Applying the fix for 'unset -f' exposed a crashing bug in lookup()
in sh/nvdisc.c, which is the function for looking up discipline
functions. This is what caused tests/variables.sh to crash.
Ref.: https://github.com/ksh93/ksh/issues/23#issuecomment-645699614
src/cmd/ksh93/sh/nvdisc.c: lookup():
- To avoid segfault, check that the function pointer nq->nvalue.rp
is actually set before checking if nq->nvalue.rp->running==1.
src/cmd/ksh93/sh/xec.c,
src/cmd/ksh93/tests/functions.sh:
- Uncomment the 'unset -f' fix from b7932e87.
Resolves#21 (again).
The fix in sh/xec.c, which was backported from the ksh 93v- beta to
delay the actual removal of a running function that unsets itself,
caused a segfault in the variables.sh regression tests (see #23).
src/cmd/ksh93/sh/xec.c:
- Comment out the backported code pending a correct fix for #21.
Now both types of functions silently fail to unset themselves
(unless they're discipline functions).
src/cmd/ksh93/tests/functions.sh:
- Disable regression tests checking that the function was actually
unset, pending a correct fix for #21.
Resolves: #23Reopens: #21
src/cmd/ksh93/sh/name.c:
- Correct the check for when a function is currently running
to fix a segmentation fault that occurred when a POSIX
function tries to unset itself while it is running.
This bug fix was backported from ksh93v-.
src/cmd/ksh93/sh/xec.c:
- If a function tries to unset itself, unset the function
with '_nv_unset(np, NV_RDONLY)' to fix a silent failure.
This fix was also backported from ksh93v-.
src/cmd/ksh93/tests/functions.sh:
- Add four regression tests for when a function unsets itself.
Resolves#21
Note that shtests simply does a 'grep -c err_exit' and substracts 1
to count the number of regression tests in a test script. Not all
test scripts make temp dirs, so subtracting 2 instead won't do.
src/cmd/ksh93/tests/*.sh:
- Escape the err_exit call in the routine to create a temporary
directory so that it is not counted as a regression test.
That bypasses the alias, so we have to pass $LINENO manually.
Four added tests did not correctly report their line numbers
upon failure and were counted as one, because the err_exit
alias/function pair was called from a shell function.
Note that shtests simply does a 'grep -c err_exit' to count the
number of regression tests in a test script.
src/cmd/ksh93/tests/subshell.sh:
- check_hash_table():
- Take line number as 1st argument.
- Quote a character in err_exit to bypass the alias when calling
it, so we can pass on the argument for the line number. This
also stops this helper function from being counted as a test.
- When calling check_hash_table(), pass $LINENO.
- Add dummy err_exit comments to have the tests counted.
Ksh was not checking for `command` when running a special builtin,
which caused preceding invocation-local variable assignments to
become global. This is the reproducer from the att/ast#72:
$ foo=BUG command eval ':'
$ echo "$foo"
This no longer prints 'BUG', as ksh now makes sure the command builtin
is not running a special builtin before making invocation-local
variable assignments global.
src/cmd/ksh93/sh/xec.c:
- Backport the bugfix for BUG_CMDSPASGN from ksh93v- 2013-10-10-alpha.
src/cmd/ksh93/tests/builtins.sh:
- Add a regression test based on the reproducer in att/ast#72.
