [[int,int]]?; + t!.0.0 = t!.0.1 = 100; + return t; +} + fun main(){} @@ -258,6 +277,8 @@ fun main(){} @testcase | 119 | 1 2 3 4 | 4 1 3 @testcase | 120 | | 3 4 [ 5 6 ] @testcase | 121 | 0 | [ 3 ] +@testcase | 122 | 1 2 | -1 -1 0 -1 -1 0 +@testcase | 123 | | [ [ 100 100 ] ] @fif_codegen """ diff --git a/tolk-tester/tests/inference-tests.tolk b/tolk-tester/tests/inference-tests.tolk index 3d451581..fe2e5933 100644 --- a/tolk-tester/tests/inference-tests.tolk +++ b/tolk-tester/tests/inference-tests.tolk @@ -18,10 +18,12 @@ fun test1(x: int, y: int) { __expect_type(random() ? x : y, "int"); __expect_type(eq(x), "int"); __expect_type(eq(x), "int"); - __expect_type(eq(null), "int"); + __expect_type(eq(null), "int?"); __expect_type(x as int, "int"); __expect_type(+x, "int"); __expect_type(~x, "int"); + __expect_type(x!, "int"); + __expect_type(x!!!, "int"); { var x: slice = beginCell().endCell().beginParse(); __expect_type(x, "slice"); @@ -62,7 +64,7 @@ fun test5(x: int) { __expect_type([], "[]"); __expect_type([x], "[int]"); __expect_type([x, x >= 1], "[int, bool]"); - __expect_type([x, x >= 1, null as slice], "[int, bool, slice]"); + __expect_type([x, x >= 1, null as slice?], "[int, bool, slice?]"); __expect_type((x, [x], [[x], x]), "(int, [int], [[int], int])"); __expect_type(getMyOriginalBalanceWithExtraCurrencies(), "[int, cell]"); } diff --git a/tolk-tester/tests/invalid-generics-1.tolk b/tolk-tester/tests/invalid-generics-1.tolk index c8ff7fec..0bbdeee6 100644 --- a/tolk-tester/tests/invalid-generics-1.tolk +++ b/tolk-tester/tests/invalid-generics-1.tolk @@ -6,5 +6,5 @@ fun failCantDeduceWithoutArgument() { /** @compilation_should_fail -@stderr can not deduce X for generic function `f` +@stderr too few arguments in call to `f`, expected 2, have 1 */ diff --git a/tolk-tester/tests/invalid-generics-13.tolk b/tolk-tester/tests/invalid-generics-13.tolk new file mode 100644 index 00000000..7574bde7 --- /dev/null +++ b/tolk-tester/tests/invalid-generics-13.tolk @@ -0,0 +1,11 @@ +fun calcSum(x: X, y: X) { return x + y; } + +fun cantApplyPlusOnNullable() { + return calcSum(((0 as int?)), null); +} + +/** +@compilation_should_fail +@stderr while instantiating generic function `calcSum` +@stderr can not apply operator `+` to `int?` and `int?` + */ diff --git a/tolk-tester/tests/invalid-mutate-18.tolk b/tolk-tester/tests/invalid-mutate-18.tolk new file mode 100644 index 00000000..bb8cde05 --- /dev/null +++ b/tolk-tester/tests/invalid-mutate-18.tolk @@ -0,0 +1,10 @@ +fun getNullableTuple(): tuple? { return createEmptyTuple(); } + +fun cantUseLValueUnwrappedNotNull() { + tuplePush(mutate getNullableTuple()!, 1); +} + +/** +@compilation_should_fail +@stderr function call can not be used as lvalue + */ diff --git a/tolk-tester/tests/invalid-mutate-19.tolk b/tolk-tester/tests/invalid-mutate-19.tolk new file mode 100644 index 00000000..bb8cde05 --- /dev/null +++ b/tolk-tester/tests/invalid-mutate-19.tolk @@ -0,0 +1,10 @@ +fun getNullableTuple(): tuple? { return createEmptyTuple(); } + +fun cantUseLValueUnwrappedNotNull() { + tuplePush(mutate getNullableTuple()!, 1); +} + +/** +@compilation_should_fail +@stderr function call can not be used as lvalue + */ diff --git a/tolk-tester/tests/invalid-mutate-20.tolk b/tolk-tester/tests/invalid-mutate-20.tolk new file mode 100644 index 00000000..f6eb2f9f --- /dev/null +++ b/tolk-tester/tests/invalid-mutate-20.tolk @@ -0,0 +1,13 @@ +fun acceptMutateNullableTensor(mutate self: (int, int)?) { +} + +fun cantModifyTupleIndexWithTypeTransition() { + var t = [1, null]; + t.1.acceptMutateNullableTensor(); +} + +/** +@compilation_should_fail +@stderr can not call method for mutate `(int, int)?` with object of type `null` +@stderr because mutation is not type compatible + */ diff --git a/tolk-tester/tests/invalid-typing-14.tolk b/tolk-tester/tests/invalid-typing-14.tolk new file mode 100644 index 00000000..657ab5f4 --- /dev/null +++ b/tolk-tester/tests/invalid-typing-14.tolk @@ -0,0 +1,14 @@ + +fun autoGetIntOrNull() { + if (random()) { return 1; } + return null; +} + +fun testAutoInferredIntOrNull() { + var b: builder = autoGetIntOrNull() as builder; +} + +/** +@compilation_should_fail +@stderr type `int?` can not be cast to `builder` + */ diff --git a/tolk-tester/tests/invalid-typing-15.tolk b/tolk-tester/tests/invalid-typing-15.tolk new file mode 100644 index 00000000..fbcff8a2 --- /dev/null +++ b/tolk-tester/tests/invalid-typing-15.tolk @@ -0,0 +1,13 @@ + +fun getNullable4(): int? { + return 4; +} + +fun testCantSumNullable() { + return 1 + getNullable4(); +} + +/** +@compilation_should_fail +@stderr can not apply operator `+` to `int` and `int?` + */ diff --git a/tolk-tester/tests/invalid-typing-16.tolk b/tolk-tester/tests/invalid-typing-16.tolk new file mode 100644 index 00000000..287754fd --- /dev/null +++ b/tolk-tester/tests/invalid-typing-16.tolk @@ -0,0 +1,10 @@ +import "@stdlib/tvm-dicts.tolk" + +fun testCantCallDictMethodsOnNullable(c: cell) { + c.beginParse().loadDict().iDictDelete(16, 1); +} + +/** +@compilation_should_fail +@stderr can not call method for `cell` with object of type `cell?` + */ diff --git a/tolk-tester/tests/invalid-typing-17.tolk b/tolk-tester/tests/invalid-typing-17.tolk new file mode 100644 index 00000000..b7302684 --- /dev/null +++ b/tolk-tester/tests/invalid-typing-17.tolk @@ -0,0 +1,10 @@ + +fun testCantUseNullableAsCondition(x: int?) { + if (x) { return 1; } + return 0; +} + +/** +@compilation_should_fail +@stderr can not use `int?` as a boolean condition + */ diff --git a/tolk-tester/tests/invalid-typing-18.tolk b/tolk-tester/tests/invalid-typing-18.tolk new file mode 100644 index 00000000..cf985add --- /dev/null +++ b/tolk-tester/tests/invalid-typing-18.tolk @@ -0,0 +1,16 @@ +fun incrementOrSetNull(mutate x: int?) { + if (random()) { x! += 1; } + else { x = null; } +} + +fun cantCallMutateMethodNotNullable() { + var x = 1; + incrementOrSetNull(mutate x); + return x; +} + +/** +@compilation_should_fail +@stderr can not pass `int` to mutate `int?` +@stderr because mutation is not type compatible + */ diff --git a/tolk-tester/tests/logical-operators.tolk b/tolk-tester/tests/logical-operators.tolk index 29cd1d10..700f2a3c 100644 --- a/tolk-tester/tests/logical-operators.tolk +++ b/tolk-tester/tests/logical-operators.tolk @@ -53,9 +53,8 @@ fun testDict(last: int) { } @method_id(105) -fun testNotNull(x: int) { - // return [x == null, null == x, !(x == null), null == null, +(null != null)]; - return [x == null, null == x, !(x == null)]; +fun testNotNull(x: int?) { + return [x == null, null == x, !(x == null), null == null, (null != null) as int]; } @method_id(106) @@ -170,8 +169,8 @@ fun main() { @testcase | 104 | 50 | 3 5 -1 @testcase | 104 | 100 | 3 5 5 @testcase | 104 | 0 | 3 -1 5 -@testcase | 105 | 0 | [ 0 0 -1 ] -@testcase | 105 | null | [ -1 -1 0 ] +@testcase | 105 | 0 | [ 0 0 -1 -1 0 ] +@testcase | 105 | null | [ -1 -1 0 -1 0 ] @testcase | 106 | | [ 0 0 0 -1 ] [ 0 0 0 ] [ -1 -1 -1 ] [ 0 -1 ] @testcase | 107 | | [ -1 -1 0 -1 ] [ 0 0 0 ] [ -1 -1 -1 ] [ -1 0 ] @testcase | 108 | 1 2 | -1 diff --git a/tolk-tester/tests/null-keyword.tolk b/tolk-tester/tests/null-keyword.tolk index eb02b624..3ea0aaa2 100644 --- a/tolk-tester/tests/null-keyword.tolk +++ b/tolk-tester/tests/null-keyword.tolk @@ -2,13 +2,13 @@ import "@stdlib/lisp-lists" @method_id(101) fun test1() { - var numbers: tuple = createEmptyList(); + var numbers: tuple? = createEmptyList(); numbers = listPrepend(1, numbers); numbers = listPrepend(2, numbers); numbers = listPrepend(3, numbers); numbers = listPrepend(4, numbers); - var (h: int, numbers redef) = listSplit(numbers); - h += listGetHead(numbers); + var (h: int, numbers redef) = listSplit(numbers!); + h += listGetHead(numbers!); _ = null; (_, _) = (null, null); @@ -22,22 +22,22 @@ fun test1() { } @method_id(102) -fun test2(x: int) { +fun test2(x: int?) { if (null != x) { - var y: int = null; + var y: int? = null; if (y != null) { return 10; } return y; } try { - return x + 10; // will throw, since not a number + return x! + 10; // will throw, since not a number } catch { return -1; } return 100; } -fun myIsNull(x: int): int { - return x == null ? -1 : x; +fun myIsNull(x: int?): int { + return x == null ? -1 : x!; } @method_id(103) @@ -64,21 +64,28 @@ fun test4(): null { @method_id(105) fun test5() { - var n: slice = getUntypedNull(); - return !(null == n) ? n.loadInt(32) : 100; + var n: slice? = getUntypedNull(); + return !(null == n) ? n!.loadInt(32) : 100; } @method_id(107) fun test7() { - var b = beginCell().storeMaybeRef(null); - var s = b.endCell().beginParse(); + var b = beginCell().storeMaybeRef(null) as builder?; + var s = b!.endCell().beginParse(); var c = s.loadMaybeRef(); return (null == c) as int * 10 + (b != null) as int; } +fun test8() { + __expect_type(null, "null"); + __expect_type([[null]], "[[null]]"); + __expect_type(null as tuple?, "tuple?"); + __expect_type(null as [int]?, "[int]?"); + __expect_type(((null)) as (int, int)?, "(int, int)?"); +} + fun main() { - // now, the compiler doesn't optimize this at compile-time, fif codegen contains ifs - var i: int = null; + var i: int? = null; if (i == null) { return 1; } @@ -120,12 +127,12 @@ fun main() { """ main PROC:<{ // - PUSHNULL // i - ISNULL // '2 - IFJMP:<{ // - 1 PUSHINT // '3=1 - }> // - 10 PUSHINT // '4=10 + PUSHNULL // i + ISNULL // '2 + IFJMP:<{ // + 1 PUSHINT // '3=1 + }> // + 10 PUSHINT // '4=10 }> """ @@ -139,8 +146,8 @@ fun main() { 10 MULCONST // b '13 SWAP // '13 b ISNULL // '13 '14 - NOT // '13 '15 - ADD // '16 + NOT // '13 '14 + ADD // '15 }> """ */ diff --git a/tolk-tester/tests/nullable-tensors.tolk b/tolk-tester/tests/nullable-tensors.tolk new file mode 100644 index 00000000..d95f5a58 --- /dev/null +++ b/tolk-tester/tests/nullable-tensors.tolk @@ -0,0 +1,404 @@ +fun getNullableInt(): int? { return 5; } + +fun sumOfNullableTensorComponents(t: (int, int)?): int { + if (t == null) { return 0; } + return t!.0 + t!.1; +} + +fun isTensorNull(t: (int, int)?) { + return t == null; +} + +fun incrementNullableTensorComponents(mutate self: (int, int)?): self { + if (self != null) { + self!.0 += 1; + self!.1 += 1; + } + return self; +} + +fun incrementTensorComponents(mutate self: (int, int)): self { + self.0 += 1; + self.1 += 1; + return self; +} + +fun assignFirstComponent(mutate t: (int, int), first: int) { + t!.0 = first; +} + +fun assignFirstComponentNullable(mutate t: (int, int)?, first: int) { + if (t == null) { + t = (first, 0); + } else { + t!.0 = first; + } +} + +fun getNullableTensor(firstComponent: int?): (int, int)? { + return firstComponent == null ? null : (firstComponent!, 2); +} + +fun sumOfTensor(x: (int, int)) { + return x.0 + x.1; +} + +fun assignNullTo(mutate x: T?) { + x = null; +} + +fun getTensor12() { + return (1,2); +} + +@method_id(101) +fun test101(): (int, int)? { + return (1, 2); +} + +@method_id(102) +fun test102(): ((int, int)?, (int, int)?) { + var t = (1, 2); + return (t, null); +} + +@method_id(103) +fun test103(t: (int, int)) { + var t2: (int, int)? = t; + return (sumOfNullableTensorComponents(t), sumOfNullableTensorComponents(t2), sumOfNullableTensorComponents(null), t2); +} + +@method_id(104) +fun test104() { + var t1_1: (int, int)? = (1, 2); + var t1_2: (int, int)? = t1_1; + var t1_3: (int, int)? = t1_1!; + var t2_1: (int, int)? = null; + var t2_2 = t2_1; + return (t1_3, t2_2); +} + +@method_id(105) +fun test105() { + return (null as (int, slice, cell)?, (1, 2, 3) as (int, int, int)?); +} + +@method_id(106) +fun test106() { + var t: (int?, int?)? = (((((1, 2))) as (int, int))); + return t; +} + +@method_id(107) +fun test107() { + var ab = (1, 2); + var ab2: (int, int)? = ab; + return (isTensorNull(ab), isTensorNull(ab2), isTensorNull(null), ab.isTensorNull(), ab2.isTensorNull(), null.isTensorNull()); +} + +@method_id(108) +fun test108(x1: (int, int)) { + incrementTensorComponents(mutate x1); + x1.incrementTensorComponents(); + var x2: (int, int)? = x1; + x2.incrementNullableTensorComponents().incrementNullableTensorComponents(); + incrementNullableTensorComponents(mutate x2); + var x3: (int, int)? = null; + x3.incrementNullableTensorComponents().incrementNullableTensorComponents(); + incrementNullableTensorComponents(mutate x3); + return (x1, x2, x3); +} + +fun isTensorNullGen(t: (T1, T2)?) { + return t == null; +} + +@method_id(109) +fun test109() { + var x1 = (1, 2); + var x2: (int, int)? = x1; + var x3: (int, int)? = x1.1 > 10 ? (1, 2) : null; + return ( + isTensorNullGen(x1), isTensorNullGen(x2), isTensorNullGen(null), + isTensorNullGen(x1), isTensorNullGen(x3), + x1.isTensorNullGen(), x2.isTensorNullGen(), x3.isTensorNullGen(), null.isTensorNullGen() + ); +} + +global g110_1: (int, int); +global g110_2: (int, int)?; + +@method_id(110) +fun test110() { + g110_1 = getNullableTensor(1)!; + incrementTensorComponents(mutate g110_1); + g110_1.incrementTensorComponents(); + g110_2 = g110_1; + g110_2.incrementNullableTensorComponents().incrementNullableTensorComponents(); + incrementNullableTensorComponents(mutate g110_2); + var tmp = g110_2; + g110_2 = null; + g110_2.incrementNullableTensorComponents(); + incrementNullableTensorComponents(mutate g110_2); + return (g110_1, g110_2, tmp); +} + +@method_id(111) +fun test111() { + var x = (1, 2); + assignFirstComponent(mutate x, 50); + var x2: (int, int)? = null; + var x3 = x2; + assignFirstComponentNullable(mutate x2, 30); + assignFirstComponentNullable(mutate x3, 70); + g110_1 = (1, 2); + g110_2 = null; + assignFirstComponent(mutate g110_1, 90); + assignFirstComponentNullable(mutate g110_2, 100); + return (x.0, x2!.0, x3!.0, g110_1.0, g110_2!.0); +} + +@method_id(112) +fun test112() { + var x: (int, int)? = (10, 20); + incrementTensorComponents(mutate x!); + x!.incrementTensorComponents(); + return x; +} + +@method_id(113) +fun test113() { + var t = [1, null]; // t.1 is always null + return isTensorNull(t.1); +} + +@method_id(114) +fun test114(): ((slice, (cell, [int, slice, tuple]))?, slice?, (int?, bool?)?) { + var t = [[null]]; + return (t.0.0, t.0.0, t.0.0); +} + +@method_id(115) +fun test115() { + var tt = getNullableTensor(null); + assignFirstComponentNullable(mutate tt, 5); + return ( + getNullableTensor(1)!.incrementTensorComponents(), + sumOfNullableTensorComponents(getNullableTensor(1).incrementNullableTensorComponents().incrementNullableTensorComponents()), + getNullableTensor(null).incrementNullableTensorComponents(), + tt, + sumOfNullableTensorComponents(getNullableTensor(null)) + ); +} + +@method_id(116) +fun test116(returnNull: bool) { + var t1: (int, int)? = returnNull ? null : getTensor12(); + var t2 = returnNull ? null as (int, int)? : getTensor12() as (int, int)?; + return (t1, t2); +} + +@method_id(117) +fun test117() { + var (a, b: (int, int)?, c) = (1, null, 3); + return (b, a, c); +} + +fun autoInferNullableTensor(a: int?, b: int) { + if (a != null) { + return (a!, b); + } + return null; +} + +@method_id(118) +fun test118(a: int?) { + return autoInferNullableTensor(a, 10); +} + +@method_id(119) +fun test119() { + var x: (int, int)? = (1, 2); + x = null; + var tt: (int, (int, int)?) = (0, (1, 2)); + tt.1 = null; + var third: (int, (int, int)?, int) = (0, (1, 2), 3); + third.2 = 100; + return (x, tt.1, third.1, third.2); +} + +@method_id(120) +fun test120(setNull: bool) { + var x: (int, int)? = (1, 2); + if (setNull) { + assignNullTo(mutate x); + } + return x; +} + +@method_id(121) +fun test121() { + var t: [int?, [int?, int?]?] = [1, [2, 3]]; + t.1 = [3, 4]; + return t; +} + +@method_id(122) +fun test122(setNull: bool) { + var t: [int?, [int?, int?]?, int?, [int?, int?]?]? = [1, [2, 3], 4, null]; + if (setNull) { + assignNullTo(mutate t!.1); + } else { + var rhs = [3, 4]; + t!!.1 = rhs; + } + return t; +} + +@method_id(123) +fun test123() { + var t: (int?, (int?, int?)?) = (1, (2, 3)); + t.1 = (3, 4); + return t; +} + +@method_id(124) +fun test124(setNull: bool) { + var t: (int?, (int?, int?)?, int?, (int?, int?)?)