[bug-diffutils] bug#60457: bug#60457: Test failure (color) with UBSAN

[Jim Meyering]

On Sat, Dec 31, 2022 at 1:55 PM Sam James <sam@gentoo.org> wrote:
> Hi folks,
>
> I'm testing out the latest release previews from platform-testers in Gentoo 
> and came across the following.
>
> diffutils' tests seem to fail (specifically the 'color' test) when built with 
> UBSAN (-fsanitize=undefined):
> ```
> [...]
> + mkfifo fifo
> + printf %1000000s-a
> + printf %1000000s-b
> + + diff --color=always ---presume-output-tty a bhead
>  -c 10
> util.c:205:3: runtime error: execution reached an unreachable program point
>     #0 0x558fa0c7ee3c in xsigismember 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/util.c:205
>     #1 0x558fa0c7f4e5 in install_signal_handlers 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/util.c:372
>     #2 0x558fa0c80f05 in check_color_output 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/util.c:822
>     #3 0x558fa0c81976 in begin_output 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/util.c:1017
>     #4 0x558fa0c7c63d in print_normal_hunk 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/normal.c:49
>     #5 0x558fa0c82cd1 in print_script 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/util.c:1303
>     #6 0x558fa0c7c5e5 in print_normal_script 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/normal.c:31
>     #7 0x558fa0c640cc in diff_2_files 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/analyze.c:665
>     #8 0x558fa0c6f28c in compare_files 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/diff.c:1488
>     #9 0x558fa0c69e11 in main 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/src/diff.c:851
>     #10 0x7fb6ab79664f in __libc_start_call_main 
> ../sysdeps/nptl/libc_start_call_main.h:58
>     #11 0x7fb6ab796708 in __libc_start_main_impl ../csu/libc-start.c:381
>     #12 0x558fa0c5a144 in _start ../sysdeps/x86_64/start.S:115
>
> + test 1 = 141
> + fail=1
> + Exit 1
> + set +e
> + exit 1
> + exit 1
> + remove_tmp_
> + __st=1
> + cleanup_
> + :
> + test  = yes
> + cd 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/tests
> + chmod -R u+rwx 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/tests/gt-colors.1cSV
> + rm -rf 
> /var/tmp/portage/sys-apps/diffutils-3.8_p20221230/work/diffutils-3.8.43-2d50/tests/gt-colors.1cSV
> + exit 1
> FAIL colors (exit status: 1)
>
> ============================================================================
> Testsuite summary for GNU diffutils 3.8.43-2d50
> ============================================================================
> # TOTAL: 23
> # PASS:  21
> # SKIP:  0
> # XFAIL: 1
> # FAIL:  1
> # XPASS: 0
> # ERROR: 0
> ============================================================================
> See tests/test-suite.log
> Please report to bug-diffutils@gnu.org
> ============================================================================
> ```
>
> This is with GCC 12.2.1_p20221224.
>
> I've made the full log available at 
> https://dev.gentoo.org/~sam/bugs/diffutils/ubsan-color-test/build.log
> and 
> https://dev.gentoo.org/~sam/bugs/diffutils/ubsan-color-test/test-suite.log.

Hi Sam, thanks for the testing and the report.
Can you still reproduce that?

So far, I am unable to, at least at this commit: v3.8-49-gcc41598.
I started with an already-configured working directory,
ran "make clean", and then this:

  ASAN_OPTIONS=detect_leaks=0 make CFLAGS='-O -ggdb3'
AM_CFLAGS=-fsanitize=undefined AM_LDFLAGS='-fsanitize=undefined
-static-libubsan' check TESTSUITEFLAGS=--jobs=55

and all tests passed.