compond_variable assignment #889
Conversation
compond_variable is not assigned properly
|
I was able to reproduce the bug on all current ksh93 releases. It appears to be a regression introduced in ksh93u- 2010-08-11. This PR has some problems though: - The PR doesn't compile because of 93u+m's changes to - The PR does not include a regression test. These problems can be fixed with a patch: diff --git a/src/cmd/ksh93/sh/name.c b/src/cmd/ksh93/sh/name.c
index b35624a62..52185d717 100644
--- a/src/cmd/ksh93/sh/name.c
+++ b/src/cmd/ksh93/sh/name.c
@@ -512,7 +512,7 @@ void nv_setlist(struct argnod *arg,int flags, Namval_t *typ)
if(!nv_isnull(np) && np->nvalue!=Empty && !nv_isvtree(np))
sub=1;
}
- else if(((np->nvalue && np->nvalue!=Empty)||nv_isvtree(np)|| nv_arrayptr(np)) && !nv_type(np))
+ else if(np->nvalue && np->nvalue!=Empty && !nv_type(np))
{
int was_assoc_array = ap && ap->fun;
nv_unset(np,NV_EXPORT); /* this can free ap */
diff --git a/src/cmd/ksh93/tests/comvar.sh b/src/cmd/ksh93/tests/comvar.sh
index 27bd77f4b..2d5aca417 100755
--- a/src/cmd/ksh93/tests/comvar.sh
+++ b/src/cmd/ksh93/tests/comvar.sh
@@ -733,5 +733,22 @@ exp=$'(\n\ttypeset -C -a c\n)'
[[ $got == "$exp" ]] || err_exit 'setting compound array c.c=() does not preserve -C attribute' \
"(expected $(printf %q "$exp"), got $(printf %q "$got"))"
+# ======
+# github.com/ksh93/ksh/pull/889
+unset complex real img img1 real1
+complex=( real=1 img=2 )
+complex=( real1=3 img1=4 )
+real=$(echo "${complex.real}")
+img=$(echo "${complex.img}")
+real1=$(echo "${complex.real1}")
+img1=$(echo "${complex.img1}")
+exp='real=1
+img=2
+real1=3
+img1=4'
+got=$(typeset -p real img real1 img1)
+[[ $exp == "$got" ]] || err_exit 'previous fields of complex compound variable are lost when adding additional fields' \
+ "(expected $(printf %q "$exp"), got $(printf %q "$got"))"
+
# ======
exit $((Errors<125?Errors:125))
After patching the pull request to make it compile, another problem is evident; the bugfix causes some regression test failures: # Run the regression tests with './bin/shtests' at the root of the repository
test arrays begins at 2025-07-07+14:55:56
arrays.sh[606]: FAIL: setting element 1 of array to compound variable failed (expected 'typeset -a foo=((11 22) (x=3))', got 'typeset -a foo=((11 22) (44 55) )')
test arrays failed at 2025-07-07+14:55:56 with exit code 1 [ 186 tests 1 error ]
test comvar begins at 2025-07-07+14:56:03
comvar.sh[659]: FAIL: assignment of compound variable to compound array element not working
test comvar failed at 2025-07-07+14:56:03 with exit code 1 [ 105 tests 1 error ]
test subshell begins at 2025-07-07+14:56:43
subshell.sh[100]: FAIL: compound variable changes after unset leaves
test subshell failed at 2025-07-07+14:56:54 with exit code 1 [ 154 tests 1 error ]
Total errors: 3The regression this PR attempts to fix was introduced via the following vague and nondescript patch to name.c in 93u- 2010-08-11: --- a/src/cmd/ksh93/sh/name.c
+++ b/src/cmd/ksh93/sh/name.c
@@ -488,7 +488,7 @@ void nv_setlist(register struct argnod *arg,register int flags, Namval_t *typ)
if(!nv_isnull(np) && np->nvalue.cp!=Empty && !nv_isvtree(np))
sub=1;
}
- else if(np->nvalue.cp && np->nvalue.cp!=Empty && !nv_type(np))
+ else if(((np->nvalue.cp && np->nvalue.cp!=Empty)||nv_isvtree(np)) && !nv_type(np))
_nv_unset(np,NV_EXPORT);
}
else
This pull request blindly reverts the 2010-08-11 change without trying to ascertain the code's intent or purpose, and consequently introduces regressions that manifest as three regression test failures. The bugfix will need some work. |
|
Thanks for your comments and suggestions! |
|
@JohnoKing Hello !! |
|
You can use an older version of the Copy it to the So, e.g.: $ cd path/to/ksh93u+m
$ git checkout f8114823 -- bin/shtests
$ cd path/to/93u+
$ cp path/to/ksh93u+m/bin/shtests ./bin
$ KSH=$(echo $PWD/arch/*/bin/ksh) bin/shtests |
|
Thank You @JohnoKing for your suggestion on testing. ksh --versionversion sh (AT&T Research) 93u+m/1.0.10 2024-08-01 AIX ksh93 https://github.com/ksh93/ksh/blob/dev/src/cmd/ksh93/tests/heredoc.sh |
|
@JohnoKing Can you please guide me out here for the above issues to understand failures in test suite. |
|
Are you running the RHEL ksh package against the latest git commit for the regression tests? The v1.0.10 release is aging, having been released in August of last year. All of these problems (except that histexpand issue) can only be reproduced when running v1.0.10 against the tests from the latest commit. In such a case failures are expected because the regression tests on the dev branch test for the presence of bugs that were fixed after the v1.0.10 release. |
|
Thank You JohnKing!! |
Adding fix on behalf on IBM AIX
This is related assignment of compound_variables.
Below is the example test case to assign compond_variable
using set-x to display the values.
set -x
complex=( real=1 img=2 )
complex=( real1=3 img1=4 )
RC=$?
real=
echo "${complex.real}"img=
echo "${complex.img}"real1=
echo "${complex.real1}"img1=
echo "${complex.img1}"The goal of the test was to ensure all parts of the complex compound variable were preserved after setting additional fields. The fix ensures that the variable isn't overwritten and that all expected values (1, 2, 3, 4) are correctly retained and verified.