? = (1, (2, 3), 4, null); + if (setNull) { + assignNullTo(mutate t!.1); + } else { + var rhs = (3, 4); + t!!.1 = rhs; + } + return t; +} + +global g125: int; +fun getT125(): (int, (int, int)?, (int?, int)?) { return (g125 += 1, null, null); } + +@method_id(125) +fun test125() { + g125 = 0; + getT125().1 = null; + getT125().2 = (1, 2); + (getT125()!! as (int, (int, int)?, (int?, int)?)).2 = null; + // test that nothing left on a stack + return g125; +} + +@method_id(126) +fun test126() { + var tt1: (int, null, int) = (1, null, 2); + var (a: int, b: (int, int)?, c: int) = tt1; + return (a, b, c); +} + +@method_id(127) +fun test127(choice: int) { + var tt1: (int, null, int) = (1, null, 2); + var tt2: (int, (int, int), int) = (1, (2, 3), 4); + var tt3: (int, (int, int)?, int) = (1, null, 5); + var abc: (int, (int, int)?, int) = choice == 1 ? tt1 : choice == 2 ? tt2 : tt3; + return abc; +} + +fun get128_1() { return (1, null, 2); } +fun get128_2() { return null; } +fun get128_3() { return (1, (2, 3), 4); } +fun takeT128(abc: (int, (int, int)?, int)?) { return abc; } + +@method_id(128) +fun test128(choice: int) { + if (choice == 1) { + return takeT128(get128_1())!; + } + if (choice == 2) { + return takeT128(get128_2()); + } + return takeT128(get128_3()); +} + +@method_id(129) +fun test129(setNull: bool) { + var t: (int?, int?) = (getNullableInt(), getNullableInt()); + var r1 = (t, t == null, t != null); + t = (setNull ? null : 1, setNull ? null : 2); + var r2 = (t, t == null, t != null); + return (r1, r2); +} + +@method_id(130) +fun test130(setNull: bool) { + var os: (int, (int, int)?) = (1, setNull ? null : (2, 3)); + return os; +} + +fun main(){} + +/** +@testcase | 101 | | 1 2 -1 +@testcase | 102 | | 1 2 -1 (null) (null) 0 +@testcase | 103 | 1 2 | 3 3 0 1 2 -1 +@testcase | 104 | | 1 2 -1 (null) (null) 0 +@testcase | 105 | | (null) (null) (null) 0 1 2 3 -1 +@testcase | 106 | | 1 2 -1 +@testcase | 107 | | 0 0 -1 0 0 -1 +@testcase | 108 | 5 6 | 7 8 10 11 -1 (null) (null) 0 +@testcase | 109 | | 0 0 -1 0 -1 0 0 -1 -1 +@testcase | 110 | | 3 4 (null) (null) 0 6 7 -1 +@testcase | 111 | | 50 30 70 90 100 +@testcase | 112 | | 12 22 -1 +@testcase | 113 | | -1 +@testcase | 114 | | (null) (null) (null) 0 (null) (null) (null) 0 +@testcase | 115 | | 2 3 7 (null) (null) 0 5 0 -1 0 +@testcase | 116 | -1 | (null) (null) 0 (null) (null) 0 +@testcase | 116 | 0 | 1 2 -1 1 2 -1 +@testcase | 117 | | (null) (null) 0 1 3 +@testcase | 118 | 5 | 5 10 -1 +@testcase | 118 | null | (null) (null) 0 +@testcase | 119 | | (null) (null) 0 (null) (null) 0 1 2 -1 100 +@testcase | 120 | -1 | (null) (null) 0 +@testcase | 120 | 0 | 1 2 -1 +@testcase | 121 | | [ 1 [ 3 4 ] ] +@testcase | 122 | 0 | [ 1 [ 3 4 ] 4 (null) ] +@testcase | 122 | -1 | [ 1 (null) 4 (null) ] +@testcase | 123 | | 1 3 4 -1 +@testcase | 124 | 0 | 1 3 4 -1 4 (null) (null) 0 -1 +@testcase | 124 | -1 | 1 (null) (null) 0 4 (null) (null) 0 -1 +@testcase | 125 | | 3 +@testcase | 126 | | 1 (null) (null) 0 2 +@testcase | 127 | 1 | 1 (null) (null) 0 2 +@testcase | 127 | 2 | 1 2 3 -1 4 +@testcase | 127 | 3 | 1 (null) (null) 0 5 +@testcase | 128 | 1 | 1 (null) (null) 0 2 -1 +@testcase | 128 | 2 | (null) (null) (null) (null) (null) 0 +@testcase | 128 | 3 | 1 2 3 -1 4 -1 +@testcase | 129 | 0 | 5 5 0 -1 1 2 0 -1 +@testcase | 129 | -1 | 5 5 0 -1 (null) (null) 0 -1 +@testcase | 130 | 0 | 1 2 3 -1 +@testcase | 130 | -1 | 1 (null) (null) 0 + +@fif_codegen +""" + isTensorNull PROC:<{ + // t.0 t.1 t.NNFlag + 2 1 BLKDROP2 // t.NNFlag + 0 EQINT // '3 + }> +""" + +@fif_codegen +""" + test113 PROC:<{ + // + 1 PUSHINT // '2=1 + PUSHNULL // '2=1 '3 + PAIR // t + 1 INDEX // '5 + PUSHNULL // '5 '6 + 0 PUSHINT // '5 '6 '7=0 + isTensorNull CALLDICT // '8 + }> +""" +*/ diff --git a/tolk-tester/tests/nullable-types.tolk b/tolk-tester/tests/nullable-types.tolk new file mode 100644 index 00000000..ebabb80d --- /dev/null +++ b/tolk-tester/tests/nullable-types.tolk @@ -0,0 +1,109 @@ + +fun getNullable4(): int? { return 4; } +fun getNullableIntNull(): int? asm "PUSHNULL"; + +fun eqInt(x: int) { return x; } +fun eq(x: T) { return x; } + +fun unwrap(x: T?): T { return x!; } +fun intOr0(x: int?): int { return null == x ? 0 : x!; } + +@method_id(101) +fun test101(x: int) { + var re = x == 0 ? null : 100; + return re == null ? re : 200 + getNullable4()!; +} + +@method_id(102) +fun test102(a: int) { + try { + throw (123, a > 10 ? null : a); + return 0; + } catch (excno, arg) { + var i = arg as int?; + return excno + (i != null ? i!!!!! : -100); + } +} + +@method_id(103) +fun test103(x: int?): (bool, bool, int) { + var x_gt_0 = x != null && eqInt(x!) > 0; + var x_lt_0 = x != null && eq(x)! < 0; + if (x == null) { + return (x_gt_0, x_lt_0, 0); + } + return (x_gt_0, x_lt_0, x!); +} + +@method_id(104) +fun test104(x: int?) { + var x2 = eq(x = 10); + var ab = (x2, getNullableIntNull()); + return (unwrap(ab.0) + (ab.1 == null ? -100 : ab.1!), ab.1); +} + +@method_id(105) +fun test105() { + var xy: (int?, int?) = (5, null); + var ab = [1 ? [xy.0, xy.1] : null]; + ab.0!.0 = intOr0(ab.0!.0); + ab.0!.1 = intOr0(ab.0!.1); + return ab.0!.0! + ab.0!.1!; +} + +global gTup106: tuple?; +global gInt106: int?; + +@method_id(106) +fun test106() { + gInt106 = 0; + gInt106! += 5; + var int106: int? = 0; + var gTup106 = createEmptyTuple(); + gTup106!.tuplePush(createEmptyTuple()); + (gTup106!.0 as tuple?)!.tuplePush(0 as int?); + tuplePush(mutate gTup106!, gInt106); + tuplePush(mutate gTup106!.0, int106! += 1); + return (gTup106 == null, null != gTup106, gTup106, gTup106!.0 as tuple?); +} + +@method_id(107) +fun test107() { + var b: builder? = beginCell(); + b!.storeInt(1, 32).storeInt(2, 32); + b = b!.storeInt(3, 32); + storeInt(mutate b!, 4, 32); + (b! as builder).storeInt(5, 32); + return b!.getBuilderBitsCount(); +} + +@method_id(108) +fun test108() { + var (a, b: cell?, c) = (1, beginCell().endCell(), 3); + b = null; + return a + (b == null ? 0 : b!.beginParse().loadInt(32)) + c; +} + +@method_id(109) +fun test109() { + var a = getNullable4(); + var b = getNullable4(); + return ([a, b] = [3, 4], a, b); +} + +fun main(x: int?, y: int?) { +} + +/** +@testcase | 101 | 0 | (null) +@testcase | 101 | -1 | 204 +@testcase | 102 | 5 | 128 +@testcase | 102 | 15 | 23 +@testcase | 103 | 10 | -1 0 10 +@testcase | 104 | 8 | -90 (null) +@testcase | 105 | | 5 +@testcase | 106 | | 0 -1 [ [ 0 1 ] 5 ] [ 0 1 ] +@testcase | 107 | | 160 +@testcase | 108 | | 4 +@testcase | 109 | | [ 3 4 ] 3 4 + */ diff --git a/tolk-tester/tests/use-before-declare.tolk b/tolk-tester/tests/use-before-declare.tolk index d3e6b165..2a0e0e7f 100644 --- a/tolk-tester/tests/use-before-declare.tolk +++ b/tolk-tester/tests/use-before-declare.tolk @@ -27,8 +27,8 @@ fun test1(): int { var demo_var: int = demo_10; var demo_slice: int = demo_20; if (demo_var > 0) { - var demo_var: tuple = null; - var demo_slice: tuple = null; + var demo_var: tuple? = null; + var demo_slice: tuple? = null; } return demo_var + demo_slice; } diff --git a/tolk-tester/tests/var-apply.tolk b/tolk-tester/tests/var-apply.tolk index 16863560..d189430f 100644 --- a/tolk-tester/tests/var-apply.tolk +++ b/tolk-tester/tests/var-apply.tolk @@ -138,6 +138,43 @@ fun testIndexedAccessApply() { return functions2.0(functions1.1(b)).loadInt(32); } +fun getNullable4(): int? { return 4; } +fun myBeginCell(): builder? asm "NEWC"; + +@method_id(108) +fun testCallingNotNull() { + var n4: () -> int? = getNullable4; + var creator: (() -> builder?)? = myBeginCell; + var end2: [int, (builder -> cell)?] = [0, endCell]; + var c: cell = end2.1!((creator!()!)!.storeInt(getNullable4()!, 32)); + return c.beginParse().loadInt(32); +} + +fun sumOfTensorIfNotNull(t: (int, int)?) { + if (t == null) { return 0; } + return t!.0 + t!.1; +} + +@method_id(109) +fun testTypeTransitionOfVarCall() { + var summer = sumOfTensorIfNotNull; + var hh1 = [1, null]; + var tt1 = (3, 4); + return (summer(null), summer((1,2)), summer(hh1.1), summer(tt1)); +} + +fun makeTensor(x1: int, x2: int, x3: int, x4: int, x5: int) { + return (x1, x2, x3, x4, x5); +} + +fun eq(x: T): T { return x; } + +@method_id(110) +fun testVarsModificationInsideVarCall(x: int) { + var cb = makeTensor; + return x > 3 ? cb(x, x += 5, eq(x *= x), x, eq(x)) : null; +} + fun main() {} /** @@ -148,4 +185,8 @@ fun main() {} @testcase | 105 | | 1 @testcase | 106 | | 1 1 [ 2 ] [ 2 ] @testcase | 107 | | 65537 +@testcase | 108 | | 4 +@testcase | 109 | | 0 3 0 7 +@testcase | 110 | 5 | 5 10 100 100 100 -1 +@testcase | 110 | 0 | (null) (null) (null) (null) (null) 0 */ diff --git a/tolk/abscode.cpp b/tolk/abscode.cpp index b465b72b..72da0ac8 100644 --- a/tolk/abscode.cpp +++ b/tolk/abscode.cpp @@ -402,7 +402,7 @@ void CodeBlob::print(std::ostream& os, int flags) const { std::vector CodeBlob::create_var(TypePtr var_type, SrcLocation loc, std::string name) { std::vector ir_idx; - int stack_w = var_type->calc_width_on_stack(); + int stack_w = var_type->get_width_on_stack(); ir_idx.reserve(stack_w); if (const TypeDataTensor* t_tensor = var_type->try_as()) { for (int i = 0; i < t_tensor->size(); ++i) { @@ -410,6 +410,10 @@ std::vector CodeBlob::create_var(TypePtr var_type, SrcLocation loc, s std::vector nested = create_var(t_tensor->items[i], loc, std::move(sub_name)); ir_idx.insert(ir_idx.end(), nested.begin(), nested.end()); } + } else if (const TypeDataNullable* t_nullable = var_type->try_as(); t_nullable && stack_w != 1) { + std::string null_flag_name = name.empty() ? name : name + ".NNFlag"; + ir_idx = create_var(t_nullable->inner, loc, std::move(name)); + ir_idx.emplace_back(create_var(TypeDataBool::create(), loc, std::move(null_flag_name))[0]); } else if (var_type != TypeDataVoid::create()) { #ifdef TOLK_DEBUG tolk_assert(stack_w == 1); diff --git a/tolk/ast-from-tokens.cpp b/tolk/ast-from-tokens.cpp index f5855bc1..fcaa1157 100644 --- a/tolk/ast-from-tokens.cpp +++ b/tolk/ast-from-tokens.cpp @@ -111,23 +111,16 @@ static void diagnose_addition_in_bitshift(SrcLocation loc, std::string_view bits } } -// replace (a == null) and similar to isNull(a) (call of a built-in function) -static AnyExprV maybe_replace_eq_null_with_isNull_call(V v) { +// replace (a == null) and similar to ast_is_null_check(a) (special AST vertex) +static AnyExprV maybe_replace_eq_null_with_isNull_check(V v) { bool has_null = v->get_lhs()->type == ast_null_keyword || v->get_rhs()->type == ast_null_keyword; bool replace = has_null && (v->tok == tok_eq || v->tok == tok_neq); if (!replace) { return v; } - auto v_ident = createV(v->loc, "__isNull"); // built-in function - auto v_ref = createV(v->loc, v_ident, nullptr); - AnyExprV v_null = v->get_lhs()->type == ast_null_keyword ? v->get_rhs() : v->get_lhs(); - AnyExprV v_arg = createV(v->loc, v_null, false); - AnyExprV v_isNull = createV(v->loc, v_ref, createV(v->loc, {v_arg})); - if (v->tok == tok_neq) { - v_isNull = createV(v->loc, "!", tok_logical_not, v_isNull); - } - return v_isNull; + AnyExprV v_nullable = v->get_lhs()->type == ast_null_keyword ? v->get_rhs() : v->get_lhs(); + return createV(v->loc, v_nullable, v->tok == tok_neq); } @@ -372,16 +365,31 @@ static AnyExprV parse_expr100(Lexer& lex) { } } -// parse E(...) (left-to-right) +// parse E(...) and E! having parsed E already (left-to-right) +static AnyExprV parse_fun_call_postfix(Lexer& lex, AnyExprV lhs) { + while (true) { + if (lex.tok() == tok_oppar) { + lhs = createV(lhs->loc, lhs, parse_argument_list(lex)); + } else if (lex.tok() == tok_logical_not) { + lex.next(); + lhs = createV(lhs->loc, lhs); + } else { + break; + } + } + return lhs; +} + +// parse E(...) and E! (left-to-right) static AnyExprV parse_expr90(Lexer& lex) { AnyExprV res = parse_expr100(lex); - while (lex.tok() == tok_oppar) { - res = createV(res->loc, res, parse_argument_list(lex)); + if (lex.tok() == tok_oppar || lex.tok() == tok_logical_not) { + res = parse_fun_call_postfix(lex, res); } return res; } -// parse E.field and E.method(...) (left-to-right) +// parse E.field and E.method(...) and E.field! (left-to-right) static AnyExprV parse_expr80(Lexer& lex) { AnyExprV lhs = parse_expr90(lex); while (lex.tok() == tok_dot) { @@ -402,8 +410,8 @@ static AnyExprV parse_expr80(Lexer& lex) { lex.unexpected("method name"); } lhs = createV(loc, lhs, v_ident, v_instantiationTs); - while (lex.tok() == tok_oppar) { - lhs = createV(lex.cur_location(), lhs, parse_argument_list(lex)); + if (lex.tok() == tok_oppar || lex.tok() == tok_logical_not) { + lhs = parse_fun_call_postfix(lex, lhs); } } return lhs; @@ -491,7 +499,7 @@ static AnyExprV parse_expr15(Lexer& lex) { AnyExprV rhs = parse_expr17(lex); lhs = createV(loc, operator_name, t, lhs, rhs); if (t == tok_eq || t == tok_neq) { - lhs = maybe_replace_eq_null_with_isNull_call(lhs->as()); + lhs = maybe_replace_eq_null_with_isNull_check(lhs->as()); } } return lhs; diff --git a/tolk/ast-replacer.h b/tolk/ast-replacer.h index c8350747..b4e0de0f 100644 --- a/tolk/ast-replacer.h +++ b/tolk/ast-replacer.h @@ -108,6 +108,8 @@ protected: virtual AnyExprV replace(V v) { return replace_children(v); } virtual AnyExprV replace(V v) { return replace_children(v); } virtual AnyExprV replace(V v) { return replace_children(v); } + virtual AnyExprV replace(V v) { return replace_children(v); } + virtual AnyExprV replace(V v) { return replace_children(v); } // statements virtual AnyV replace(V v) { return replace_children(v); } virtual AnyV replace(V v) { return replace_children(v); } @@ -144,6 +146,8 @@ protected: case ast_binary_operator: return replace(v->as()); case ast_ternary_operator: return replace(v->as()); case ast_cast_as_operator: return replace(v->as()); + case ast_not_null_operator: return replace(v->as()); + case ast_is_null_check: return replace(v->as()); default: throw UnexpectedASTNodeType(v, "ASTReplacerInFunctionBody::replace"); } diff --git a/tolk/ast-replicator.h b/tolk/ast-replicator.h index 02198adb..16bbbeb8 100644 --- a/tolk/ast-replicator.h +++ b/tolk/ast-replicator.h @@ -121,6 +121,12 @@ protected: virtual V clone(V v) { return createV(v->loc, clone(v->get_expr()), clone(v->cast_to_type)); } + virtual V clone(V v) { + return createV(v->loc, clone(v->get_expr())); + } + virtual V clone(V v) { + return createV(v->loc, clone(v->get_expr()), v->is_negated); + } // statements @@ -200,6 +206,8 @@ protected: case ast_binary_operator: return clone(v->as()); case ast_ternary_operator: return clone(v->as()); case ast_cast_as_operator: return clone(v->as()); + case ast_not_null_operator: return clone(v->as()); + case ast_is_null_check: return clone(v->as()); default: throw UnexpectedASTNodeType(v, "ASTReplicatorFunction::clone"); } diff --git a/tolk/ast-stringifier.h b/tolk/ast-stringifier.h index 1211d63f..a7f260de 100644 --- a/tolk/ast-stringifier.h +++ b/tolk/ast-stringifier.h @@ -56,6 +56,8 @@ class ASTStringifier final : public ASTVisitor { {ast_binary_operator, "ast_binary_operator"}, {ast_ternary_operator, "ast_ternary_operator"}, {ast_cast_as_operator, "ast_cast_as_operator"}, + {ast_not_null_operator, "ast_not_null_operator"}, + {ast_is_null_check, "ast_is_null_check"}, // statements {ast_empty_statement, "ast_empty_statement"}, {ast_sequence, "ast_sequence"}, @@ -268,6 +270,8 @@ public: case ast_binary_operator: return handle_vertex(v->as()); case ast_ternary_operator: return handle_vertex(v->as()); case ast_cast_as_operator: return handle_vertex(v->as()); + case ast_not_null_operator: return handle_vertex(v->as()); + case ast_is_null_check: return handle_vertex(v->as()); // statements case ast_empty_statement: return handle_vertex(v->as()); case ast_sequence: return handle_vertex(v->as()); diff --git a/tolk/ast-visitor.h b/tolk/ast-visitor.h index a54cb13b..c0ab7c2f 100644 --- a/tolk/ast-visitor.h +++ b/tolk/ast-visitor.h @@ -109,6 +109,8 @@ protected: virtual void visit(V v) { return visit_children(v); } virtual void visit(V v) { return visit_children(v); } virtual void visit(V v) { return visit_children(v); } + virtual void visit(V v) { return visit_children(v); } + virtual void visit(V v) { return visit_children(v); } // statements virtual void visit(V v) { return visit_children(v); } virtual void visit(V v) { return visit_children(v); } @@ -146,6 +148,8 @@ protected: case ast_binary_operator: return visit(v->as()); case ast_ternary_operator: return visit(v->as()); case ast_cast_as_operator: return visit(v->as()); + case ast_not_null_operator: return visit(v->as()); + case ast_is_null_check: return visit(v->as()); // statements case ast_empty_statement: return visit(v->as()); case ast_sequence: return visit(v->as()); diff --git a/tolk/ast.cpp b/tolk/ast.cpp index 092260ff..4b6e8f90 100644 --- a/tolk/ast.cpp +++ b/tolk/ast.cpp @@ -169,6 +169,10 @@ void Vertex::assign_fun_ref(const FunctionData* fun_ref) { this->fun_ref = fun_ref; } +void Vertex::assign_is_negated(bool is_negated) { + this->is_negated = is_negated; +} + void Vertex::assign_target(const DotTarget& target) { this->target = target; } diff --git a/tolk/ast.h b/tolk/ast.h index d2db49f8..b8fc61ba 100644 --- a/tolk/ast.h +++ b/tolk/ast.h @@ -88,6 +88,8 @@ enum ASTNodeType { ast_binary_operator, ast_ternary_operator, ast_cast_as_operator, + ast_not_null_operator, + ast_is_null_check, // statements ast_empty_statement, ast_sequence, @@ -684,6 +686,32 @@ struct Vertex final : ASTExprUnary { , cast_to_type(cast_to_type) {} }; +template<> +// ast_not_null_operator is non-null assertion: like TypeScript ! or Kotlin !! +// examples: `nullableInt!