As previously reported in rhbz#1112306 (https://bugzilla.redhat.com/show_bug.cgi?id=1112306),
ksh may crash when receiving SIGCHLD because GCC's optimizer will fail to generate
`addl` and `sub` instructions to increment and decrement `job.in_critical` in the
`job_subsave` function. This bug *does* occur in GCC 10 with `-O2`, but not `-O1`;
it doesn't appear this bug has been fixed. As a reference, here is the relevant
debug assembly output of `job_subsave` when KSH is compiled with `CCFLAGS` set to
`-g -O1`:
0000000000034c97 <job_subsave>:
void *job_subsave(void)
{
34c97: 53 push %rbx
struct back_save *bp = new_of(struct back_save,0);
34c98: bf 18 00 00 00 mov $0x18,%edi
34c9d: e8 34 4a 0a 00 callq d96d6 <_ast_malloc>
34ca2: 48 89 c3 mov %rax,%rbx
job_lock();
34ca5: 83 05 3c 50 13 00 01 addl $0x1,0x13503c(%rip) # 169ce8 <job+0x28>
*bp = bck;
34cac: 66 0f 6f 05 4c 5a 13 movdqa 0x135a4c(%rip),%xmm0 # 16a700 <bck>
34cb3: 00
34cb4: 0f 11 00 movups %xmm0,(%rax)
34cb7: 48 8b 05 52 5a 13 00 mov 0x135a52(%rip),%rax # 16a710 <bck+0x10>
34cbe: 48 89 43 10 mov %rax,0x10(%rbx)
bp->prev = bck.prev;
34cc2: 48 8b 05 47 5a 13 00 mov 0x135a47(%rip),%rax # 16a710 <bck+0x10>
34cc9: 48 89 43 10 mov %rax,0x10(%rbx)
bck.count = 0;
34ccd: c7 05 29 5a 13 00 00 movl $0x0,0x135a29(%rip) # 16a700 <bck>
34cd4: 00 00 00
bck.list = 0;
34cd7: 48 c7 05 26 5a 13 00 movq $0x0,0x135a26(%rip) # 16a708 <bck+0x8>
34cde: 00 00 00 00
bck.prev = bp;
34ce2: 48 89 1d 27 5a 13 00 mov %rbx,0x135a27(%rip) # 16a710 <bck+0x10>
job_unlock();
34ce9: 8b 05 f9 4f 13 00 mov 0x134ff9(%rip),%eax # 169ce8 <job+0x28>
34cef: 83 e8 01 sub $0x1,%eax
34cf2: 89 05 f0 4f 13 00 mov %eax,0x134ff0(%rip) # 169ce8 <job+0x28>
34cf8: 75 2b jne 34d25 <job_subsave+0x8e>
34cfa: 8b 3d ec 4f 13 00 mov 0x134fec(%rip),%edi # 169cec <job+0x2c>
34d00: 85 ff test %edi,%edi
34d02: 74 21 je 34d25 <job_subsave+0x8e>
34d04: c7 05 da 4f 13 00 01 movl $0x1,0x134fda(%rip) # 169ce8 <job+0x28>
When `-O2` is used instead of `-O1`, the `addl` and `sub` instructions for
incrementing and decrementing the lock are removed. GCC instead generates a
broken `mov` instruction for `job_lock` and removes the initial `sub` instruction
in job_unlock (this is also seen in Red Hat's bug report):
job_lock();
*bp = bck;
37d7c: 66 0f 6f 05 7c 79 14 movdqa 0x14797c(%rip),%xmm0 # 17f700 <bck>
37d83: 00
struct back_save *bp = new_of(struct back_save,0);
37d84: 49 89 c4 mov %rax,%r12
job_lock();
37d87: 8b 05 5b 6f 14 00 mov 0x146f5b(%rip),%eax # 17ece8 <job+0x28>
...
job_unlock();
37dc6: 89 05 1c 6f 14 00 mov %eax,0x146f1c(%rip) # 17ece8 <job+0x28>
37dcc: 85 c0 test %eax,%eax
37dce: 75 2b jne 37dfb <job_subsave+0x8b>
The original patch works around this bug by using the legacy `__sync_fetch_and_add/sub`
GCC builtins. This forces GCC to generate instructions that change the lock with
`lock addl`, `lock xadd` and `lock subl`:
job_lock();
37d9f: f0 83 05 41 6f 14 00 lock addl $0x1,0x146f41(%rip) # 17ece8 <job+0x28>
37da6: 01
...
job_unlock();
37deb: f0 0f c1 05 f5 6e 14 lock xadd %eax,0x146ef5(%rip) # 17ece8 <job+0x28>
37df2: 00
37df3: 83 f8 01 cmp $0x1,%eax
37df6: 74 08 je 37e00 <job_subsave+0x70>
...
37e25: 74 11 je 37e38 <job_subsave+0xa8>
37e27: f0 83 2d b9 6e 14 00 lock subl $0x1,0x146eb9(%rip) # 17ece8 <job+0x28>
While this does work, it isn't portable. This patch implements a different
workaround for this compiler bug. If `job_lock` is put at the beginning of
`job_subsave`, GCC will generate the required `addl` and `sub` instructions:
job_lock();
37d67: 83 05 7a 5f 14 00 01 addl $0x1,0x145f7a(%rip) # 17dce8 <job+0x28>
...
job_unlock();
37dbb: 83 e8 01 sub $0x1,%eax
37dbe: 89 05 24 5f 14 00 mov %eax,0x145f24(%rip) # 17dce8 <job+0x28>
It is odd that moving a single line of code fixes this problem, although
GCC _should_ have generated these instructions from the original code anyway.
I'll note that this isn't the only way to get these instructions to generate.
The problem also seems to go away when inserting almost anything else inside
of the code for `job_subsave`. This is just a simple workaround for a strange
compiler bug.
This bug was previously reported in att/ast#37.
