Update comsub-with-alias anti-leak hack (re: fe20311f)

In the 93v- beta, they add a newline instead of a space.
This has fewer side effects as final newlines get stripped.
It's still a hack and it would still be nice to have a real fix,
but it seems even the AT&T guys couldn't come up with one.

src/cmd/ksh93/sh/macro.c:
- To somehow avoid a memory leak involving alias substitution,
  append a linefeed instead of a space to the comsub buffer.

src/cmd/ksh93/tests/subshell.sh:
- Add test for minor regression caused by the RedHat version.

src/cmd/ksh93/sh/macro.c

@@ -2118,7 +2118,7 @@ static void comsubst(Mac_t *mp,register Shnode_t* t, int type)
 			}
 			sfputc(stkp,c);
 		}
-		sfputc(stkp,' '); /* rhbz#982142: a=`some_alias` leaked memory, a=`some_alias ` did not! TODO: non-hack fix */
+		sfputc(stkp,'\n');  /* a=`some_alias` leaked memory, a=`some_alias<LF>` did not! TODO: non-hack fix */
 		c = stktell(stkp);
 		str=stkfreeze(stkp,1);
 		/* disable verbose and don't save in history file */

src/cmd/ksh93/tests/subshell.sh

@@ -950,5 +950,15 @@ v=main
 (v=sub; (d=${ v=shared; }; [[ $v == shared ]]) ) || err_exit "shared comsub in nested subshell wrongly scoped (2)"
 [[ $v == main ]] || err_exit "shared comsub leaks out of subshell (7)"
 
+# ======
+# After the memory leak patch for rhbz#982142, this minor regression was introduced:
+# if a backtick command substitution expanded an alias, an extra space was inserted.
+exp='word'
+alias a='print -n wo\'
+got=$(eval 'echo "`a`rd"')
+unalias a
+[[ $got == "$exp" ]] || err_exit 'backtick comsub with alias:' \
+	"expected $(printf %q "$exp"), got $(printf %q "$got")"
+
 # ======
 exit $((Errors<125?Errors:125))