` / `getNullableBuilder()!` +struct Vertex final : ASTExprUnary { + AnyExprV get_expr() const { return child; } + + Vertex(SrcLocation loc, AnyExprV expr) + : ASTExprUnary(ast_not_null_operator, loc, expr) {} +}; + +template<> +// ast_is_null_check is an artificial vertex for "expr == null" / "expr != null" / same but null on the left +// it's created instead of a general binary expression to emphasize its purpose +struct Vertex final : ASTExprUnary { + bool is_negated; + + AnyExprV get_expr() const { return child; } + + Vertex* mutate() const { return const_cast(this); } + void assign_is_negated(bool is_negated); + + Vertex(SrcLocation loc, AnyExprV expr, bool is_negated) + : ASTExprUnary(ast_is_null_check, loc, expr) + , is_negated(is_negated) {} +}; + // // --------------------------------------------------------- diff --git a/tolk/codegen.cpp b/tolk/codegen.cpp index ad61b8a5..5b2c50cc 100644 --- a/tolk/codegen.cpp +++ b/tolk/codegen.cpp @@ -348,9 +348,9 @@ bool Op::generate_code_step(Stack& stack) { std::vector args0, res; int w_arg = 0; for (const LocalVarData& param : f_sym->parameters) { - w_arg += param.declared_type->calc_width_on_stack(); + w_arg += param.declared_type->get_width_on_stack(); } - int w_ret = f_sym->inferred_return_type->calc_width_on_stack(); + int w_ret = f_sym->inferred_return_type->get_width_on_stack(); tolk_assert(w_ret >= 0 && w_arg >= 0); for (int i = 0; i < w_ret; i++) { res.emplace_back(0); diff --git a/tolk/generics-helpers.cpp b/tolk/generics-helpers.cpp index 7a2dd83f..6cae8da3 100644 --- a/tolk/generics-helpers.cpp +++ b/tolk/generics-helpers.cpp @@ -37,12 +37,38 @@ static TypePtr replace_genericT_with_deduced(TypePtr orig, const GenericsDeclara if (idx == -1) { throw Fatal("can not replace generic " + asT->nameT); } + if (substitutionTs[idx] == nullptr) { + throw GenericDeduceError("can not deduce " + asT->nameT); + } return substitutionTs[idx]; } return child; }); } +GenericSubstitutionsDeduceForCall::GenericSubstitutionsDeduceForCall(const FunctionData* fun_ref) + : fun_ref(fun_ref) { + substitutionTs.resize(fun_ref->genericTs->size()); // filled with nullptr (nothing deduced) +} + +void GenericSubstitutionsDeduceForCall::provide_deducedT(const std::string& nameT, TypePtr deduced) { + if (deduced == TypeDataNullLiteral::create() || deduced->has_unknown_inside()) { + return; // just 'null' doesn't give sensible info + } + + int idx = fun_ref->genericTs->find_nameT(nameT); + if (substitutionTs[idx] == nullptr) { + substitutionTs[idx] = deduced; + } else if (substitutionTs[idx] != deduced) { + throw GenericDeduceError(nameT + " is both " + substitutionTs[idx]->as_human_readable() + " and " + deduced->as_human_readable()); + } +} + +void GenericSubstitutionsDeduceForCall::provide_manually_specified(std::vector&& substitutionTs) { + this->substitutionTs = std::move(substitutionTs); + this->manually_specified = true; +} + // purpose: having `f(value: T)` and call `f(5)`, deduce T = int // generally, there may be many generic Ts for declaration, and many arguments // for every argument, `consider_next_condition()` is called @@ -51,71 +77,67 @@ static TypePtr replace_genericT_with_deduced(TypePtr orig, const GenericsDeclara // - next condition: param_type = `T1`, arg_type = `int`, deduce T1 = int // - next condition: param_type = `(T1, T2)`, arg_type = `(int, slice)`, deduce T1 = int, T2 = slice // for call `f(6, cs, (8, cs))` T1 will be both `slice` and `int`, fired an error -class GenericSubstitutionsDeduceForFunctionCall final { - const FunctionData* fun_ref; - std::vector substitutions; - - void provideDeducedT(const std::string& nameT, TypePtr deduced) { - if (deduced == TypeDataNullLiteral::create() || deduced->has_unknown_inside()) { - return; // just 'null' doesn't give sensible info +void GenericSubstitutionsDeduceForCall::consider_next_condition(TypePtr param_type, TypePtr arg_type) { + if (const auto* asT = param_type->try_as()) { + // `(arg: T)` called as `f([1, 2])` => T is [int, int] + provide_deducedT(asT->nameT, arg_type); + } else if (const auto* p_nullable = param_type->try_as()) { + // `arg: T?` called as `f(nullableInt)` => T is int + if (const auto* a_nullable = arg_type->try_as()) { + consider_next_condition(p_nullable->inner, a_nullable->inner); } - - int idx = fun_ref->genericTs->find_nameT(nameT); - if (substitutions[idx] == nullptr) { - substitutions[idx] = deduced; - } else if (substitutions[idx] != deduced) { - throw std::runtime_error(nameT + " is both " + substitutions[idx]->as_human_readable() + " and " + deduced->as_human_readable()); + // `arg: T?` called as `f(int)` => T is int + else { + consider_next_condition(p_nullable->inner, arg_type); } - } - -public: - explicit GenericSubstitutionsDeduceForFunctionCall(const FunctionData* fun_ref) - : fun_ref(fun_ref) { - substitutions.resize(fun_ref->genericTs->size()); // filled with nullptr (nothing deduced) - } - - void consider_next_condition(TypePtr param_type, TypePtr arg_type) { - if (const auto* asT = param_type->try_as()) { - // `(arg: T)` called as `f([1, 2])` => T is [int, int] - provideDeducedT(asT->nameT, arg_type); - } else if (const auto* p_tensor = param_type->try_as()) { - // `arg: (int, T)` called as `f((5, cs))` => T is slice - if (const auto* a_tensor = arg_type->try_as(); a_tensor && a_tensor->size() == p_tensor->size()) { - for (int i = 0; i < a_tensor->size(); ++i) { - consider_next_condition(p_tensor->items[i], a_tensor->items[i]); - } - } - } else if (const auto* p_tuple = param_type->try_as()) { - // `arg: [int, T]` called as `f([5, cs])` => T is slice - if (const auto* a_tuple = arg_type->try_as(); a_tuple && a_tuple->size() == p_tuple->size()) { - for (int i = 0; i < a_tuple->size(); ++i) { - consider_next_condition(p_tuple->items[i], a_tuple->items[i]); - } - } - } else if (const auto* p_callable = param_type->try_as()) { - // `arg: fun(TArg) -> TResult` called as `f(calcTupleLen)` => TArg is tuple, TResult is int - if (const auto* a_callable = arg_type->try_as(); a_callable && a_callable->params_size() == p_callable->params_size()) { - for (int i = 0; i < a_callable->params_size(); ++i) { - consider_next_condition(p_callable->params_types[i], a_callable->params_types[i]); - } - consider_next_condition(p_callable->return_type, a_callable->return_type); + } else if (const auto* p_tensor = param_type->try_as()) { + // `arg: (int, T)` called as `f((5, cs))` => T is slice + if (const auto* a_tensor = arg_type->try_as(); a_tensor && a_tensor->size() == p_tensor->size()) { + for (int i = 0; i < a_tensor->size(); ++i) { + consider_next_condition(p_tensor->items[i], a_tensor->items[i]); } } - } - - int get_first_not_deduced_idx() const { - for (int i = 0; i < static_cast(substitutions.size()); ++i) { - if (substitutions[i] == nullptr) { - return i; + } else if (const auto* p_tuple = param_type->try_as()) { + // `arg: [int, T]` called as `f([5, cs])` => T is slice + if (const auto* a_tuple = arg_type->try_as(); a_tuple && a_tuple->size() == p_tuple->size()) { + for (int i = 0; i < a_tuple->size(); ++i) { + consider_next_condition(p_tuple->items[i], a_tuple->items[i]); } } - return -1; + } else if (const auto* p_callable = param_type->try_as()) { + // `arg: fun(TArg) -> TResult` called as `f(calcTupleLen)` => TArg is tuple, TResult is int + if (const auto* a_callable = arg_type->try_as(); a_callable && a_callable->params_size() == p_callable->params_size()) { + for (int i = 0; i < a_callable->params_size(); ++i) { + consider_next_condition(p_callable->params_types[i], a_callable->params_types[i]); + } + consider_next_condition(p_callable->return_type, a_callable->return_type); + } } +} - std::vector flush() { - return {std::move(substitutions)}; +TypePtr GenericSubstitutionsDeduceForCall::replace_by_manually_specified(TypePtr param_type) const { + return replace_genericT_with_deduced(param_type, fun_ref->genericTs, substitutionTs); +} + +TypePtr GenericSubstitutionsDeduceForCall::auto_deduce_from_argument(SrcLocation loc, TypePtr param_type, TypePtr arg_type) { + try { + if (!manually_specified) { + consider_next_condition(param_type, arg_type); + } + return replace_genericT_with_deduced(param_type, fun_ref->genericTs, substitutionTs); + } catch (const GenericDeduceError& ex) { + throw ParseError(loc, ex.message + " for generic function `" + fun_ref->as_human_readable() + "`"); } -}; +} + +int GenericSubstitutionsDeduceForCall::get_first_not_deduced_idx() const { + for (int i = 0; i < static_cast(substitutionTs.size()); ++i) { + if (substitutionTs[i] == nullptr) { + return i; + } + } + return -1; +} // clone the body of `f` replacing T everywhere with a substitution // before: `fun f(v: T) { var cp: [T] = [v]; }` @@ -198,28 +220,6 @@ std::string generate_instantiated_name(const std::string& orig_name, const std:: return name; } -td::Result> deduce_substitutionTs_on_generic_func_call(const FunctionData* called_fun, std::vector&& arg_types, TypePtr return_hint) { - try { - GenericSubstitutionsDeduceForFunctionCall deducing(called_fun); - for (const LocalVarData& param : called_fun->parameters) { - if (param.declared_type->has_genericT_inside() && param.param_idx < static_cast(arg_types.size())) { - deducing.consider_next_condition(param.declared_type, arg_types[param.param_idx]); - } - } - int idx = deducing.get_first_not_deduced_idx(); - if (idx != -1 && return_hint && called_fun->declared_return_type->has_genericT_inside()) { - deducing.consider_next_condition(called_fun->declared_return_type, return_hint); - idx = deducing.get_first_not_deduced_idx(); - } - if (idx != -1) { - return td::Status::Error(td::Slice{"can not deduce " + called_fun->genericTs->get_nameT(idx)}); - } - return deducing.flush(); - } catch (const std::runtime_error& ex) { - return td::Status::Error(td::Slice{ex.what()}); - } -} - const FunctionData* instantiate_generic_function(SrcLocation loc, const FunctionData* fun_ref, const std::string& inst_name, std::vector&& substitutionTs) { tolk_assert(fun_ref->genericTs); diff --git a/tolk/generics-helpers.h b/tolk/generics-helpers.h index 2a304f55..0c734c75 100644 --- a/tolk/generics-helpers.h +++ b/tolk/generics-helpers.h @@ -57,8 +57,46 @@ struct GenericsInstantiation { } }; +// this class helps to deduce Ts on the fly +// purpose: having `f(value: T)` and call `f(5)`, deduce T = int +// while analyzing a call, arguments are handled one by one, by `auto_deduce_from_argument()` +// this class also handles manually specified substitutions like `f(5)` +class GenericSubstitutionsDeduceForCall { + const FunctionData* fun_ref; + std::vector substitutionTs; + bool manually_specified = false; + + void provide_deducedT(const std::string& nameT, TypePtr deduced); + void consider_next_condition(TypePtr param_type, TypePtr arg_type); + +public: + explicit GenericSubstitutionsDeduceForCall(const FunctionData* fun_ref); + + bool is_manually_specified() const { + return manually_specified; + } + + void provide_manually_specified(std::vector&& substitutionTs); + TypePtr replace_by_manually_specified(TypePtr param_type) const; + TypePtr auto_deduce_from_argument(SrcLocation loc, TypePtr param_type, TypePtr arg_type); + int get_first_not_deduced_idx() const; + + std::vector&& flush() { + return std::move(substitutionTs); + } +}; + +struct GenericDeduceError final : std::exception { + std::string message; + explicit GenericDeduceError(std::string message) + : message(std::move(message)) { } + + const char* what() const noexcept override { + return message.c_str(); + } +}; + std::string generate_instantiated_name(const std::string& orig_name, const std::vector& substitutions); -td::Result> deduce_substitutionTs_on_generic_func_call(const FunctionData* called_fun, std::vector&& arg_types, TypePtr return_hint); const FunctionData* instantiate_generic_function(SrcLocation loc, const FunctionData* fun_ref, const std::string& inst_name, std::vector&& substitutionTs); } // namespace tolk diff --git a/tolk/pipe-ast-to-legacy.cpp b/tolk/pipe-ast-to-legacy.cpp index be657e25..6f88f617 100644 --- a/tolk/pipe-ast-to-legacy.cpp +++ b/tolk/pipe-ast-to-legacy.cpp @@ -32,12 +32,22 @@ * So, if execution reaches this pass, the input is (almost) correct, and code generation should succeed. * (previously, there was a check for one variable modified twice like `(t.0, t.0) = rhs`, but after changing * execution order of assignment to "first lhs, then lhs", it was removed for several reasons) +* + * A noticeable property for IR generation is "target_type" used to extend/shrink stack. + * Example: `var a: (int,int)? = null`. This `null` has inferred_type "null literal", but target_type "nullable tensor", + * and when it's assigned, it's "expanded" from 1 stack slot to 3 (int + int + null flag). + * Example: `fun analyze(t: (int,int)?)` and a call `analyze((1,2))`. `(1,2)` is `(int,int)` (2 stack slots), + * and when passed to target (3 slots, one for null flag), this null flag is implicitly added (zero value). + * Example: `nullableInt!`; for `nullableInt` inferred_type is `int?`, and target_type is `int` + * (this doesn't lead to stack reorganization, but in case `nullableTensor!` does) + * (inferred_type of `nullableInt!` is `int`, and its target_type depends on its usage). + * The same mechanism will work for union types in the future. */ namespace tolk { class LValContext; -std::vector pre_compile_expr(AnyExprV v, CodeBlob& code, LValContext* lval_ctx = nullptr); +std::vector pre_compile_expr(AnyExprV v, CodeBlob& code, TypePtr target_type = nullptr, LValContext* lval_ctx = nullptr); std::vector pre_compile_symbol(SrcLocation loc, const Symbol* sym, CodeBlob& code, LValContext* lval_ctx); void process_any_statement(AnyV v, CodeBlob& code); @@ -102,7 +112,7 @@ class LValContext { // example: `globalInt = 9`, implicit var is created `$tmp = 9`, and `SetGlob "globalInt" $tmp` is done after struct ModifiedGlobal { const GlobalVarData* glob_ref; - std::vector lval_ir_idx; // typically 1, generally calc_width_on_stack() of global var (tensors) + std::vector lval_ir_idx; // typically 1, generally get_width_on_stack() of global var (tensors) // for 1-slot globals int/cell/slice, assigning to them is just SETGLOB // same for tensors, if they are fully rewritten in an expression: `gTensor = (5,6)` @@ -139,13 +149,13 @@ class LValContext { void apply(CodeBlob& code, SrcLocation loc) const { LValContext local_lval; local_lval.enter_rval_inside_lval(); - std::vector obj_ir_idx = pre_compile_expr(tensor_obj, code, &local_lval); + std::vector obj_ir_idx = pre_compile_expr(tensor_obj, code, nullptr, &local_lval); const TypeDataTensor* t_tensor = tensor_obj->inferred_type->try_as(); tolk_assert(t_tensor); - int stack_width = t_tensor->items[index_at]->calc_width_on_stack(); + int stack_width = t_tensor->items[index_at]->get_width_on_stack(); int stack_offset = 0; for (int i = 0; i < index_at; ++i) { - stack_offset += t_tensor->items[i]->calc_width_on_stack(); + stack_offset += t_tensor->items[i]->get_width_on_stack(); } std::vector field_ir_idx = {obj_ir_idx.begin() + stack_offset, obj_ir_idx.begin() + stack_offset + stack_width}; tolk_assert(field_ir_idx.size() == lval_ir_idx.size()); @@ -167,7 +177,7 @@ class LValContext { void apply(CodeBlob& code, SrcLocation loc) const { LValContext local_lval; local_lval.enter_rval_inside_lval(); - std::vector tuple_ir_idx = pre_compile_expr(tuple_obj, code, &local_lval); + std::vector tuple_ir_idx = pre_compile_expr(tuple_obj, code, nullptr, &local_lval); std::vector index_ir_idx = code.create_tmp_var(TypeDataInt::create(), loc, "(tuple-idx)"); code.emplace_back(loc, Op::_IntConst, index_ir_idx, td::make_refint(index_at)); @@ -245,29 +255,39 @@ public: } }; +// given `{some_expr}!`, return some_expr +static AnyExprV unwrap_not_null_operator(AnyExprV v) { + while (auto v_notnull = v->try_as()) { + v = v_notnull->get_expr(); + } + return v; +} + // given `{some_expr}.