Ksh ignores `-r` when `read -r -d` is run because when
the bit for `D_FLAG` is set, the bit for `R_FLAG` is unset
as a side effect of setting `D_FLAG`. The following set
of commands fails to print a backslash:
$ printf '\\\000' | read -r -d ''
$ echo $REPLY
The fix for this bug is to set `D_FLAG` with `D_FLAG + 1`,
which prevents `R_FLAG` from being unset. This bugfix
has been backported from ksh93v- 2013-10-10-alpha.
src/cmd/ksh93/bltins/read.c:
- Set `D_FLAG` with `D_FLAG + 1` to prevent the bit for
`R_FLAG` from being unset.
src/cmd/ksh93/tests/builtins.sh:
- Add the regression test for `read -r -d` from ksh93v-.
Running shbench after undoing the incorrect subshell optimisation
showed that the performance of ${ subshare; }-type command
substitutions went down very slightly, but consistently.
The main purpose of using this ksh-specific type of command
substitution vs. a normal one is performance. Thus, it *is*
appropriate to eke every last bit of performance out of it that
we can, provided correctness is completely preserved.
It is also a type of command substitution where every change is
supposed to be shared with the main shell environment; only command
output is captured in a subshell-like fashion.
Thus, on the face of it, it would be a logical optimisation for
sh_assignok() to avoid bothering with saving a subshell context for
variables if we're in a subshare.
Lo and behold, applying it does not introduce any regress fails.
Here are my average results of the braces.ksh benchmark from
shbench <http://fossil.0branch.com/csb/tktnew> against stock
/bin/ksh 93u+ vs. current 93u+m (same compiler flags),
100 runs pre-optimisation and 100 runs post-optimisation:
Stock /bin/ksh: Pre-optimisation (at 3d38270b):
93u+: 0.743 secs 93u+m: 0.739 secs
Stock /bin/ksh: Post-optimisation (now):
93u+: 0.744 secs 93u+m: 0.726 secs
The left column shows only a small margin of error with 100 runs;
the right one shows a very small, but not insignificant difference.
However, these tests were not very rigorous with 100 runs each.
If anyone wants to do it properly, please report results to
korn-shell@googlegroups.com. I'm happy enough with this, though.
Thanks to Joerg van den Hoff for providing shbench, without
which it would not have occurred to me to try this.
src/cmd/ksh93/sh/subshell.c: sh_assignok():
- Don't bother if we're in a ${ subshare; }.
The fact that every ksh builtin command self-documents with the
options --help and --man (and others, see 'getopts --man'; but
these are the essential ones) is poorly known; the information is
buried somewhere deep in the sh.1 manual page, and is incomplete at
that. None of the terse usage messages displayed on error point the
user to these options. So let's fix that.
src/lib/libast/misc/optget.c:
- Change generic 'options' placeholder, used in all terse usage
messages, to 'options | --help | --man'.
src/cmd/ksh93/sh.1:
- Edit documentation of --man/-?, adding documentation on --help
which was completely undocumented. Refer to 'getopts --man' for
more advanced info.
- Separate these from the (important) documentation on special
builtins using a paragraph break.
ksh 93v- beta introduced a regression with nested command
substitutions: backticks nested in $( ) result in misdirected
output. This has never been in 93u+, but since we're often
backporting things, let's avoid backporting this bug. It is also
useful if this shows up when running our bin/shtests against the
actual beta by adding a SHELL=... argument.
Ref.: https://github.com/att/ast/issues/478
src/cmd/ksh93/tests/subshell.sh:
- Add reproducer submitted by the reporter as a regression test.
This bug was originally reported by @lijog in att/ast#7 and has been
reported again in #15. KSH does not save the state of a variable if it
is in a newer scope. This is because of an optimization in sh_assignok
first introduced in ksh93t+ 2010-05-24. Here is the code change in that
version:
return(np);
/* don't bother to save if in newer scope */
- if(!(rp=shp->st.real_fun) || !(dp=rp->sdict))
- dp = sp->var;
- if(np->nvenv && !nv_isattr(np,NV_MINIMAL|NV_EXPORT) && shp->last_root)
- dp = shp->last_root;
- if((mp=nv_search((char*)np,dp,HASH_BUCKET))!=np)
- {
- if(mp || !np->nvfun || np->nvfun->subshell>=sh.subshell)
- return(np);
- }
+ if(sp->var!=shp->var_tree && shp->last_root==shp->var_tree)
+ return(np);
if((ap=nv_arrayptr(np)) && (mp=nv_opensub(np)))
{
This change was originally made to replace a buggier optimization.
