Pádraig Brady <P@draigBrady.com> writes:
> On 01/12/2025 22:14, Bruno Haible via GNU coreutils General Discussion wrote:
>> Hi,
>> On Ubuntu 24.04, the unit test that was added in commit
>> d0a51c614def6cf4f39fafe8ca7dcef925c4a503 fails when clang with ASAN and UBSAN
>> is used: The file timeout.status is expected to contain the number 125,
>> but it contains the number 124.
>> How to reproduce:
>> 1. Set the environment variables
>> CC="$CLANG_INST_DIR/bin/clang -Wl,-rpath,$CLANG_INST_DIR/lib"
>> CC="$CC
>> -fsanitize=address,undefined,signed-integer-overflow,shift,integer-divide-by-zero
>> -fno-sanitize-recover=undefined"; export CC
>> CFLAGS="-O0 -fno-omit-frame-pointer -ggdb"; export CFLAGS
>> ASAN_OPTIONS="detect_leaks=0 abort_on_error=1
>> allocator_may_return_null=1"
>> export ASAN_OPTIONS
>> 2. Configure and build coreutils.
>> 3.
>> $ cd src
>> $ { ./timeout -v .1 sleep 10 2>&1; echo $? >timeout.status; } | :
>> $ cat timeout.status
>> 124
>
>
> Ah right, the close_out() atexit handlerdoesn't call fclose(stderr) if
> SANITIZE_ADDRESS is defined,
> "as sanitizers may report to stderr after this function returns".
>
> Otherwise timeout(1) would have noticed the EPIPE error
> trying to write to stderr, and exit with 125 rather than 124.
>
> The attached updates timeout(1) to exit with status 124 in either case,
> as it would be generally bad for 125 to hide the fact that we
> have actually timed out the command.
>
> I'll apply the attached a little later.
Ah, I saw that standard error wasn't closed when ASAN was enabled. But
for some reason I didn't think that close_stdout() would behave
differently when ASAN was enabled.
Exiting with 124 seems more correct in this case, since we want to know
the program timed out.
Thanks,
Collin