{i}`, check it for pattern `some_var.0` / `some_var.0.1` / etc. // return some_var if satisfies (it may be a local or a global var, a tensor or a tuple) // return nullptr otherwise: `f().0` / `(v = rhs).0` / `some_var.method().0` / etc. static V calc_sink_leftmost_obj(V v) { - AnyExprV leftmost_obj = v->get_obj(); + AnyExprV leftmost_obj = unwrap_not_null_operator(v->get_obj()); while (auto v_dot = leftmost_obj->try_as()) { if (!v_dot->is_target_indexed_access()) { break; } - leftmost_obj = v_dot->get_obj(); + leftmost_obj = unwrap_not_null_operator(v_dot->get_obj()); } return leftmost_obj->type == ast_reference ? leftmost_obj->as() : nullptr; } static std::vector> pre_compile_tensor_inner(CodeBlob& code, const std::vector& args, - LValContext* lval_ctx) { + const TypeDataTensor* tensor_target_type, LValContext* lval_ctx) { const int n = static_cast(args.size()); if (n == 0) { // just `()` return {}; } + tolk_assert(!tensor_target_type || tensor_target_type->size() == n); if (n == 1) { // just `(x)`: even if x is modified (e.g. `f(x=x+2)`), there are no next arguments - return {pre_compile_expr(args[0], code, lval_ctx)}; + TypePtr child_target_type = tensor_target_type ? tensor_target_type->items[0] : nullptr; + return {pre_compile_expr(args[0], code, child_target_type, lval_ctx)}; } // the purpose is to handle such cases: `return (x, x += y, x)` @@ -321,7 +341,8 @@ static std::vector> pre_compile_tensor_inner(CodeBlob& co WatchingVarList watched_vars(n); for (int arg_idx = 0; arg_idx < n; ++arg_idx) { - std::vector vars_of_ith_arg = pre_compile_expr(args[arg_idx], code, lval_ctx); + TypePtr child_target_type = tensor_target_type ? tensor_target_type->items[arg_idx] : nullptr; + std::vector vars_of_ith_arg = pre_compile_expr(args[arg_idx], code, child_target_type, lval_ctx); watched_vars.add_and_watch_modifications(std::move(vars_of_ith_arg), code); } return watched_vars.clear_and_stop_watching(); @@ -329,7 +350,13 @@ static std::vector> pre_compile_tensor_inner(CodeBlob& co static std::vector pre_compile_tensor(CodeBlob& code, const std::vector& args, LValContext* lval_ctx = nullptr) { - std::vector> res_lists = pre_compile_tensor_inner(code, args, lval_ctx); + std::vector types_list; + types_list.reserve(args.size()); + for (AnyExprV item : args) { + types_list.push_back(item->inferred_type); + } + const TypeDataTensor* tensor_target_type = TypeDataTensor::create(std::move(types_list))->try_as(); + std::vector> res_lists = pre_compile_tensor_inner(code, args, tensor_target_type, lval_ctx); std::vector res; for (const std::vector& list : res_lists) { res.insert(res.end(), list.cbegin(), list.cend()); @@ -340,20 +367,23 @@ static std::vector pre_compile_tensor(CodeBlob& code, const std::vect static std::vector pre_compile_let(CodeBlob& code, AnyExprV lhs, AnyExprV rhs, SrcLocation loc) { // [lhs] = [rhs]; since type checking is ok, it's the same as "lhs = rhs" if (lhs->type == ast_typed_tuple && rhs->type == ast_typed_tuple) { + // note: there are no type transitions (adding nullability flag, etc.), since only 1-slot elements allowed in tuples LValContext local_lval; std::vector left = pre_compile_tensor(code, lhs->as()->get_items(), &local_lval); vars_modification_watcher.trigger_callbacks(left, loc); std::vector right = pre_compile_tensor(code, rhs->as()->get_items()); code.emplace_back(loc, Op::_Let, left, right); local_lval.after_let(std::move(left), code, loc); - return right; + std::vector rvect = code.create_tmp_var(TypeDataTuple::create(), loc, "(tuple)"); + code.emplace_back(lhs->loc, Op::_Tuple, rvect, std::move(right)); + return rvect; } // [lhs] = rhs; it's un-tuple to N left vars if (lhs->type == ast_typed_tuple) { LValContext local_lval; std::vector left = pre_compile_tensor(code, lhs->as()->get_items(), &local_lval); vars_modification_watcher.trigger_callbacks(left, loc); - std::vector right = pre_compile_expr(rhs, code); + std::vector right = pre_compile_expr(rhs, code, nullptr); const TypeDataTypedTuple* inferred_tuple = rhs->inferred_type->try_as(); std::vector types_list = inferred_tuple->items; std::vector rvect = code.create_tmp_var(TypeDataTensor::create(std::move(types_list)), rhs->loc, "(unpack-tuple)"); @@ -364,17 +394,17 @@ static std::vector pre_compile_let(CodeBlob& code, AnyExprV lhs, AnyE } // small optimization: `var x = rhs` or `local_var = rhs` (90% cases), LValContext not needed actually if (lhs->type == ast_local_var_lhs || (lhs->type == ast_reference && lhs->as()->sym->try_as())) { - std::vector left = pre_compile_expr(lhs, code); // effectively, local_var->ir_idx + std::vector left = pre_compile_expr(lhs, code, nullptr); // effectively, local_var->ir_idx vars_modification_watcher.trigger_callbacks(left, loc); - std::vector right = pre_compile_expr(rhs, code); + std::vector right = pre_compile_expr(rhs, code, lhs->inferred_type); code.emplace_back(loc, Op::_Let, std::move(left), right); return right; } // lhs = rhs LValContext local_lval; - std::vector left = pre_compile_expr(lhs, code, &local_lval); + std::vector left = pre_compile_expr(lhs, code, nullptr, &local_lval); vars_modification_watcher.trigger_callbacks(left, loc); - std::vector right = pre_compile_expr(rhs, code); + std::vector right = pre_compile_expr(rhs, code, lhs->inferred_type); code.emplace_back(loc, Op::_Let, left, right); local_lval.after_let(std::move(left), code, loc); return right; @@ -390,6 +420,143 @@ static std::vector gen_op_call(CodeBlob& code, TypePtr ret_type, SrcL return rvect; } +// "Transition to target (runtime) type" is the following process. +// Imagine `fun analyze(t: (int,int)?)` and a call `analyze((1,2))`. +// `(1,2)` (inferred_type) is 2 stack slots, but `t` (target_type) is 3 (one for null-flag). +// So, this null flag should be implicitly added (zero value, since a variable is not null). +// Another example: `var t: (int, int)? = null`. +// `null` (inferred_type) is 1 stack slots, but target_type is 3, we should add 2 nulls. +// Another example: `var t1 = (1, null); var t2: (int, (int,int)?) = t1;`. +// Then t1's rvect is 2 vars (1 and null), but t1's `null` should be converted to 3 stack slots (resulting in 4 total). +// The same mechanism will work for union types in the future. +// Here rvect is a list of IR vars for inferred_type, probably patched due to target_type. +GNU_ATTRIBUTE_NOINLINE +static std::vector transition_expr_to_runtime_type_impl(std::vector&& rvect, CodeBlob& code, TypePtr original_type, TypePtr target_type, SrcLocation loc) { + // pass `T` to `T` + // could occur for passing tensor `(..., T, ...)` to `(..., T, ...)` while traversing tensor's components + if (target_type == original_type) { + return rvect; + } + + int target_w = target_type->get_width_on_stack(); + const TypeDataNullable* t_nullable = target_type->try_as(); + const TypeDataNullable* o_nullable = original_type->try_as(); + + // pass `null` to `T?` + // for primitives like `int?`, no changes in rvect, null occupies the same TVM slot + // for tensors like `(int,int)?`, `null` is represented as N nulls + 1 null flag, insert N nulls + if (t_nullable && original_type == TypeDataNullLiteral::create()) { + tolk_assert(rvect.size() == 1); + if (target_w > 1) { + const FunctionData* builtin_sym = lookup_global_symbol("__null")->as(); + rvect.reserve(target_w + 1); + for (int i = 1; i < target_w - 1; ++i) { + std::vector ith_null = code.create_tmp_var(TypeDataNullLiteral::create(), loc, "(null-literal)"); + code.emplace_back(loc, Op::_Call, ith_null, std::vector{}, builtin_sym); + rvect.push_back(ith_null[0]); + } + std::vector null_flag_ir = code.create_tmp_var(TypeDataInt::create(), loc, "(NNFlag)"); + var_idx_t null_flag_ir_idx = null_flag_ir[0]; + code.emplace_back(loc, Op::_IntConst, std::move(null_flag_ir), td::make_refint(0)); + rvect.push_back(null_flag_ir_idx); + } + return rvect; + } + // pass `T` to `T?` + // for primitives like `int?`, no changes in rvect: `int` and `int?` occupy the same TVM slot (null is represented as NULL TVM value) + // for passing `(int, int)` to `(int, int)?` / `(int, null)` to `(int, (int,int)?)?`, add a null flag equals to 0 + if (t_nullable && !o_nullable) { + if (target_w > 1) { + rvect = transition_expr_to_runtime_type_impl(std::move(rvect), code, original_type, t_nullable->inner, loc); + tolk_assert(target_w == static_cast(rvect.size() + 1)); + std::vector null_flag_ir = code.create_tmp_var(TypeDataInt::create(), loc, "(NNFlag)"); + var_idx_t null_flag_ir_idx = null_flag_ir[0]; + code.emplace_back(loc, Op::_IntConst, std::move(null_flag_ir), td::make_refint(-1)); + rvect.push_back(null_flag_ir_idx); + } + return rvect; + } + // pass `T1?` to `T2?` + // for example, `int8?` to `int16?` + // transition inner types, leaving nullable flag unchanged for tensors + if (t_nullable && o_nullable) { + if (target_w > 1) { + var_idx_t null_flag_ir_idx = rvect.back(); + rvect.pop_back(); + rvect = transition_expr_to_runtime_type_impl(std::move(rvect), code, o_nullable->inner, t_nullable->inner, loc); + rvect.push_back(null_flag_ir_idx); + } + return rvect; + } + // pass `T?` to `T` + // it may occur due to operator `!` or smart cast + // for primitives like `int?`, no changes in rvect + // for passing `(int, int)?` to `(int, int)`, drop the null flag from the tail + if (!t_nullable && original_type->try_as() && original_type->try_as()->inner == target_type) { + if (original_type->get_width_on_stack() > 1) { + rvect.pop_back(); + } + return rvect; + } + // pass `bool` to `int` + // in code, it's done via `as` operator, like `boolVar as int` + // no changes in rvect, boolVar is guaranteed to be -1 or 0 at TVM level + if (target_type == TypeDataInt::create() && original_type == TypeDataBool::create()) { + return rvect; + } + // pass something to `unknown` + // probably, it comes from `_ = rhs`, type of `_` is unknown, it's target_type of rhs + // no changes in rvect + if (target_type == TypeDataUnknown::create()) { + return rvect; + } + // pass `unknown` to something + // probably, it comes from `arg` in exception, it's inferred as `unknown` and could be cast to any value + if (original_type == TypeDataUnknown::create()) { + tolk_assert(rvect.size() == 1); + return rvect; + } + // pass tensor to tensor, e.g. `(1, null)` to `(int, slice?)` / `(1, null)` to `(int, (int,int)?)` + // every element of rhs tensor should be transitioned + if (target_type->try_as() && original_type->try_as()) { + const TypeDataTensor* target_tensor = target_type->try_as(); + const TypeDataTensor* inferred_tensor = original_type->try_as(); + tolk_assert(target_tensor->size() == inferred_tensor->size()); + tolk_assert(inferred_tensor->get_width_on_stack() == static_cast(rvect.size())); + std::vector result_rvect; + result_rvect.reserve(target_w); + int stack_offset = 0; + for (int i = 0; i < inferred_tensor->size(); ++i) { + int ith_w = inferred_tensor->items[i]->get_width_on_stack(); + std::vector rvect_i{rvect.begin() + stack_offset, rvect.begin() + stack_offset + ith_w}; + std::vector result_i = transition_expr_to_runtime_type_impl(std::move(rvect_i), code, inferred_tensor->items[i], target_tensor->items[i], loc); + result_rvect.insert(result_rvect.end(), result_i.begin(), result_i.end()); + stack_offset += ith_w; + } + return result_rvect; + } + // pass tuple to tuple, e.g. `[1, null]` to `[int, int?]` / `[1, null]` to `[int, [int?,int?]?]` + // to changes to rvect, since tuples contain only 1-slot elements + if (target_type->try_as() && original_type->try_as()) { + tolk_assert(target_type->get_width_on_stack() == original_type->get_width_on_stack()); + return rvect; + } + + throw Fatal("unhandled transition_expr_to_runtime_type_impl() combination"); +} + +// invoke the function above only if potentially needed to +// (if an expression is targeted to another type) +#ifndef TOLK_DEBUG +GNU_ATTRIBUTE_ALWAYS_INLINE +#endif +static std::vector transition_to_target_type(std::vector&& rvect, CodeBlob& code, TypePtr target_type, AnyExprV v) { + if (target_type != nullptr && target_type != v->inferred_type) { + rvect = transition_expr_to_runtime_type_impl(std::move(rvect), code, v->inferred_type, target_type, v->loc); + } + return rvect; +} + std::vector pre_compile_symbol(SrcLocation loc, const Symbol* sym, CodeBlob& code, LValContext* lval_ctx) { if (const auto* glob_ref = sym->try_as()) { @@ -426,37 +593,47 @@ std::vector pre_compile_symbol(SrcLocation loc, const Symbol* sym, Co } if (const auto* var_ref = sym->try_as()) { #ifdef TOLK_DEBUG - tolk_assert(static_cast(var_ref->ir_idx.size()) == var_ref->declared_type->calc_width_on_stack()); + tolk_assert(static_cast(var_ref->ir_idx.size()) == var_ref->declared_type->get_width_on_stack()); #endif return var_ref->ir_idx; } throw Fatal("pre_compile_symbol"); } -static std::vector process_assignment(V v, CodeBlob& code) { +static std::vector process_reference(V v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) { + std::vector rvect = pre_compile_symbol(v->loc, v->sym, code, lval_ctx); + return transition_to_target_type(std::move(rvect), code, target_type, v); +} + +static std::vector process_assignment(V v, CodeBlob& code, TypePtr target_type) { if (auto lhs_decl = v->get_lhs()->try_as()) { - return pre_compile_let(code, lhs_decl->get_expr(), v->get_rhs(), v->loc); + std::vector rvect = pre_compile_let(code, lhs_decl->get_expr(), v->get_rhs(), v->loc); + return transition_to_target_type(std::move(rvect), code, target_type, v); } else { - return pre_compile_let(code, v->get_lhs(), v->get_rhs(), v->loc); + std::vector rvect = pre_compile_let(code, v->get_lhs(), v->get_rhs(), v->loc); + return transition_to_target_type(std::move(rvect), code, target_type, v); } } -static std::vector process_set_assign(V v, CodeBlob& code) { +static std::vector process_set_assign(V v, CodeBlob& code, TypePtr target_type) { // for "a += b", emulate "a = a + b" // seems not beautiful, but it works; probably, this transformation should be done at AST level in advance std::string_view calc_operator = v->operator_name; // "+" for operator += auto v_apply = createV(v->loc, calc_operator, static_cast(v->tok - 1), v->get_lhs(), v->get_rhs()); v_apply->assign_inferred_type(v->inferred_type); v_apply->assign_fun_ref(v->fun_ref); - return pre_compile_let(code, v->get_lhs(), v_apply, v->loc); + + std::vector rvect = pre_compile_let(code, v->get_lhs(), v_apply, v->loc); + return transition_to_target_type(std::move(rvect), code, target_type, v); } -static std::vector process_binary_operator(V v, CodeBlob& code) { +static std::vector process_binary_operator(V v, CodeBlob& code, TypePtr target_type) { TokenType t = v->tok; if (v->fun_ref) { // almost all operators, fun_ref was assigned at type inferring std::vector args_vars = pre_compile_tensor(code, {v->get_lhs(), v->get_rhs()}); - return gen_op_call(code, v->inferred_type, v->loc, std::move(args_vars), v->fun_ref, "(binary-op)"); + std::vector rvect = gen_op_call(code, v->inferred_type, v->loc, std::move(args_vars), v->fun_ref, "(binary-op)"); + return transition_to_target_type(std::move(rvect), code, target_type, v); } if (t == tok_logical_and || t == tok_logical_or) { // do the following transformations: @@ -469,42 +646,86 @@ static std::vector process_binary_operator(V v, auto v_b_ne_0 = createV(v->loc, "!=", tok_neq, v->get_rhs(), v_0); v_b_ne_0->mutate()->assign_inferred_type(TypeDataInt::create()); v_b_ne_0->mutate()->assign_fun_ref(lookup_global_symbol("_!