However, the current optimization causes variables set in subshells
to wrongly affect the environment outside of the subshell, as the
variable does not get set back to its original value. This patch
simply removes the buggy optimization to fix this problem.
src/cmd/ksh93/sh/subshell.c:
- Remove a buggy optimization that caused variables set in subshells
to affect the environment outside of the subshell.
src/cmd/ksh93/tests/subshell.sh:
- Add a regression test for setting variables in subshells. This
test has to be run from the disk after being created with a here
document because it always returns the expected result when run
directly in the regression test script.
The 'source' alias is now converted into a regular built-in command
so that 'unalias -a' does not remove it, and something like
cmd=source; $cmd name args
will now work.
This is part of the project to replace default aliases that define
essential commands by proper builtins that act identically (except
you now get the actual command's name in any error/usage messages).
src/cmd/ksh93/data/aliases.c:
- Remove 'source' default alias.
src/cmd/ksh93/data/builtins.c,
src/cmd/ksh93/include/builtins.h:
- Define 'source' regular builtin with extra parser ID "SYSSOURCE".
Same definition as '.', minus the BLT_SPC flag indicating a
special builtin. This preserves the behaviour of 'command .'.
- Update sh_optdot[] to include info for 'source --man'.
(Note that \f?\f expands to the current command name.
This allows several commands to share a single --man page.)
src/cmd/ksh93/sh/parse.c:
- In the two places that SYSDOT is checked for, also check for
SYSSOURCE, making sure the two commands are parsed identically.
src/cmd/ksh93/sh.1:
- Remove 'source' default alias.
- Document 'source' regular builtin.
This reverts the patch for the job_lock and job_unlock macros.
As I said in 58560db7, this is very probably a workaround for an
optimiser bug in certain versions of GCC, at least 2017 versions.
In the version I committed, that workaround version never gets
used, because you cannot use #if defined(...) to check for the
presence of a compiler builtin. Thanks to Johnothan King for
keeping an eye on my code and pointing that out to me.
What is needed to test for the presence of a compiler builtin is a
builtin macro called __has_builtin (and it *can* be tested for
using #if defined...()).. This is a clang invention. It was not
added to gcc until version 10, which was only released in a first
stable version just over a month ago.
Ref.: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66970#c14https://gcc.gnu.org/gcc-10/
However, for gcc 10, it seems unlikely the patch is still needed.
(Although it would certainly be a good idea to test that.)
And for the older gcc versions that do need it, we cannot use
__has_builtin, which means we need to define a dummy that always
returns false, so the workaround version is never used.
Ref.: https://github.com/ksh93/ksh/commit/58560db7#commitcomment-39900259
And we cannot use the workaround version unconditionally either,
because it would cause build failures on compilers without the
__sync_fetch_and_add() and __sync_fetch_and_sub() builtins.
Which means the only sensible thing left to do is to revert the
patch for now.
As far as I can tell at this point, for the patch to return to this
upstream in a sensible manner, someone would need to:
1. Write a small C program that tests these macros and somehow
checks for the presence of that optimiser bug. (This is not
going to come from me; my C-fu is simply not good enough.)
2. Incorporate that into the distribution as a test for iffe.
(Few know how iffe works, but it's probably not that hard as
there are plenty of existing tests to use as a template.)
3. Reinsert the workaround version using the macro defined (or not)
by that iffe test, so that it is only compiled in when using a
compiler that actually has the bug in question.
Until then, this can just continue to be an OS-specific patch for
systems with GCC versions that might have that bug.
The more I think about it, the more it seems obvious that commit
07cc71b8 (PR #14) is quite simply a workaround for a GCC optimiser
bug, and (who knows?) possibly an old, long-fixed one, as the bug
report is years old.
The commit also caused ksh to fail to build on HP-UX B.11.11 with
GCC 4.2.3 (hosted at polarhome.com), because it doesn't have
__sync_fetch_and_add() and __sync_fetch_and_sub(). It may fail on
other systems. The GCC documentation says these are legacy:
https://gcc.gnu.org/onlinedocs/gcc/_005f_005fsync-Builtins.html
HELP WANTED: what I would like best is if someone could come up
with some way of detecting this optimiser bug and then error out
with a message along the lines of "please upgrade your broken
compiler". It would probably need to be a new iffe test.
Meanwhile, let's try it this way for a while and see what happens:
src/cmd/ksh93/include/jobs.h:
- Restore original ksh version of job_lock()/job_unlock() macros.