=_")->as()); - std::vector cond = pre_compile_expr(v->get_lhs(), code); + std::vector cond = pre_compile_expr(v->get_lhs(), code, nullptr); tolk_assert(cond.size() == 1); - std::vector rvect = code.create_tmp_var(v->inferred_type, v->loc, "(cond)"); + std::vector rvect = code.create_tmp_var(v->inferred_type, v->loc, "(ternary)"); Op& if_op = code.emplace_back(v->loc, Op::_If, cond); code.push_set_cur(if_op.block0); - code.emplace_back(v->loc, Op::_Let, rvect, pre_compile_expr(t == tok_logical_and ? v_b_ne_0 : v_1, code)); + code.emplace_back(v->loc, Op::_Let, rvect, pre_compile_expr(t == tok_logical_and ? v_b_ne_0 : v_1, code, nullptr)); code.close_pop_cur(v->loc); code.push_set_cur(if_op.block1); - code.emplace_back(v->loc, Op::_Let, rvect, pre_compile_expr(t == tok_logical_and ? v_0 : v_b_ne_0, code)); + code.emplace_back(v->loc, Op::_Let, rvect, pre_compile_expr(t == tok_logical_and ? v_0 : v_b_ne_0, code, nullptr)); code.close_pop_cur(v->loc); - return rvect; + return transition_to_target_type(std::move(rvect), code, target_type, v); } throw UnexpectedASTNodeType(v, "process_binary_operator"); } -static std::vector process_unary_operator(V v, CodeBlob& code) { - std::vector args_vars = pre_compile_tensor(code, {v->get_rhs()}); - return gen_op_call(code, v->inferred_type, v->loc, std::move(args_vars), v->fun_ref, "(unary-op)"); +static std::vector process_unary_operator(V v, CodeBlob& code, TypePtr target_type) { + std::vector rhs_vars = pre_compile_expr(v->get_rhs(), code, nullptr); + std::vector rvect = gen_op_call(code, v->inferred_type, v->loc, std::move(rhs_vars), v->fun_ref, "(unary-op)"); + return transition_to_target_type(std::move(rvect), code, target_type, v); } -static std::vector process_ternary_operator(V v, CodeBlob& code) { - std::vector cond = pre_compile_expr(v->get_cond(), code); +static std::vector process_ternary_operator(V v, CodeBlob& code, TypePtr target_type) { + std::vector cond = pre_compile_expr(v->get_cond(), code, nullptr); tolk_assert(cond.size() == 1); std::vector rvect = code.create_tmp_var(v->inferred_type, v->loc, "(cond)"); Op& if_op = code.emplace_back(v->loc, Op::_If, cond); code.push_set_cur(if_op.block0); - code.emplace_back(v->get_when_true()->loc, Op::_Let, rvect, pre_compile_expr(v->get_when_true(), code)); + code.emplace_back(v->get_when_true()->loc, Op::_Let, rvect, pre_compile_expr(v->get_when_true(), code, target_type)); code.close_pop_cur(v->get_when_true()->loc); code.push_set_cur(if_op.block1); - code.emplace_back(v->get_when_false()->loc, Op::_Let, rvect, pre_compile_expr(v->get_when_false(), code)); + code.emplace_back(v->get_when_false()->loc, Op::_Let, rvect, pre_compile_expr(v->get_when_false(), code, target_type)); code.close_pop_cur(v->get_when_false()->loc); - return rvect; + return rvect; // no transition to target type, since every branch already transitions } -static std::vector process_dot_access(V v, CodeBlob& code, LValContext* lval_ctx) { +static std::vector process_cast_as_operator(V v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) { + TypePtr child_target_type = v->cast_to_type; + std::vector rvect = pre_compile_expr(v->get_expr(), code, child_target_type, lval_ctx); + return transition_to_target_type(std::move(rvect), code, target_type, v); +} + +static std::vector process_not_null_operator(V v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) { + TypePtr child_target_type = v->get_expr()->inferred_type; + if (const auto* as_nullable = child_target_type->try_as()) { + child_target_type = as_nullable->inner; + } + std::vector rvect = pre_compile_expr(v->get_expr(), code, child_target_type, lval_ctx); + return transition_to_target_type(std::move(rvect), code, target_type, v); +} + +static std::vector process_is_null_check(V v, CodeBlob& code, TypePtr target_type) { + std::vector expr_ir_idx = pre_compile_expr(v->get_expr(), code, nullptr); + std::vector isnull_ir_idx = code.create_tmp_var(TypeDataBool::create(), v->loc, "(is-null)"); + TypePtr expr_type = v->get_expr()->inferred_type; + + if (expr_type->try_as()) { + bool has_separate_null_flag = expr_ir_idx.size() > 1; + if (has_separate_null_flag) { + std::vector zero_ir_idx = code.create_tmp_var(TypeDataInt::create(), v->loc, "(zero)"); + code.emplace_back(v->loc, Op::_IntConst, zero_ir_idx, td::make_refint(0)); + const FunctionData* eq_sym = lookup_global_symbol("_==_")->try_as(); + code.emplace_back(v->loc, Op::_Call, isnull_ir_idx, std::vector{expr_ir_idx.back(), zero_ir_idx[0]}, eq_sym); + } else { + const FunctionData* builtin_sym = lookup_global_symbol("__isNull")->try_as(); + code.emplace_back(v->loc, Op::_Call, isnull_ir_idx, expr_ir_idx, builtin_sym); + } + } else { + bool always_null = expr_type == TypeDataNullLiteral::create(); + code.emplace_back(v->loc, Op::_IntConst, isnull_ir_idx, td::make_refint(always_null ? -1 : 0)); + } + + if (v->is_negated) { + const FunctionData* not_sym = lookup_global_symbol("!b_")->try_as(); + code.emplace_back(v->loc, Op::_Call, isnull_ir_idx, std::vector{isnull_ir_idx}, not_sym); + } + return transition_to_target_type(std::move(isnull_ir_idx), code, target_type, v); +} + +static std::vector process_dot_access(V v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) { // it's NOT a method call `t.tupleSize()` (since such cases are handled by process_function_call) // it's `t.0`, `getUser().id`, and `t.tupleSize` (as a reference, not as a call) if (!v->is_target_fun_ref()) { @@ -521,13 +742,14 @@ static std::vector process_dot_access(V v, CodeBlob& } } // since a tensor of N elems are N vars on a stack actually, calculate offset - std::vector lhs_vars = pre_compile_expr(v->get_obj(), code, lval_ctx); - int stack_width = t_tensor->items[index_at]->calc_width_on_stack(); + std::vector lhs_vars = pre_compile_expr(v->get_obj(), code, nullptr, lval_ctx); + int stack_width = t_tensor->items[index_at]->get_width_on_stack(); int stack_offset = 0; for (int i = 0; i < index_at; ++i) { - stack_offset += t_tensor->items[i]->calc_width_on_stack(); + stack_offset += t_tensor->items[i]->get_width_on_stack(); } - return {lhs_vars.begin() + stack_offset, lhs_vars.begin() + stack_offset + stack_width}; + std::vector rvect{lhs_vars.begin() + stack_offset, lhs_vars.begin() + stack_offset + stack_width}; + return transition_to_target_type(std::move(rvect), code, target_type, v); } // `tupleVar.0` if (obj_type->try_as() || obj_type->try_as()) { @@ -548,7 +770,9 @@ static std::vector process_dot_access(V v, CodeBlob& if (lval_ctx && calc_sink_leftmost_obj(v)) { // `tupleVar.0.1 = rhs`, then `tupleVar.0` is rval inside lval lval_ctx->capture_tuple_index_modification(v->get_obj(), index_at, field_ir_idx); } - return field_ir_idx; + // like tensor index, `tupleVar.1` also might be smart cast, for example we're in `if (tupleVar.1 != null)` + // but since tuple's elements are only 1-slot width (no tensors and unions), no stack transformations required + return transition_to_target_type(std::move(field_ir_idx), code, target_type, v); } tolk_assert(false); } @@ -557,26 +781,36 @@ static std::vector process_dot_access(V v, CodeBlob& // (currently, nothing except a global function can be referenced, no object-scope methods exist) const FunctionData* fun_ref = std::get(v->target); tolk_assert(fun_ref); - return pre_compile_symbol(v->loc, fun_ref, code, lval_ctx); + std::vector rvect = pre_compile_symbol(v->loc, fun_ref, code, lval_ctx); + return transition_to_target_type(std::move(rvect), code, target_type, v); } -static std::vector process_function_call(V v, CodeBlob& code) { +static std::vector process_function_call(V v, CodeBlob& code, TypePtr target_type) { // v is `globalF(args)` / `globalF(args)` / `obj.method(args)` / `local_var(args)` / `getF()(args)` const FunctionData* fun_ref = v->fun_maybe; if (!fun_ref) { + // it's `local_var(args)`, treat args like a tensor: + // 1) when variables are modified like `local_var(x, x += 2, x)`, regular mechanism of watching automatically works + // 2) when `null` is passed to `(int, int)?`, or any other type transitions, it automatically works std::vector args; args.reserve(v->get_num_args()); for (int i = 0; i < v->get_num_args(); ++i) { args.push_back(v->get_arg(i)->get_expr()); } - std::vector args_vars = pre_compile_tensor(code, args); - std::vector tfunc = pre_compile_expr(v->get_callee(), code); + std::vector params_types = v->get_callee()->inferred_type->try_as()->params_types; + const TypeDataTensor* tensor_tt = TypeDataTensor::create(std::move(params_types))->try_as(); + std::vector> vars_per_arg = pre_compile_tensor_inner(code, args, tensor_tt, nullptr); + std::vector args_vars; + for (const std::vector& list : vars_per_arg) { + args_vars.insert(args_vars.end(), list.cbegin(), list.cend()); + } + std::vector tfunc = pre_compile_expr(v->get_callee(), code, nullptr); tolk_assert(tfunc.size() == 1); args_vars.push_back(tfunc[0]); std::vector rvect = code.create_tmp_var(v->inferred_type, v->loc, "(call-ind)"); Op& op = code.emplace_back(v->loc, Op::_CallInd, rvect, std::move(args_vars)); op.set_impure_flag(); - return rvect; + return transition_to_target_type(std::move(rvect), code, target_type, v); } int delta_self = v->is_dot_call(); @@ -593,7 +827,11 @@ static std::vector process_function_call(V v, Code for (int i = 0; i < v->get_num_args(); ++i) { args.push_back(v->get_arg(i)->get_expr()); } - std::vector> vars_per_arg = pre_compile_tensor_inner(code, args, nullptr); + // the purpose of tensor_tt ("tensor target type") is to transition `null` to `(int, int)?` and so on + // the purpose of calling `pre_compile_tensor_inner` is to have 0-th IR vars to handle return self + std::vector params_types = fun_ref->inferred_full_type->try_as()->params_types; + const TypeDataTensor* tensor_tt = TypeDataTensor::create(std::move(params_types))->try_as(); + std::vector> vars_per_arg = pre_compile_tensor_inner(code, args, tensor_tt, nullptr); TypePtr op_call_type = v->inferred_type; TypePtr real_ret_type = v->inferred_type; @@ -607,7 +845,7 @@ static std::vector process_function_call(V v, Code std::vector types_list; for (int i = 0; i < delta_self + v->get_num_args(); ++i) { if (fun_ref->parameters[i].is_mutate_parameter()) { - types_list.push_back(args[i]->inferred_type); + types_list.push_back(fun_ref->parameters[i].declared_type); } } types_list.push_back(real_ret_type); @@ -628,7 +866,7 @@ static std::vector process_function_call(V v, Code AnyExprV arg_i = obj_leftmost && i == 0 ? obj_leftmost : args[i]; tolk_assert(arg_i->is_lvalue || i == 0); if (arg_i->is_lvalue) { - std::vector ith_var_idx = pre_compile_expr(arg_i, code, &local_lval); + std::vector ith_var_idx = pre_compile_expr(arg_i, code, nullptr, &local_lval); left.insert(left.end(), ith_var_idx.begin(), ith_var_idx.end()); } else { left.insert(left.end(), vars_per_arg[0].begin(), vars_per_arg[0].end()); @@ -645,36 +883,39 @@ static std::vector process_function_call(V v, Code if (obj_leftmost && fun_ref->does_return_self()) { if (obj_leftmost->is_lvalue) { // to handle if obj is global var, potentially re-assigned inside a chain - rvect_apply = pre_compile_expr(obj_leftmost, code); + rvect_apply = pre_compile_expr(obj_leftmost, code, nullptr); } else { // temporary object, not lvalue, pre_compile_expr rvect_apply = vars_per_arg[0]; } } - return rvect_apply; + return transition_to_target_type(std::move(rvect_apply), code, target_type, v); } -static std::vector process_tensor(V v, CodeBlob& code, LValContext* lval_ctx) { - return pre_compile_tensor(code, v->get_items(), lval_ctx); +static std::vector process_tensor(V v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) { + // tensor is compiled "as is", for example `(1, null)` occupies 2 slots + // and if assigned/passed to something other, like `(int, (int,int)?)`, a whole tensor is transitioned, it works + std::vector rvect = pre_compile_tensor(code, v->get_items(), lval_ctx); + return transition_to_target_type(std::move(rvect), code, target_type, v); } -static std::vector process_typed_tuple(V v, CodeBlob& code, LValContext* lval_ctx) { +static std::vector process_typed_tuple(V v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) { if (lval_ctx) { // todo some time, make "var (a, [b,c]) = (1, [2,3])" work v->error("[...] can not be used as lvalue here"); } std::vector left = code.create_tmp_var(v->inferred_type, v->loc, "(pack-tuple)"); std::vector right = pre_compile_tensor(code, v->get_items(), lval_ctx); code.emplace_back(v->loc, Op::_Tuple, left, std::move(right)); - return left; + return transition_to_target_type(std::move(left), code, target_type, v); } -static std::vector process_int_const(V v, CodeBlob& code) { +static std::vector process_int_const(V v, CodeBlob& code, TypePtr target_type) { std::vector rvect = code.create_tmp_var(v->inferred_type, v->loc, "(int-const)"); code.emplace_back(v->loc, Op::_IntConst, rvect, v->intval); - return rvect; + return transition_to_target_type(std::move(rvect), code, target_type, v); } -static std::vector process_string_const(V v, CodeBlob& code) { +static std::vector process_string_const(V v, CodeBlob& code, TypePtr target_type) { ConstantValue value = eval_const_init_value(v); std::vector rvect = code.create_tmp_var(v->inferred_type, v->loc, "(str-const)"); if (value.is_int()) { @@ -682,27 +923,31 @@ static std::vector process_string_const(V v, CodeBl } else { code.emplace_back(v->loc, Op::_SliceConst, rvect, value.as_slice()); } - return rvect; + return transition_to_target_type(std::move(rvect), code, target_type, v); } -static std::vector process_bool_const(V v, CodeBlob& code) { +static std::vector process_bool_const(V v, CodeBlob& code, TypePtr target_type) { const FunctionData* builtin_sym = lookup_global_symbol(v->bool_val ? "__true" : "__false")->as(); - return gen_op_call(code, v->inferred_type, v->loc, {}, builtin_sym, "(bool-const)"); + std::vector rvect = gen_op_call(code, v->inferred_type, v->loc, {}, builtin_sym, "(bool-const)"); + return transition_to_target_type(std::move(rvect), code, target_type, v); } -static std::vector process_null_keyword(V v, CodeBlob& code) { +static std::vector process_null_keyword(V v, CodeBlob& code, TypePtr target_type) { const FunctionData* builtin_sym = lookup_global_symbol("__null")->as(); - return gen_op_call(code, v->inferred_type, v->loc, {}, builtin_sym, "(null-literal)"); + std::vector rvect = gen_op_call(code, v->inferred_type, v->loc, {}, builtin_sym, "(null-literal)"); + return transition_to_target_type(std::move(rvect), code, target_type, v); } -static std::vector process_local_var(V v, CodeBlob& code) { +static std::vector process_local_var(V v, CodeBlob& code, TypePtr target_type) { if (v->marked_as_redef) { - return pre_compile_symbol(v->loc, v->var_ref, code, nullptr); + std::vector rvect = pre_compile_symbol(v->loc, v->var_ref, code, nullptr); + return transition_to_target_type(std::move(rvect), code, target_type, v); } tolk_assert(v->var_ref->ir_idx.empty()); v->var_ref->mutate()->assign_ir_idx(code.create_var(v->inferred_type, v->loc, v->var_ref->name)); - return v->var_ref->ir_idx; + std::vector rvect = v->var_ref->ir_idx; + return transition_to_target_type(std::move(rvect), code, target_type, v); } static std::vector process_local_vars_declaration(V, CodeBlob&) { @@ -716,42 +961,46 @@ static std::vector process_underscore(V v, CodeBlob& return code.create_tmp_var(v->inferred_type, v->loc, "(underscore)"); } -std::vector pre_compile_expr(AnyExprV v, CodeBlob& code, LValContext* lval_ctx) { +std::vector pre_compile_expr(AnyExprV v, CodeBlob& code, TypePtr target_type, LValContext* lval_ctx) { switch (v->type) { case ast_reference: - return pre_compile_symbol(v->loc, v->as()->sym, code, lval_ctx); + return process_reference(v->as(), code, target_type, lval_ctx); case ast_assign: - return process_assignment(v->as(), code); + return process_assignment(v->as(), code, target_type); case ast_set_assign: - return process_set_assign(v->as(), code); + return process_set_assign(v->as(), code, target_type); case ast_binary_operator: - return process_binary_operator(v->as(), code); + return process_binary_operator(v->as(), code, target_type); case ast_unary_operator: - return process_unary_operator(v->as(), code); + return process_unary_operator(v->as(), code, target_type); case ast_ternary_operator: - return process_ternary_operator(v->as(), code); + return process_ternary_operator(v->as(), code, target_type); case ast_cast_as_operator: - return pre_compile_expr(v->as()->get_expr(), code, lval_ctx); + return process_cast_as_operator(v->as(), code, target_type, lval_ctx); + case ast_not_null_operator: + return process_not_null_operator(v->as(), code, target_type, lval_ctx); + case ast_is_null_check: + return process_is_null_check(v->as(), code, target_type); case ast_dot_access: - return process_dot_access(v->as(), code, lval_ctx); + return process_dot_access(v->as(), code, target_type, lval_ctx); case ast_function_call: - return process_function_call(v->as(), code); + return process_function_call(v->as(), code, target_type); case ast_parenthesized_expression: - return pre_compile_expr(v->as()->get_expr(), code, lval_ctx); + return pre_compile_expr(v->as()->get_expr(), code, target_type, lval_ctx); case ast_tensor: - return process_tensor(v->as(), code, lval_ctx); + return process_tensor(v->as(), code, target_type, lval_ctx); case ast_typed_tuple: - return process_typed_tuple(v->as(), code, lval_ctx); + return process_typed_tuple(v->as(), code, target_type, lval_ctx); case ast_int_const: - return process_int_const(v->as(), code); + return process_int_const(v->as(), code, target_type); case ast_string_const: - return process_string_const(v->as(), code); + return process_string_const(v->as(), code, target_type); case ast_bool_const: - return process_bool_const(v->as(), code); + return process_bool_const(v->as(), code, target_type); case ast_null_keyword: - return process_null_keyword(v->as(), code); + return process_null_keyword(v->as(), code, target_type); case ast_local_var_lhs: - return process_local_var(v->as(), code); + return process_local_var(v->as(), code, target_type); case ast_local_vars_declaration: return process_local_vars_declaration(v->as(), code); case ast_underscore: @@ -814,7 +1063,7 @@ static void process_try_catch_statement(V v, CodeBlob& } static void process_repeat_statement(V v, CodeBlob& code) { - std::vector tmp_vars = pre_compile_expr(v->get_cond(), code); + std::vector tmp_vars = pre_compile_expr(v->get_cond(), code, nullptr); Op& repeat_op = code.emplace_back(v->loc, Op::_Repeat, tmp_vars); code.push_set_cur(repeat_op.block0); process_any_statement(v->get_body(), code); @@ -822,7 +1071,7 @@ static void process_repeat_statement(V v, CodeBlob& code) } static void process_if_statement(V v, CodeBlob& code) { - std::vector tmp_vars = pre_compile_expr(v->get_cond(), code); + std::vector tmp_vars = pre_compile_expr(v->get_cond(), code, nullptr); Op& if_op = code.emplace_back(v->loc, Op::_If, std::move(tmp_vars)); code.push_set_cur(if_op.block0); process_any_statement(v->get_if_body(), code); @@ -872,14 +1121,16 @@ static void process_do_while_statement(V v, CodeBlob& co v_un->mutate()->assign_fun_ref(lookup_global_symbol(static_cast(v_un->operator_name) + "_")->as()); } - until_op.left = pre_compile_expr(until_cond, code); + until_op.left = pre_compile_expr(until_cond, code, nullptr); + tolk_assert(until_op.left.size() == 1); code.close_pop_cur(v->get_body()->loc_end); } static void process_while_statement(V v, CodeBlob& code) { Op& while_op = code.emplace_back(v->loc, Op::_While); code.push_set_cur(while_op.block0); - while_op.left = pre_compile_expr(v->get_cond(), code); + while_op.left = pre_compile_expr(v->get_cond(), code, nullptr); + tolk_assert(while_op.left.size() == 1); code.close_pop_cur(v->get_body()->loc); code.push_set_cur(while_op.block1); process_any_statement(v->get_body(), code); @@ -899,7 +1150,14 @@ static void process_throw_statement(V v, CodeBlob& code) { } static void process_return_statement(V v, CodeBlob& code) { - std::vector return_vars = v->has_return_value() ? pre_compile_expr(v->get_return_value(), code) : std::vector{}; + std::vector return_vars; + if (v->has_return_value()) { + TypePtr child_target_type = code.fun_ref->inferred_return_type; + if (code.fun_ref->does_return_self()) { + child_target_type = code.fun_ref->parameters[0].declared_type; + } + return_vars = pre_compile_expr(v->get_return_value(), code, child_target_type); + } if (code.fun_ref->does_return_self()) { return_vars = {}; } @@ -951,7 +1209,7 @@ void process_any_statement(AnyV v, CodeBlob& code) { case ast_empty_statement: return; default: - pre_compile_expr(reinterpret_cast(v), code); + pre_compile_expr(reinterpret_cast(v), code, nullptr); } } @@ -962,7 +1220,7 @@ static void convert_function_body_to_CodeBlob(const FunctionData* fun_ref, Funct std::vector rvect_import; int total_arg_width = 0; for (int i = 0; i < fun_ref->get_num_params(); ++i) { - total_arg_width += fun_ref->parameters[i].declared_type->calc_width_on_stack(); + total_arg_width += fun_ref->parameters[i].declared_type->get_width_on_stack(); } rvect_import.reserve(total_arg_width); @@ -990,7 +1248,7 @@ static void convert_function_body_to_CodeBlob(const FunctionData* fun_ref, Funct static void convert_asm_body_to_AsmOp(const FunctionData* fun_ref, FunctionBodyAsm* asm_body) { int cnt = fun_ref->get_num_params(); - int width = fun_ref->inferred_return_type->calc_width_on_stack(); + int width = fun_ref->inferred_return_type->get_width_on_stack(); std::vector asm_ops; for (AnyV v_child : fun_ref->ast_root->as()->get_body()->as()->get_asm_commands()) { std::string_view ops = v_child->as()->str_val; // \n\n... @@ -1029,7 +1287,7 @@ public: int total_arg_mutate_width = 0; bool has_arg_width_not_1 = false; for (const LocalVarData& param : fun_ref->parameters) { - int arg_width = param.declared_type->calc_width_on_stack(); + int arg_width = param.declared_type->get_width_on_stack(); has_arg_width_not_1 |= arg_width != 1; total_arg_mutate_width += param.is_mutate_parameter() * arg_width; } @@ -1043,7 +1301,7 @@ public: cum_arg_width.reserve(1 + fun_ref->get_num_params()); cum_arg_width.push_back(0); for (const LocalVarData& param : fun_ref->parameters) { - cum_arg_width.push_back(total_arg_width += param.declared_type->calc_width_on_stack()); + cum_arg_width.push_back(total_arg_width += param.declared_type->get_width_on_stack()); } std::vector arg_order; for (int i = 0; i < fun_ref->get_num_params(); ++i) { @@ -1060,7 +1318,7 @@ public: // ret_order is a shuffled range 0...N // validate N: a function should return value and mutated arguments onto a stack if (!fun_ref->ret_order.empty()) { - size_t expected_width = fun_ref->inferred_return_type->calc_width_on_stack() + total_arg_mutate_width; + size_t expected_width = fun_ref->inferred_return_type->get_width_on_stack() + total_arg_mutate_width; if (expected_width != fun_ref->ret_order.size()) { v_function->get_body()->error("ret_order (after ->) expected to contain " + std::to_string(expected_width) + " numbers"); } diff --git a/tolk/pipe-calc-rvalue-lvalue.cpp b/tolk/pipe-calc-rvalue-lvalue.cpp index 041aec89..c0fa323b 100644 --- a/tolk/pipe-calc-rvalue-lvalue.cpp +++ b/tolk/pipe-calc-rvalue-lvalue.cpp @@ -177,6 +177,18 @@ class CalculateRvalueLvalueVisitor final : public ASTVisitorFunctionBody { parent::visit(v->get_expr()); // leave lvalue state unchanged, for `mutate (t.0 as int)` both `t.0 as int` and `t.0` are lvalue } + void visit(V v) override { + mark_vertex_cur_or_rvalue(v); + parent::visit(v->get_expr()); // leave lvalue state unchanged, for `mutate x!` both `x!` and `x` are lvalue + } + + void visit(V v) override { + mark_vertex_cur_or_rvalue(v); + MarkingState saved = enter_state(MarkingState::RValue); + parent::visit(v->get_expr()); + restore_state(saved); + } + void visit(V v) override { tolk_assert(cur_state == MarkingState::LValue); mark_vertex_cur_or_rvalue(v); diff --git a/tolk/pipe-check-rvalue-lvalue.cpp b/tolk/pipe-check-rvalue-lvalue.cpp index a824cc5d..7b208816 100644 --- a/tolk/pipe-check-rvalue-lvalue.cpp +++ b/tolk/pipe-check-rvalue-lvalue.cpp @@ -97,6 +97,18 @@ class CheckRValueLvalueVisitor final : public ASTVisitorFunctionBody { parent::visit(v->get_expr()); } + void visit(V v) override { + // if `x!` is lvalue, then `x` is also lvalue, so check that `x` is ok + parent::visit(v->get_expr()); + } + + void visit(V v) override { + if (v->is_lvalue) { + fire_error_cannot_be_used_as_lvalue(v, v->is_negated ? "operator !=" : "operator =="); + } + parent::visit(v->get_expr()); + } + void visit(V v) override { if (v->is_lvalue) { fire_error_cannot_be_used_as_lvalue(v, "literal"); diff --git a/tolk/pipe-constant-folding.cpp b/tolk/pipe-constant-folding.cpp index 05d543b3..7ab9af7d 100644 --- a/tolk/pipe-constant-folding.cpp +++ b/tolk/pipe-constant-folding.cpp @@ -88,6 +88,17 @@ class ConstantFoldingReplacer final : public ASTReplacerInFunctionBody { return v; } + AnyExprV replace(V v) override { + parent::replace(v); + + // `null == null` / `null != null` + if (v->get_expr()->type == ast_null_keyword) { + return create_bool_const(v->loc, !v->is_negated); + } + + return v; + } + public: bool should_visit_function(const FunctionData* fun_ref) override { return fun_ref->is_code_function() && !fun_ref->is_generic_function(); diff --git a/tolk/pipe-infer-types-and-calls.cpp b/tolk/pipe-infer-types-and-calls.cpp index abb060a2..b247c831 100644 --- a/tolk/pipe-infer-types-and-calls.cpp +++ b/tolk/pipe-infer-types-and-calls.cpp @@ -82,11 +82,6 @@ static std::string to_string(AnyExprV v_with_type) { return "`" + v_with_type->inferred_type->as_human_readable() + "`"; } -GNU_ATTRIBUTE_NOINLINE -static std::string to_string(const LocalVarData& var_ref) { - return "`" + var_ref.declared_type->as_human_readable() + "`"; -} - GNU_ATTRIBUTE_NOINLINE static std::string to_string(const FunctionData* fun_ref) { return "`" + fun_ref->as_human_readable() + "`"; @@ -97,7 +92,7 @@ static std::string to_string(const FunctionData* fun_ref) { // (in FunC, `forall` type just couldn't be unified with non-primitives; in Tolk, generic T is expectedly inferred) GNU_ATTRIBUTE_NORETURN GNU_ATTRIBUTE_COLD static void fire_error_calling_asm_function_with_non1_stack_width_arg(SrcLocation loc, const FunctionData* fun_ref, const std::vector& substitutions, int arg_idx) { - throw ParseError(loc, "can not call `" + fun_ref->as_human_readable() + "` with " + fun_ref->genericTs->get_nameT(arg_idx) + "=" + substitutions[arg_idx]->as_human_readable() + ", because it occupies " + std::to_string(substitutions[arg_idx]->calc_width_on_stack()) + " stack slots in TVM, not 1"); + throw ParseError(loc, "can not call `" + fun_ref->as_human_readable() + "` with " + fun_ref->genericTs->get_nameT(arg_idx) + "=" + substitutions[arg_idx]->as_human_readable() + ", because it occupies " + std::to_string(substitutions[arg_idx]->get_width_on_stack()) + " stack slots in TVM, not 1"); } // fire an error on `var n = null` @@ -134,34 +129,26 @@ static void fire_error_cannot_deduce_untyped_tuple_access(SrcLocation loc, int i // fire an error on `untypedTupleVar.0` when inferred as (int,int), or `[int, (int,int)]`, or other non-1 width in a tuple GNU_ATTRIBUTE_NORETURN GNU_ATTRIBUTE_COLD static void fire_error_tuple_cannot_have_non1_stack_width_elem(SrcLocation loc, TypePtr inferred_type) { - throw ParseError(loc, "a tuple can not have " + to_string(inferred_type) + " inside, because it occupies " + std::to_string(inferred_type->calc_width_on_stack()) + " stack slots in TVM, not 1"); + throw ParseError(loc, "a tuple can not have " + to_string(inferred_type) + " inside, because it occupies " + std::to_string(inferred_type->get_width_on_stack()) + " stack slots in TVM, not 1"); } -// check correctness of called arguments counts and their type matching -static void check_function_arguments(const FunctionData* fun_ref, V v, AnyExprV lhs_of_dot_call) { - int delta_self = lhs_of_dot_call ? 1 : 0; - int n_arguments = v->size() + delta_self; - int n_parameters = fun_ref->get_num_params(); - - // Tolk doesn't have optional parameters currently, so just compare counts - if (!n_parameters && lhs_of_dot_call) { - v->error("`" + fun_ref->name + "` has no parameters and can not be called as method"); - } - if (n_parameters < n_arguments) { - v->error("too many arguments in call to `" + fun_ref->name + "`, expected " + std::to_string(n_parameters - delta_self) + ", have " + std::to_string(n_arguments - delta_self)); - } - if (n_arguments < n_parameters) { - v->error("too few arguments in call to `" + fun_ref->name + "`, expected " + std::to_string(n_parameters - delta_self) + ", have " + std::to_string(n_arguments - delta_self)); - } - - if (lhs_of_dot_call) { - if (!fun_ref->parameters[0].declared_type->can_rhs_be_assigned(lhs_of_dot_call->inferred_type)) { - lhs_of_dot_call->error("can not call method for " + to_string(fun_ref->parameters[0]) + " with object of type " + to_string(lhs_of_dot_call)); +// check type correctness of a passed argument when calling a function/method +static void check_function_argument(TypePtr param_type, bool is_mutate_param, AnyExprV ith_arg, bool is_obj_of_dot_call) { + // given `f(x: int)` and a call `f(expr)`, check that expr_type is assignable to `int` + if (!param_type->can_rhs_be_assigned(ith_arg->inferred_type)) { + if (is_obj_of_dot_call) { + ith_arg->error("can not call method for " + to_string(param_type) + " with object of type " + to_string(ith_arg)); + } else { + ith_arg->error("can not pass " + to_string(ith_arg) + " to " + to_string(param_type)); } } - for (int i = 0; i < v->size(); ++i) { - if (!fun_ref->parameters[i + delta_self].declared_type->can_rhs_be_assigned(v->get_arg(i)->inferred_type)) { - v->get_arg(i)->error("can not pass " + to_string(v->get_arg(i)) + " to " + to_string(fun_ref->parameters[i + delta_self])); + // given `f(x: mutate int?)` and a call `f(expr)`, check that `int?` is assignable to expr_type + // (for instance, can't call such a function with `f(mutate intVal)`, since f can potentially assign null to it) + if (is_mutate_param && !ith_arg->inferred_type->can_rhs_be_assigned(param_type)) { + if (is_obj_of_dot_call) { + ith_arg->error("can not call method for mutate " + to_string(param_type) + " with object of type " + to_string(ith_arg) + ", because mutation is not type compatible"); + } else { + ith_arg->error("can not pass " + to_string(ith_arg) + " to mutate " + to_string(param_type) + ", because mutation is not type compatible"); } } } @@ -189,6 +176,13 @@ class TypeInferringUnifyStrategy { return t2; } + if (t1 == TypeDataNullLiteral::create()) { + return TypeDataNullable::create(t2); + } + if (t2 == TypeDataNullLiteral::create()) { + return TypeDataNullable::create(t1); + } + const auto* tensor1 = t1->try_as(); const auto* tensor2 = t2->try_as(); if (tensor1 && tensor2 && tensor1->size() == tensor2->size()) { @@ -365,6 +359,10 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { return infer_ternary_operator(v->as(), hint); case ast_cast_as_operator: return infer_cast_as_operator(v->as()); + case ast_not_null_operator: + return infer_not_null_operator(v->as()); + case ast_is_null_check: + return infer_is_null_check(v->as()); case ast_parenthesized_expression: return infer_parenthesized(v->as(), hint); case ast_reference: @@ -388,14 +386,29 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { } } + static TypePtr unwrap_nullable(TypePtr type) { + while (const TypeDataNullable* as_nullable = type->try_as()) { + type = as_nullable->inner; + } + return type; + } + static bool expect_integer(AnyExprV v_inferred) { return v_inferred->inferred_type == TypeDataInt::create(); } + static bool expect_integer(TypePtr inferred_type) { + return inferred_type == TypeDataInt::create(); + } + static bool expect_boolean(AnyExprV v_inferred) { return v_inferred->inferred_type == TypeDataBool::create(); } + static bool expect_boolean(TypePtr inferred_type) { + return inferred_type == TypeDataBool::create(); + } + static void infer_int_const(V v) { assign_inferred_type(v, TypeDataInt::create()); } @@ -520,7 +533,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { // check `untypedTuple.0 = rhs_tensor` and other non-1 width elements if (auto lhs_dot = lhs->try_as()) { if (lhs_dot->is_target_indexed_access() && lhs_dot->get_obj()->inferred_type == TypeDataTuple::create()) { - if (rhs_type->calc_width_on_stack() != 1) { + if (rhs_type->get_width_on_stack() != 1) { fire_error_tuple_cannot_have_non1_stack_width_elem(err_loc->loc, rhs_type); } } @@ -617,8 +630,8 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { // == != can compare both integers and booleans, (int == bool) is NOT allowed case tok_eq: case tok_neq: { - bool both_int = expect_integer(lhs) && expect_integer(rhs); - bool both_bool = expect_boolean(lhs) && expect_boolean(rhs); + bool both_int = expect_integer(unwrap_nullable(lhs->inferred_type)) && expect_integer(unwrap_nullable(rhs->inferred_type)); + bool both_bool = expect_boolean(unwrap_nullable(lhs->inferred_type)) && expect_boolean(unwrap_nullable(rhs->inferred_type)); if (!both_int && !both_bool) { if (lhs->inferred_type == rhs->inferred_type) { // compare slice with slice v->error("type " + to_string(lhs) + " can not be compared with `== !=`"); @@ -706,6 +719,22 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { assign_inferred_type(v, v->cast_to_type); } + void infer_is_null_check(V v) { + infer_any_expr(v->get_expr()); + assign_inferred_type(v, TypeDataBool::create()); + } + + void infer_not_null_operator(V v) { + infer_any_expr(v->get_expr()); + if (const auto* as_nullable = v->get_expr()->inferred_type->try_as()) { + // operator `!