- Use the workaround version only if the compiler has the builtins
__sync_fetch_and_add() and __sync_fetch_and_sub().
ksh segfaults in job_chksave after receiving SIGCHLD
https://bugs.launchpad.net/ubuntu/+source/ksh/+bug/1697501
Eric Desrochers wrote on 2017-06-12:
[Impact]
* The compiler optimization dropped parts from the ksh job
locking mechanism from the binary code. As a consequence, ksh
could terminate unexpectedly with a segmentation fault after
it received the SIGCHLD signal.
[Test Case]
Unfortunately, there is no clear and easy way to reproduce the
segfault.
* But the original reporter of this bug can randomly reproduce
the problem using an in-house ksh script that only works
inside his infrastructure as follow : "ksh
<in-house-script.ksh>" and then once in a while ksh will
segfault as follow :
(gdb) bt
#0 job_chksave (pid=pid@entry=19003) at /build/ksh-6IEHIC/ksh-93u+20120801/src/cmd/ksh93/sh/jobs.c:1948
#1 0x00000000004282ab in job_reap (sig=17) at /build/ksh-6IEHIC/ksh-93u+20120801/src/cmd/ksh93/sh/jobs.c:428
#2 <signal handler called>
...
[Regression Potential]
* Regression risk : low/none expected, the package has been
highly/intensively tested by a user who run over 18M ksh
scripts a day on each of their clusters.
[...]
* The fix has been written by RH and has been proven to work for
them for the last 3 years.
* A test package including the RH fix has been intensively tested
and verified (pre-SRU) by an affected user with positive
feedbacks using a reproducer that segfault without the RH
patch.
* Test package (pre-SRU) feedbacks :
https://bugs.launchpad.net/ubuntu/xenial/+source/ksh/+bug/1697501/comments/7
[Other Info]
* Details about the RH bug :
- https://bugzilla.redhat.com/show_bug.cgi?id=1123467
- https://bugzilla.redhat.com/show_bug.cgi?id=1112306
- https://access.redhat.com/solutions/1253243
- http://rhn.redhat.com/errata/RHBA-2014-1015.html
- ksh.spec
* Fri Jul 25 2014 Michal Hlavinka <email address hidden> - 20120801-10.8
* job locking mechanism did not survive compiler optimization (#1123467)
- patch
* ksh-20120801-locking.patch
Debian bug:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=867181
[Original Description]
# gdb
[New LWP 3882]
Core was generated by `/bin/ksh <KSH_SCRIPT>.ksh'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0 job_chksave (pid=pid@entry=19385) at /build/ksh-6IEHIC/ksh-93u+20120801/src/cmd/ksh93/sh/jobs.c:1948
1948 if(jp->pid==pid)
(gdb) p *jp
Cannot access memory at address 0xb
(gdb) p *jp->pid
Cannot access memory at address 0x13
(gdb) p pid
$2 = 19385
(gdb) p *jpold
$1 = {next = 0xb, pid = -604008960, exitval = 11124}
The struct is corrupted at some point looking at the next,pid and
exitval struct members values which isn't valid data.
# assembly code
=> 0x0000000000427159 <+41>: cmp %edi,0x8(%rdx)
(gdb) p $edi ## pid variable
$1 = 19385
(gdb) p *($rdx + 8) ## jp->pid struct
Cannot access memory at address 0x13
--
ksh is segfaulting because it can't access struct "jp" ($rdx)
thus cannot de-reference the struct member "jp>pid" ($rdx + 8) at
line : src/cmd/ksh93/sh/jobs.c:1948 when looking if jp->pid is
equal to pid ($edi) variable.
Prior to this bugfix, the following set of commands would
fail to print two double quotes:
IFS=',' read -S a b c <<<'foo,"""title"" data",bar'
echo $b
This fix is from ksh93v- 2013-10-10-alpha, although it has
been revised to use stakputc to put the required double quote
into the buffer for consistency with the ksh93u+ codebase.
src/cmd/ksh93/bltins/read.c:
- When handling nested double quotes, put the required double
quote in read's buffer with stakputc.
src/cmd/ksh93/tests/builtins.sh:
- Add the regression test for `read -S` from ksh93v-.
src/cmd/ksh93/sh.1:
- Fix a minor formatting error to highlight '-S' in the ksh(1)
man page.
Fixes for implicit declaration warnings
This is to upstream some minor fixes for implicit declaration warnings currently in Debian.