` used for `T?`, leave `T` + assign_inferred_type(v, as_nullable->inner); + } else { + // operator `!` used for non-nullable, probably a warning should be printed + assign_inferred_type(v, v->get_expr()); + } + } + void infer_parenthesized(V v, TypePtr hint) { infer_any_expr(v->get_expr(), hint); assign_inferred_type(v, v->get_expr()); @@ -782,7 +811,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { // T for asm function must be a TVM primitive (width 1), otherwise, asm would act incorrectly if (fun_ref->is_asm_function() || fun_ref->is_builtin_function()) { for (int i = 0; i < static_cast(substitutionTs.size()); ++i) { - if (substitutionTs[i]->calc_width_on_stack() != 1) { + if (substitutionTs[i]->get_width_on_stack() != 1) { fire_error_calling_asm_function_with_non1_stack_width_arg(loc, fun_ref, substitutionTs, i); } } @@ -836,7 +865,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { if (hint == nullptr) { fire_error_cannot_deduce_untyped_tuple_access(v->loc, index_at); } - if (hint->calc_width_on_stack() != 1) { + if (hint->get_width_on_stack() != 1) { fire_error_tuple_cannot_have_non1_stack_width_elem(v->loc, hint); } item_type = hint; @@ -857,7 +886,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { // `t.tupleSize` is ok, `cs.tupleSize` not if (!fun_ref->parameters[0].declared_type->can_rhs_be_assigned(obj_type)) { - v_ident->error("referencing a method for " + to_string(fun_ref->parameters[0]) + " with object of type " + to_string(obj_type)); + v_ident->error("referencing a method for " + to_string(fun_ref->parameters[0].declared_type) + " with object of type " + to_string(obj_type)); } if (fun_ref->is_generic_function() && !v_instantiationTs) { @@ -921,30 +950,26 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { // fun_ref remains nullptr } - // infer argument types, looking at fun_ref's parameters as hints - for (int i = 0; i < v->get_num_args(); ++i) { - TypePtr param_type = fun_ref && i < fun_ref->get_num_params() - delta_self ? fun_ref->parameters[delta_self + i].declared_type : nullptr; - auto arg_i = v->get_arg(i); - infer_any_expr(arg_i->get_expr(), param_type && !param_type->has_genericT_inside() ? param_type : nullptr); - assign_inferred_type(arg_i, arg_i->get_expr()); - } - // handle `local_var()` / `getF()()` / `5()` / `SOME_CONST()` / `obj.method()()()` / `tensorVar.0()` if (!fun_ref) { // treat callee like a usual expression, which must have "callable" inferred type infer_any_expr(callee); const TypeDataFunCallable* f_callable = callee->inferred_type->try_as(); if (!f_callable) { // `5()` / `SOME_CONST()` / `null()` - v->error("calling a non-function"); + v->error("calling a non-function " + to_string(callee->inferred_type)); } // check arguments count and their types if (v->get_num_args() != static_cast(f_callable->params_types.size())) { v->error("expected " + std::to_string(f_callable->params_types.size()) + " arguments, got " + std::to_string(v->get_arg_list()->size())); } for (int i = 0; i < v->get_num_args(); ++i) { - if (!f_callable->params_types[i]->can_rhs_be_assigned(v->get_arg(i)->inferred_type)) { - v->get_arg(i)->error("can not pass " + to_string(v->get_arg(i)) + " to " + to_string(f_callable->params_types[i])); + auto arg_i = v->get_arg(i)->get_expr(); + TypePtr param_type = f_callable->params_types[i]; + infer_any_expr(arg_i, param_type); + if (!param_type->can_rhs_be_assigned(arg_i->inferred_type)) { + arg_i->error("can not pass " + to_string(arg_i) + " to " + to_string(param_type)); } + assign_inferred_type(v->get_arg(i), arg_i); } v->mutate()->assign_fun_ref(nullptr); // no fun_ref to a global function assign_inferred_type(v, f_callable->return_type); @@ -952,30 +977,75 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { } // so, we have a call `f(args)` or `obj.f(args)`, f is a global function (fun_ref) (code / asm / builtin) + // we're going to iterate over passed arguments, check type compatibility, and (if generic) infer substitutionTs + // at first, check arguments count (Tolk doesn't have optional parameters, so just compare counts) + int n_arguments = v->get_num_args() + delta_self; + int n_parameters = fun_ref->get_num_params(); + if (!n_parameters && dot_obj) { + v->error("`" + fun_ref->name + "` has no parameters and can not be called as method"); + } + if (n_parameters < n_arguments) { + v->error("too many arguments in call to `" + fun_ref->name + "`, expected " + std::to_string(n_parameters - delta_self) + ", have " + std::to_string(n_arguments - delta_self)); + } + if (n_arguments < n_parameters) { + v->error("too few arguments in call to `" + fun_ref->name + "`, expected " + std::to_string(n_parameters - delta_self) + ", have " + std::to_string(n_arguments - delta_self)); + } + + // now, for every passed argument, we need to infer its type, and check it against parameter type + // for regular functions, it's obvious + // but for generic functions, we need to infer type arguments (substitutionTs) on the fly + // (unless Ts are specified by a user like `f(args)` / `t.tupleAt()`, take them) + GenericSubstitutionsDeduceForCall* deducingTs = fun_ref->is_generic_function() ? new GenericSubstitutionsDeduceForCall(fun_ref) : nullptr; + if (deducingTs && v_instantiationTs) { + deducingTs->provide_manually_specified(collect_fun_generic_substitutions_from_manually_specified(v->loc, fun_ref, v_instantiationTs)); + } + + // loop over every argument, for `obj.method()` obj is the first one + // if genericT deducing has a conflict, ParseError is thrown + // note, that deducing Ts one by one is important to manage control flow (mutate params work like assignments) + // a corner case, e.g. `f(v1:T?, v2:T?)` and `f(null,2)` will fail on first argument, won't try the second one + if (dot_obj) { + const LocalVarData& param_0 = fun_ref->parameters[0]; + TypePtr param_type = param_0.declared_type; + if (param_type->has_genericT_inside()) { + param_type = deducingTs->auto_deduce_from_argument(dot_obj->loc, param_type, dot_obj->inferred_type); + } + check_function_argument(param_type, param_0.is_mutate_parameter(), dot_obj, true); + } + for (int i = 0; i < v->get_num_args(); ++i) { + const LocalVarData& param_i = fun_ref->parameters[delta_self + i]; + AnyExprV arg_i = v->get_arg(i)->get_expr(); + TypePtr param_type = param_i.declared_type; + if (param_type->has_genericT_inside() && deducingTs->is_manually_specified()) { // `f(a)` + param_type = deducingTs->replace_by_manually_specified(param_type); + } + if (param_type->has_genericT_inside()) { // `f(a)` where f is generic: use `a` to infer param type + infer_any_expr(arg_i); // then arg_i is inferred without any hint + param_type = deducingTs->auto_deduce_from_argument(arg_i->loc, param_type, arg_i->inferred_type); + } else { + infer_any_expr(arg_i, param_type); // param_type is hint, helps infer arg_i + } + assign_inferred_type(v->get_arg(i), arg_i); // arg itself is an expression + check_function_argument(param_type, param_i.is_mutate_parameter(), arg_i, false); + } + // if it's a generic function `f`, we need to instantiate it, like `f` // same for generic methods `t.tupleAt`, need to achieve `t.tupleAt` - if (fun_ref->is_generic_function() && v_instantiationTs) { - // if Ts are specified by a user like `f(args)` / `t.tupleAt()`, take them - std::vector substitutions = collect_fun_generic_substitutions_from_manually_specified(v->loc, fun_ref, v_instantiationTs); - fun_ref = check_and_instantiate_generic_function(v->loc, fun_ref, std::move(substitutions)); - - } else if (fun_ref->is_generic_function()) { - // if `f` called like `f(args)`, deduce T from arg types - std::vector arg_types; - arg_types.reserve(delta_self + v->get_num_args()); - if (dot_obj) { - arg_types.push_back(dot_obj->inferred_type); + if (fun_ref->is_generic_function()) { + // if `f(args)` was called, Ts were inferred; check that all of them are known + int idx = deducingTs->get_first_not_deduced_idx(); + if (idx != -1 && hint && fun_ref->declared_return_type->has_genericT_inside()) { + // example: `t.tupleFirst()`, T doesn't depend on arguments, but is determined by return type + // if used like `var x: int = t.tupleFirst()` / `t.tupleFirst() as int` / etc., use hint + deducingTs->auto_deduce_from_argument(v->loc, fun_ref->declared_return_type, hint); + idx = deducingTs->get_first_not_deduced_idx(); } - for (int i = 0; i < v->get_num_args(); ++i) { - arg_types.push_back(v->get_arg(i)->inferred_type); + if (idx != -1) { + v->error("can not deduce " + fun_ref->genericTs->get_nameT(idx)); } - - td::Result> deduced = deduce_substitutionTs_on_generic_func_call(fun_ref, std::move(arg_types), hint); - if (deduced.is_error()) { - v->error(deduced.error().message().str() + " for generic function " + to_string(fun_ref)); - } - fun_ref = check_and_instantiate_generic_function(v->loc, fun_ref, deduced.move_as_ok()); + fun_ref = check_and_instantiate_generic_function(v->loc, fun_ref, deducingTs->flush()); + delete deducingTs; } else if (UNLIKELY(v_instantiationTs != nullptr)) { // non-generic function/method called with type arguments, like `c.cellHash()` / `beginCell()` @@ -988,8 +1058,6 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { v->get_callee()->as()->mutate()->assign_target(fun_ref); v->get_callee()->as()->mutate()->assign_inferred_type(fun_ref->inferred_full_type); } - // check arguments count and their types - check_function_arguments(fun_ref, v->get_arg_list(), dot_obj); // get return type either from user-specified declaration or infer here on demand traversing its body get_or_infer_return_type(fun_ref); TypePtr inferred_type = dot_obj && fun_ref->does_return_self() ? dot_obj->inferred_type : fun_ref->inferred_return_type; @@ -1020,7 +1088,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { for (int i = 0; i < v->size(); ++i) { AnyExprV item = v->get_item(i); infer_any_expr(item, tuple_hint && i < tuple_hint->size() ? tuple_hint->items[i] : nullptr); - if (item->inferred_type->calc_width_on_stack() != 1) { + if (item->inferred_type->get_width_on_stack() != 1) { fire_error_tuple_cannot_have_non1_stack_width_elem(v->get_item(i)->loc, item->inferred_type); } types_list.emplace_back(item->inferred_type); @@ -1134,7 +1202,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { v->get_thrown_code()->error("excNo of `throw` must be an integer, got " + to_string(v->get_thrown_code())); } infer_any_expr(v->get_thrown_arg()); - if (v->has_thrown_arg() && v->get_thrown_arg()->inferred_type->calc_width_on_stack() != 1) { + if (v->has_thrown_arg() && v->get_thrown_arg()->inferred_type->get_width_on_stack() != 1) { v->get_thrown_arg()->error("can not throw " + to_string(v->get_thrown_arg()) + ", exception arg must occupy exactly 1 stack slot"); } } @@ -1163,7 +1231,7 @@ class InferCheckTypesAndCallsAndFieldsVisitor final { // `catch` has exactly 2 variables: excNo and arg (when missing, they are implicit underscores) // `arg` is a curious thing, it can be any TVM primitive, so assign unknown to it - // hence, using `fInt(arg)` (int from parameter is a hint) or `arg as slice` works well + // hence, using `fInt(arg)` (int from parameter is a target type) or `arg as slice` works well // it's not truly correct, because `arg as (int,int)` also compiles, but can never happen, but let it be user responsibility tolk_assert(v->get_catch_expr()->size() == 2); std::vector types_list = {TypeDataInt::create(), TypeDataUnknown::create()}; diff --git a/tolk/pipe-optimize-boolean-expr.cpp b/tolk/pipe-optimize-boolean-expr.cpp index c0251295..946e79cb 100644 --- a/tolk/pipe-optimize-boolean-expr.cpp +++ b/tolk/pipe-optimize-boolean-expr.cpp @@ -120,6 +120,11 @@ protected: } v = createV(v->loc, !v->is_ifnot, v_cond_unary->get_rhs(), v->get_if_body(), v->get_else_body()); } + // `if (x != null)` -> ifnot(x == null) + if (auto v_cond_isnull = v->get_cond()->try_as(); v_cond_isnull && v_cond_isnull->is_negated) { + v_cond_isnull->mutate()->assign_is_negated(!v_cond_isnull->is_negated); + v = createV(v->loc, !v->is_ifnot, v_cond_isnull, v->get_if_body(), v->get_else_body()); + } return v; } diff --git a/tolk/pipe-refine-lvalue-for-mutate.cpp b/tolk/pipe-refine-lvalue-for-mutate.cpp index 540d7413..48bcd2dc 100644 --- a/tolk/pipe-refine-lvalue-for-mutate.cpp +++ b/tolk/pipe-refine-lvalue-for-mutate.cpp @@ -86,6 +86,8 @@ class RefineLvalueForMutateArgumentsVisitor final : public ASTVisitorFunctionBod leftmost_obj = as_par->get_expr(); } else if (auto as_cast = leftmost_obj->try_as()) { leftmost_obj = as_cast->get_expr(); + } else if (auto as_nn = leftmost_obj->try_as()) { + leftmost_obj = as_nn->get_expr(); } else { break; } diff --git a/tolk/tolk.h b/tolk/tolk.h index 4086d7f7..89fe6d5f 100644 --- a/tolk/tolk.h +++ b/tolk/tolk.h @@ -45,7 +45,7 @@ typedef int const_idx_t; struct TmpVar { var_idx_t ir_idx; // every var in IR represents 1 stack slot - TypePtr v_type; // calc_width_on_stack() is 1 + TypePtr v_type; // get_width_on_stack() is 1 std::string name; // "x" for vars originated from user sources; "x.0" for tensor components; empty for implicitly created tmp vars SrcLocation loc; // location of var declaration in sources or where a tmp var was originated #ifdef TOLK_DEBUG diff --git a/tolk/type-system.cpp b/tolk/type-system.cpp index c7122e10..73546be8 100644 --- a/tolk/type-system.cpp +++ b/tolk/type-system.cpp @@ -108,6 +108,20 @@ void type_system_init() { // and creates an object only if it isn't found in a global hashtable // +TypePtr TypeDataNullable::create(TypePtr inner) { + TypeDataTypeIdCalculation hash(1774084920039440885ULL); + hash.feed_child(inner); + + if (TypePtr existing = hash.get_existing()) { + return existing; + } + // most types (int?, slice?, etc.), when nullable, still occupy 1 stack slot (holding TVM NULL at runtime) + // but for example for `(int, int)` we need an extra stack slot "null flag" + int inner_w = inner->get_width_on_stack(); + int width_on_stack = inner_w == 1 ? 