Author: Anuradha Weeraman aweeraman@gmail.com
Reviewed-By: Thorsten Glaser t.glaser@tarent.de
Last-Update: 2020-02-09
src/cmd/ksh93/edit/history.c:
- The `time_t` data type is usually 64-bit, so cast `t` to an
unsigned long and use the correct format specifier with sfprintf.
src/lib/libast/tm/tmdata.c:
- Update the leap second array by adding the current leap
seconds that were absent, being:
2016-12-31+23:59:60 1483228826
2015-06-30+23:59:60 1435708825
2012-06-30+23:59:60 1341100824
I couldn't find a proper list of leap seconds in Unix Epoch time,
so I used an Epoch Converter:
https://www.epochconverter.com/
The converter doesn't support leap seconds, so each conversion
had to be corrected by adding the number of leap seconds required
(e.g. 1230767999 + 24 = 1230768023, for 2008-12-31+23:59:60 in GMT).
'cd' with no argument is supposed to go to your home directory, but
in this override function it produced an error due to an incorrect
use of $@ within another parameter substitution, preventing $@ from
expanding to zero words.
Ref.: https://groups.google.com/d/msgid/korn-shell/781ed34b-6860-5d47-dfe8-be21280a6b31%40inlv.org
src/cmd/ksh93/fun/dirs: _cd():
- Fix the bug by setting a positional parameter if needed,
then using a normal and correct "$@" expansion for \cd.
- While we're here, restore a historic comment about using
'command' to bypass aliases that exists in the listing of this
function in the official 1995 KornShell book (pp. 274-276).
Presumably, that comment was deleted when they decided to add the
obnoxious and broken default alias command='command ', whenever
that was. This branch got rid of that default alias in 61d9bca5.
This fix for `printf '%T' now` on FreeBSD was written by
@krader1961. This is from https://github.com/att/ast/pull/591:
On FreeBSD calling tzset() does not guarantee the tzname array will
be correctly populated. On most systems that works but on FreeBSD you
have to call localtime() or a related function (e.g., ctime()).
This change also eliminates a potential, very small, memory leak due to
the strdup()'ed tznames not being freed.
src/lib/libast/tm/tminit.c:
- Fix timezone name determination on FreeBSD and a memory leak.
The regression this commit fixes was first introduced in ksh93t
2008-07-25. It was previously worked around in 6f0e008c by forking
subshells if any special environment variable is unset.
The reason why this problem doesn't occur in ksh93s+ is because in
that version of ksh sh_assignok never moves nodes, it only clones
them. The second argument doesn't set NV_MOVE, which makes
`sh_assignok(np,0)` is similar to `sh_assignok(np,1)`. In ksh93t and
higher, setting the second argument to zero causes the node to be moved
with NV_MOVE, which causes the discipline function associated with
the variable node to be removed when `np->nvfun` is set to zero (i.e.
NULL). This is why a command like `(unset LC_NUMERIC; LC_NUMERIC=invalid)`
doesn't print a diagnostic, as it looses its discipline function.
This patch fixes the problem by cloning the node with sh_assignok
if it is a special variable with a discipline function. This allows
special variables to work as expected in virtual subshells. The
original workaround has been kept for the $PATH variable only, as
hash tables are still broken in virtual subshells. It has been updated
accordingly to only fork subshells if it detects the variable node
for PATH. I have added two more regression tests for changing the
PATH in subshells to make sure hash tables continue working as
expected with this fix.
src/cmd/ksh93/bltins/typeset.c:
- Only fork virtual subshells if the PATH will be changed. If a
variable is a special variable with a discipline function, it
should be just be cloned, not moved.
src/cmd/ksh93/sh/nvdisc.c:
- Add a comment to clarify that NV_MOVE will delete the discipline
function associated with the node.
src/cmd/ksh93/tests/subshells.sh:
- Add two more regression tests for unsetting the PATH in subshells,
one for if PATH is being pointed to by a nameref. Condense the
hash table tests by moving the main test into a single function.
Test 137(C) was failing on some systems because $TMPDIR was set and
the local vi(1) honours it, so that the expected '/tmp/' string was
never output by vi. For compatibility with vi programs that honour
$TMPDIR and those that always use /tmp, we must export TMPDIR=/tmp.
src/cmd/ksh93/tests/pty.sh:
- Export TMPDIR=/tmp for test 137(C).
Note that this exports TMPDIR to the environment for the
duration of the 'tst' function run because the function was
defined using the ksh 'function tst { ...; }' syntax.
The lib/package/ast-open.* are almost entirely obsolete after
most of the AST utilities were removed to concentrate on ksh93
development. The only part that is both relevant and not repeated
elsewhere is AT&T's introductory blurb on ksh93.
README.md:
- Add the official AT&T ksh93 description from ast-open.html,
reformatted to Markdown with a couple of added hyperlinks.
lib/package/ast-open.README,
lib/package/ast-open.html:
- Removed. (Note that the historic ksh93 changelog in these files
is also included in src/cmd/ksh93/RELEASE).
We cannot identify the new builtins as coming from "AT&T Research"
as they don't. 'ksh community' is accurate but a bit long. Since
we're now at a GitHub org called ksh93, let's just use that.
The man page for the builtin command says special builtins cannot
be deleted. This wasn't the case though, running `builtin -d` on
a special builtin was deleting it. As an example, the following
set of commands was ending with 'export: not found':
$ builtin -d export
$ export foo=bar
This commit backports the bugfix from ksh93v- (2014-12-24-beta),
which added an error check to prevent special builtins from being
deleted.
src/cmd/ksh93/sh/nvdisc.c:
- Add an error check to prevent special builtins from being deleted.
src/cmd/ksh93/tests/builtins.sh
- Add a regression test for using `builtin -d` on special builtins.
This enables the tests/pty.sh regression tests for the interactive
shell to work. I've managed to restore these without restoring
nmake (which would cause unresolved build failures). It involved
hand-editing the Mamfile originally generated by nmake.
src/cmd/builtin/Makefile:
- Restored. Even though nmake is not installed, apparently mamake
still needs the Makefile for the 'install' target to work. This
will probably need editing if nmake is ever restored.
src/cmd/builtin/Mamfile:
- Restored and edited by hand to remove the commands to build all
the utilities except pty, as well as all test targets (pty has no
test). Most of this was systematically deleting blocks delimited
by 'make FOO' and 'done FOO generated' until the build stopped
failing -- a task made so much easier by the build now failing
gracefully on error (see d0dfb37c). To see the diff, do:
git diff cc1f2bf8 src/cmd/builtin/Mamfile
src/cmd/builtin/RELEASE:
- Restored for completeness' sake with all non-pty entries removed.
src/cmd/builtin/features/pty,
src/cmd/builtin/pty.c:
- Restored.
src/cmd/ksh93/tests/io.sh:
- 16 tests for the 'redirect' builtin used the ((arithmetic
command)) to check the result. This does not tolerate the result
being a non-number, such as the empty string, which may occur on
test failure. So use [[ ... -eq ArithExpression ]] to check these
results instead; it simply returns false, failing gracefully.
This commit removes the undocumented 'login' and 'newgrp' builtin
commands. They already stopped blocking shell functions by that
name by changing from special to regular builtins in 04b91718 (a
change I forgot to mention in that commit message), but there is
another obnoxious aspect to these: being glorified hooks into
'exec', they replaced your shell session with the external commands
by the same name. This makes argument and error checking
impossible, so if you made so much as a typo, you would be
immediately logged out.
Even if that behaviour is wanted by a few, having it as the default
is user-hostile enough to be called a bug. It also violates the
POSIX definition of the 'newgrp' utility which explicitly says that
it "shall create a new shell execution environment", not replace
the existing one.
https://pubs.opengroup.org/onlinepubs/9699919799/utilities/newgrp.html
Users who do want this behaviour can easily restore it by setting:
alias login='exec login'
alias newgrp='exec newgrp'
src/cmd/ksh93/bltins/misc.c:
- As there is no more 'login' builtin, combine b_exec() and
B_login() functions, which allows eliminating a few variables.
Note that most of 'exec' was actually implemented in B_login()!
src/cmd/ksh93/data/builtins.c:
- Remove "login" and "newgrp" table entries.
src/cmd/ksh93/include/builtins.h:
- Remove SYSLOGIN parser ID. As this was the first, all the others
needed renumbering.
src/cmd/ksh93/sh/xec.c:
- Remove SYSLOGIN parser check that made 'login' and 'newgrp' act
like 'exec' and replace the shell.
This commit converts the redirect='command exec' alias to a regular
'redirect' builtin command that only accepts I/O redirections, which
persist as in 'exec'. This means that:
* 'unlias -a' no longer removes the 'redirect' command;
* users no longer accidentally get logged out of their shells if
they type something intuitive but wrong, like 'redirect ls >file'.
This should not introduce any legitimate change in behaviour. If
someone did accidentally pass non-redirection arguments to
'redirect', unexpected behaviour would occur; this now produces
an 'incorrect syntax' error.
src/cmd/ksh93/bltins/misc.c: b_exec():
- Recognise 'redirect' when parsing options.
- If invoked as 'redirect', produce error if there are arguments.
src/cmd/ksh93/data/aliases.c:
- Remove redirect='command exec' alias.
src/cmd/ksh93/data/builtins.c:
- Update/improve comments re ordering.
- Add 'redirect' builtin entry.
- sh_optexec[]: Abbreviate redirection-related documentation;
refer to redirect(1) instead.
- sh_optredirect[]: Add documentation.
src/cmd/ksh93/include/builtins.h:
- Add SYSREDIR parser ID, renumbering those following it.