1 : inner_w + 1; + return hash.register_unique(new TypeDataNullable(hash.type_id(), hash.children_flags(), width_on_stack, inner)); +} + TypePtr TypeDataFunCallable::create(std::vector&& params_types, TypePtr return_type) { TypeDataTypeIdCalculation hash(3184039965511020991ULL); for (TypePtr param : params_types) { @@ -143,7 +157,11 @@ TypePtr TypeDataTensor::create(std::vector&& items) { if (TypePtr existing = hash.get_existing()) { return existing; } - return hash.register_unique(new TypeDataTensor(hash.type_id(), hash.children_flags(), std::move(items))); + int width_on_stack = 0; + for (TypePtr item : items) { + width_on_stack += item->get_width_on_stack(); + } + return hash.register_unique(new TypeDataTensor(hash.type_id(), hash.children_flags(), width_on_stack, std::move(items))); } TypePtr TypeDataTypedTuple::create(std::vector&& items) { @@ -178,6 +196,12 @@ TypePtr TypeDataUnresolved::create(std::string&& text, SrcLocation loc) { // only non-trivial implementations are here; trivial are defined in .h file // +std::string TypeDataNullable::as_human_readable() const { + std::string nested = inner->as_human_readable(); + bool embrace = inner->try_as(); + return embrace ? "(" + nested + ")?" : nested + "?"; +} + std::string TypeDataFunCallable::as_human_readable() const { std::string result = "("; for (TypePtr param : params_types) { @@ -223,6 +247,11 @@ std::string TypeDataTypedTuple::as_human_readable() const { // only non-trivial implementations are here; by default (no children), `callback(this)` is executed // +void TypeDataNullable::traverse(const TraverserCallbackT& callback) const { + callback(this); + inner->traverse(callback); +} + void TypeDataFunCallable::traverse(const TraverserCallbackT& callback) const { callback(this); for (TypePtr param : params_types) { @@ -254,6 +283,10 @@ void TypeDataTypedTuple::traverse(const TraverserCallbackT& callback) const { // only non-trivial implementations are here; by default (no children), `return callback(this)` is executed // +TypePtr TypeDataNullable::replace_children_custom(const ReplacerCallbackT& callback) const { + return callback(create(inner->replace_children_custom(callback))); +} + TypePtr TypeDataFunCallable::replace_children_custom(const ReplacerCallbackT& callback) const { std::vector mapped; mapped.reserve(params_types.size()); @@ -282,53 +315,17 @@ TypePtr TypeDataTypedTuple::replace_children_custom(const ReplacerCallbackT& cal } -// -------------------------------------------- -// calc_width_on_stack() -// -// returns the number of stack slots occupied by a variable of this type -// only non-trivial implementations are here; by default (most types) occupy 1 stack slot -// - -int TypeDataGenericT::calc_width_on_stack() const { - // this function is invoked only in functions with generics already instantiated - assert(false); - return -999999; -} - -int TypeDataTensor::calc_width_on_stack() const { - int sum = 0; - for (TypePtr item : items) { - sum += item->calc_width_on_stack(); - } - return sum; -} - -int TypeDataUnresolved::calc_width_on_stack() const { - // since early pipeline stages, no unresolved types left - assert(false); - return -999999; -} - -int TypeDataVoid::calc_width_on_stack() const { - return 0; -} - - // -------------------------------------------- // can_rhs_be_assigned() // // on `var lhs: = rhs`, having inferred rhs_type, check that it can be assigned without any casts // the same goes for passing arguments, returning values, etc. — where the "receiver" (lhs) checks "applier" (rhs) -// for now, `null` can be assigned to any TVM primitive, be later we'll have T? types and null safety // bool TypeDataInt::can_rhs_be_assigned(TypePtr rhs) const { if (rhs == this) { return true; } - if (rhs == TypeDataNullLiteral::create()) { - return true; - } return false; } @@ -336,9 +333,6 @@ bool TypeDataBool::can_rhs_be_assigned(TypePtr rhs) const { if (rhs == this) { return true; } - if (rhs == TypeDataNullLiteral::create()) { - return true; - } return false; } @@ -346,9 +340,6 @@ bool TypeDataCell::can_rhs_be_assigned(TypePtr rhs) const { if (rhs == this) { return true; } - if (rhs == TypeDataNullLiteral::create()) { - return true; - } return false; } @@ -356,9 +347,6 @@ bool TypeDataSlice::can_rhs_be_assigned(TypePtr rhs) const { if (rhs == this) { return true; } - if (rhs == TypeDataNullLiteral::create()) { - return true; - } return false; } @@ -366,9 +354,6 @@ bool TypeDataBuilder::can_rhs_be_assigned(TypePtr rhs) const { if (rhs == this) { return true; } - if (rhs == TypeDataNullLiteral::create()) { - return true; - } return false; } @@ -376,9 +361,6 @@ bool TypeDataTuple::can_rhs_be_assigned(TypePtr rhs) const { if (rhs == this) { return true; } - if (rhs == TypeDataNullLiteral::create()) { - return true; - } return false; } @@ -386,9 +368,6 @@ bool TypeDataContinuation::can_rhs_be_assigned(TypePtr rhs) const { if (rhs == this) { return true; } - if (rhs == TypeDataNullLiteral::create()) { - return true; - } return false; } @@ -396,6 +375,19 @@ bool TypeDataNullLiteral::can_rhs_be_assigned(TypePtr rhs) const { return rhs == this; } +bool TypeDataNullable::can_rhs_be_assigned(TypePtr rhs) const { + if (rhs == this) { + return true; + } + if (rhs == TypeDataNullLiteral::create()) { + return true; + } + if (const TypeDataNullable* rhs_nullable = rhs->try_as()) { + return inner->can_rhs_be_assigned(rhs_nullable->inner); + } + return inner->can_rhs_be_assigned(rhs); +} + bool TypeDataFunCallable::can_rhs_be_assigned(TypePtr rhs) const { return rhs == this; } @@ -414,7 +406,6 @@ bool TypeDataTensor::can_rhs_be_assigned(TypePtr rhs) const { } return true; } - // note, that tensors can not accept null return false; } @@ -427,9 +418,6 @@ bool TypeDataTypedTuple::can_rhs_be_assigned(TypePtr rhs) const { } return true; } - if (rhs == TypeDataNullLiteral::create()) { - return true; - } return false; } @@ -455,41 +443,69 @@ bool TypeDataVoid::can_rhs_be_assigned(TypePtr rhs) const { // bool TypeDataInt::can_be_casted_with_as_operator(TypePtr cast_to) const { + if (const auto* to_nullable = cast_to->try_as()) { // `int` as `int?` + return can_be_casted_with_as_operator(to_nullable->inner); + } return cast_to == this; } bool TypeDataBool::can_be_casted_with_as_operator(TypePtr cast_to) const { + if (const auto* to_nullable = cast_to->try_as()) { + return can_be_casted_with_as_operator(to_nullable->inner); + } return cast_to == this || cast_to == TypeDataInt::create(); } bool TypeDataCell::can_be_casted_with_as_operator(TypePtr cast_to) const { + if (const auto* to_nullable = cast_to->try_as()) { + return can_be_casted_with_as_operator(to_nullable->inner); + } return cast_to == this; } bool TypeDataSlice::can_be_casted_with_as_operator(TypePtr cast_to) const { + if (const auto* to_nullable = cast_to->try_as()) { + return can_be_casted_with_as_operator(to_nullable->inner); + } return cast_to == this; } bool TypeDataBuilder::can_be_casted_with_as_operator(TypePtr cast_to) const { + if (const auto* to_nullable = cast_to->try_as()) { + return can_be_casted_with_as_operator(to_nullable->inner); + } return cast_to == this; } bool TypeDataTuple::can_be_casted_with_as_operator(TypePtr cast_to) const { + if (const auto* to_nullable = cast_to->try_as()) { + return can_be_casted_with_as_operator(to_nullable->inner); + } return cast_to == this; } bool TypeDataContinuation::can_be_casted_with_as_operator(TypePtr cast_to) const { + if (const auto* to_nullable = cast_to->try_as()) { + return can_be_casted_with_as_operator(to_nullable->inner); + } return cast_to == this; } bool TypeDataNullLiteral::can_be_casted_with_as_operator(TypePtr cast_to) const { - return cast_to == this - || cast_to == TypeDataInt::create() || cast_to == TypeDataBool::create() || cast_to == TypeDataCell::create() || cast_to == TypeDataSlice::create() - || cast_to == TypeDataBuilder::create() || cast_to == TypeDataContinuation::create() || cast_to == TypeDataTuple::create() - || cast_to->try_as(); + return cast_to == this || cast_to->try_as(); +} + +bool TypeDataNullable::can_be_casted_with_as_operator(TypePtr cast_to) const { + if (const auto* to_nullable = cast_to->try_as()) { + return inner->can_be_casted_with_as_operator(to_nullable->inner); + } + return false; } bool TypeDataFunCallable::can_be_casted_with_as_operator(TypePtr cast_to) const { + if (const auto* to_nullable = cast_to->try_as()) { + return can_be_casted_with_as_operator(to_nullable->inner); + } return this == cast_to; } @@ -506,6 +522,9 @@ bool TypeDataTensor::can_be_casted_with_as_operator(TypePtr cast_to) const { } return true; } + if (const auto* to_nullable = cast_to->try_as()) { + return can_be_casted_with_as_operator(to_nullable->inner); + } return false; } @@ -518,14 +537,15 @@ bool TypeDataTypedTuple::can_be_casted_with_as_operator(TypePtr cast_to) const { } return true; } + if (const auto* to_nullable = cast_to->try_as()) { + return can_be_casted_with_as_operator(to_nullable->inner); + } return false; } bool TypeDataUnknown::can_be_casted_with_as_operator(TypePtr cast_to) const { - // 'unknown' can be cast to any type - // (though it's not valid for exception arguments when casting them to non-1 stack width, - // but to ensure it, we need a special type "unknown TVM primitive", which is overwhelming I think) - return true; + // 'unknown' can be cast to any TVM value + return cast_to->get_width_on_stack() == 1; } bool TypeDataUnresolved::can_be_casted_with_as_operator(TypePtr cast_to) const { @@ -542,7 +562,7 @@ bool TypeDataVoid::can_be_casted_with_as_operator(TypePtr cast_to) const { // // here we implement parsing types (mostly after colon) to TypeData // example: `var v: int` is TypeDataInt -// example: `var v: (builder, [cell])` is TypeDataTensor(TypeDataBuilder, TypeDataTypedTuple(TypeDataCell)) +// example: `var v: (builder?, [cell])` is TypeDataTensor(TypeDataNullable(TypeDataBuilder), TypeDataTypedTuple(TypeDataCell)) // example: `fun f(): ()` is TypeDataTensor() (an empty one) // // note, that unrecognized type names (MyEnum, MyStruct, T) are parsed as TypeDataUnresolved, @@ -633,7 +653,8 @@ static TypePtr parse_type_nullable(Lexer& lex) { TypePtr result = parse_simple_type(lex); if (lex.tok() == tok_question) { - lex.error("nullable types are not supported yet"); + lex.next(); + result = TypeDataNullable::create(result); } return result; diff --git a/tolk/type-system.h b/tolk/type-system.h index 482039e6..ab793f1d 100644 --- a/tolk/type-system.h +++ b/tolk/type-system.h @@ -50,6 +50,8 @@ class TypeData { const uint64_t type_id; // bits of flag_mask, to store often-used properties and return them without tree traversing const int flags; + // how many slots on a stack this type occupies (calculated on creation), e.g. `int`=1, `(int,int)`=2, `(int,int)?`=3 + const int width_on_stack; friend class TypeDataTypeIdCalculation; @@ -60,9 +62,10 @@ protected: flag_contains_unresolved_inside = 1 << 3, }; - explicit TypeData(uint64_t type_id, int flags_with_children) + explicit TypeData(uint64_t type_id, int flags_with_children, int width_on_stack) : type_id(type_id) - , flags(flags_with_children) { + , flags(flags_with_children) + , width_on_stack(width_on_stack) { } public: @@ -74,6 +77,7 @@ public: } uint64_t get_type_id() const { return type_id; } + int get_width_on_stack() const { return width_on_stack; } bool has_unknown_inside() const { return flags & flag_contains_unknown_inside; } bool has_genericT_inside() const { return flags & flag_contains_genericT_inside; } @@ -93,17 +97,13 @@ public: virtual TypePtr replace_children_custom(const ReplacerCallbackT& callback) const { return callback(this); } - - virtual int calc_width_on_stack() const { - return 1; - } }; /* * `int` is TypeDataInt, representation of TVM int. */ class TypeDataInt final : public TypeData { - TypeDataInt() : TypeData(1ULL, 0) {} + TypeDataInt() : TypeData(1ULL, 0, 1) {} static TypePtr singleton; friend void type_system_init(); @@ -121,7 +121,7 @@ public: * From the type system point of view, int and bool are different, not-autocastable types. */ class TypeDataBool final : public TypeData { - TypeDataBool() : TypeData(2ULL, 0) {} + TypeDataBool() : TypeData(2ULL, 0, 1) {} static TypePtr singleton; friend void type_system_init(); @@ -138,7 +138,7 @@ public: * `cell` is TypeDataCell, representation of TVM cell. */ class TypeDataCell final : public TypeData { - TypeDataCell() : TypeData(3ULL, 0) {} + TypeDataCell() : TypeData(3ULL, 0, 1) {} static TypePtr singleton; friend void type_system_init(); @@ -155,7 +155,7 @@ public: * `slice` is TypeDataSlice, representation of TVM slice. */ class TypeDataSlice final : public TypeData { - TypeDataSlice() : TypeData(4ULL, 0) {} + TypeDataSlice() : TypeData(4ULL, 0, 1) {} static TypePtr singleton; friend void type_system_init(); @@ -172,7 +172,7 @@ public: * `builder` is TypeDataBuilder, representation of TVM builder. */ class TypeDataBuilder final : public TypeData { - TypeDataBuilder() : TypeData(5ULL, 0) {} + TypeDataBuilder() : TypeData(5ULL, 0, 1) {} static TypePtr singleton; friend void type_system_init(); @@ -191,7 +191,7 @@ public: * so getting its element results in TypeDataUnknown (which must be assigned/cast explicitly). */ class TypeDataTuple final : public TypeData { - TypeDataTuple() : TypeData(6ULL, 0) {} + TypeDataTuple() : TypeData(6ULL, 0, 1) {} static TypePtr singleton; friend void type_system_init(); @@ -209,7 +209,7 @@ public: * It's like "untyped callable", not compatible with other types. */ class TypeDataContinuation final : public TypeData { - TypeDataContinuation() : TypeData(7ULL, 0) {} + TypeDataContinuation() : TypeData(7ULL, 0, 1) {} static TypePtr singleton; friend void type_system_init(); @@ -224,12 +224,12 @@ public: /* * `null` has TypeDataNullLiteral type. - * Currently, it can be assigned to int/slice/etc., but later Tolk will have T? types and null safety. + * It can be assigned only to nullable types (`int?`, etc.), to ensure null safety. * Note, that `var i = null`, though valid (i would be constant null), fires an "always-null" compilation error * (it's much better for user to see an error here than when he passes this variable somewhere). */ class TypeDataNullLiteral final : public TypeData { - TypeDataNullLiteral() : TypeData(8ULL, 0) {} + TypeDataNullLiteral() : TypeData(8ULL, 0, 1) {} static TypePtr singleton; friend void type_system_init(); @@ -242,6 +242,27 @@ public: bool can_be_casted_with_as_operator(TypePtr cast_to) const override; }; +/* + * `T?` is "nullable T". + * It can be converted to T either with ! (non-null assertion operator) or with smart casts. + */ +class TypeDataNullable final : public TypeData { + TypeDataNullable(uint64_t type_id, int children_flags, int width_on_stack, TypePtr inner) + : TypeData(type_id, children_flags, width_on_stack) + , inner(inner) {} + +public: + const TypePtr inner; + + static TypePtr create(TypePtr inner); + + std::string as_human_readable() const override; + bool can_rhs_be_assigned(TypePtr rhs) const override; + bool can_be_casted_with_as_operator(TypePtr cast_to) const override; + void traverse(const TraverserCallbackT& callback) const override; + TypePtr replace_children_custom(const ReplacerCallbackT& callback) const override; +}; + /* * `fun(int, int) -> void` is TypeDataFunCallable, think of is as a typed continuation. * A type of function `fun f(x: int) { return x; }` is actually `fun(int) -> int`. @@ -249,7 +270,7 @@ public: */ class TypeDataFunCallable final : public TypeData { TypeDataFunCallable(uint64_t type_id, int children_flags, std::vector&& params_types, TypePtr return_type) - : TypeData(type_id, children_flags) + : TypeData(type_id, children_flags, 1) , params_types(std::move(params_types)) , return_type(return_type) {} @@ -275,7 +296,7 @@ public: */ class TypeDataGenericT final : public TypeData { TypeDataGenericT(uint64_t type_id, std::string&& nameT) - : TypeData(type_id, flag_contains_genericT_inside) + : TypeData(type_id, flag_contains_genericT_inside, -999999) // width undefined until instantiated , nameT(std::move(nameT)) {} public: @@ -286,7 +307,6 @@ public: std::string as_human_readable() const override { return nameT; } bool can_rhs_be_assigned(TypePtr rhs) const override; bool can_be_casted_with_as_operator(TypePtr cast_to) const override; - int calc_width_on_stack() const override; }; /* @@ -296,8 +316,8 @@ public: * A tensor can be empty. */ class TypeDataTensor final : public TypeData { - TypeDataTensor(uint64_t type_id, int children_flags, std::vector&& items) - : TypeData(type_id, children_flags) + TypeDataTensor(uint64_t type_id, int children_flags, int width_on_stack, std::vector&& items) + : TypeData(type_id, children_flags, width_on_stack) , items(std::move(items)) {} public: @@ -312,7 +332,6 @@ public: bool can_be_casted_with_as_operator(TypePtr cast_to) const override; void traverse(const TraverserCallbackT& callback) const override; TypePtr replace_children_custom(const ReplacerCallbackT& callback) const override; - int calc_width_on_stack() const override; }; /* @@ -322,7 +341,7 @@ public: */ class TypeDataTypedTuple final : public TypeData { TypeDataTypedTuple(uint64_t type_id, int children_flags, std::vector&& items) - : TypeData(type_id, children_flags) + : TypeData(type_id, children_flags, 1) , items(std::move(items)) {} public: @@ -346,7 +365,7 @@ public: * The only thing available to do with unknown is to cast it: `catch (excNo, arg) { var i = arg as int; }` */ class TypeDataUnknown final : public TypeData { - TypeDataUnknown() : TypeData(20ULL, flag_contains_unknown_inside) {} + TypeDataUnknown() : TypeData(20ULL, flag_contains_unknown_inside, 1) {} static TypePtr singleton; friend void type_system_init(); @@ -367,7 +386,7 @@ public: */ class TypeDataUnresolved final : public TypeData { TypeDataUnresolved(uint64_t type_id, std::string&& text, SrcLocation loc) - : TypeData(type_id, flag_contains_unresolved_inside) + : TypeData(type_id, flag_contains_unresolved_inside, -999999) , text(std::move(text)) , loc(loc) {} @@ -380,7 +399,6 @@ public: std::string as_human_readable() const override { return text + "*"; } bool can_rhs_be_assigned(TypePtr rhs) const override; bool can_be_casted_with_as_operator(TypePtr cast_to) const override; - int calc_width_on_stack() const override; }; /* @@ -389,7 +407,7 @@ public: * Empty tensor is not compatible with void, although at IR level they are similar, 0 stack slots. */ class TypeDataVoid final : public TypeData { - TypeDataVoid() : TypeData(10ULL, 0) {} + TypeDataVoid() : TypeData(10ULL, 0, 0) {} static TypePtr singleton; friend void type_system_init(); @@ -400,7 +418,6 @@ public: std::string as_human_readable() const override { return "void"; } bool can_rhs_be_assigned(TypePtr rhs) const override; bool can_be_casted_with_as_operator(TypePtr cast_to) const override; - int calc_width_on_stack() const override; };