- Improve comments.
- Add extern sh_optredirect[].
src/cmd/ksh93/sh.1:
- exec: Abbreviate redirection-related documentation; refer to
'redirect' instead.
- redirect: Add documentation.
src/cmd/ksh93/sh/xec.c:
- Recognise SYSREDIR parser ID in addition to SYSEXEC when
determining whether to make redirections persistent.
src/cmd/ksh93/tests/io.sh:
- To regress-test the new builtin, change most 'command exec' uses
to 'redirect'.
- Add tests verifying the exit behaviour of 'exec', 'command exec',
'redirect' on redirections.
The exit status fix introduced in cherry-picked commit 22c3a6e1
introduced a problem: interrupting the build only interrupted the
main process, and the background job would happily continue.
bin/package, src/cmd/INIT/package.sh:
- Give up on the idea of a background job and put up with not
cleaning up the FIFO. It's in a build directory that is going to
get cleaned up anyway.
src/cmd/INIT/cc.darwin*:
- Build on ancient Mac OS X gcc versions (darwin 7+, Mac OS X
10.3+), more recent gcc versions (darwin 11+, Mac OS X 10.7+), as
well as current clang versions. So there are now three cc.darwin*
wrapper scripts for these different darwin versions.
bin/package, src/cmd/INIT/package.sh:
- Differentiate the host ID and pick the correct Darwin script based on
the detected OS release version.
Both 'alias' and 'unalias' are specified as regular builtins. Among
a few other things, that means it ought to be portable to use these
names for shell functions. But ksh93 disallowed that until now.
src/cmd/ksh93/data/builtins.c:
- Make 'unalias' a regular builtin by removing the BLT_SPC flag.
- (same fix for 'alias' was already done in afa68dca)
- Add the BLT_ENV flag to the 'alias' and 'hash' commands. In
include/name.h, this flag is commented: "non-stoppable, can
modify environment". The "non-stoppable" bit seems like a good
idea: these operations should not be interruptable as that would
cause an inconsistent state.
src/cmd/ksh93/sh.1:
- Remove the '-', indicating special builtin, from 'alias' entry.
- Minor cosmetic fix: space after the '-' for 'unset'.
(cherry picked from commit a4315d7672204acb543010b4d4916b22dcb9cb08)
The b_hash() function duplicated much of its code from b_alias(),
while b_alias() retained some code to support being called as
'hash'. There is no reason why 'hash' and 'alias' can't be handled
with a single function, as is the case several other builtins. Note
that option parsing can easily be made dependent on the name the
command was invoked with (in this case, argv[0]=='h').
The new hash builtin's -r option cleared the hash table by
assigning to PATH its existing value, triggering its associated
discipline function (put_restricted() in init.c) which then
actually cleared the hash table. That's a bit of a hack. It's nicer
if we can just do that directly. This requires taking a static
handler function rehash() from init.c, which invalidates one hash
table entry, and making it available to the builtin.
src/cmd/ksh93/bltins/typeset.c,
src/cmd/ksh93/include/builtins.h,
src/cmd/ksh93/include/nval.h,
src/cmd/ksh93/sh/init.c,
src/cmd/ksh93/sh/name.c:
- Merge b_hash() into b_alias().
- The -x option was still uselessly setting the NV_EXPORT flag.
Exported aliases were in ksh88 but were removed in ksh93.
- Rename rehash() handler function from init.c to nv_rehash
(avoiding a possible conflict with another rehash() in cd_pwd.c)
and move it to name.c just above nv_scan(), which it's meant to
be used with. Make it an extern so typeset.c can use it.
- b_alias(): Replace the PATH assignment by an nv_scan() call to
clear the hash table directly using the nv_rehash() handler.
src/cmd/ksh93/data/builtins.c:
- POSIX compliance fix: Remove BLT_SPC (special builtin) flag from
"alias" definition. 'alias' is specified as a regular builtin.
- sh_optalias[]: Fix uninformative -t option documentation.
- sh_opthash[]: Edit for conciseness and clarity.
src/cmd/ksh93/sh.1:
- Edit the 'alias -t' and 'hash' documentation.
- Remove the -- prefix from the 'alias' entry, which indicated that
it was supposed to be a declaration builtin like 'typeset', with
assignment-arguments expanding tildes and not being subject to
field splitting. However, my testing shows that 'alias' has never
actually behaved that way on ksh93. Even adding the BLT_DCL flag
in data/builtins.c doesn't seem to change that.
(cherry picked from commit afa68dca5c786daa13213973e8b0f9bf3a1dadf6)
