On 15/09/2025 15:47, Pádraig Brady wrote:
On 14/09/2025 15:43, Bruno Haible via GNU coreutils Bug Reports wrote:
Pádraig Brady wrote:
p.s. an ASAN build would be good for CI
The CI already includes an ASAN + UBSAN build: see
https://github.com/coreutils/ci-check/blob/master/.github/workflows/many-platforms.yml#L850
named "make check with sanitizers" in
https://github.com/coreutils/ci-check/actions/runs/17705498910
If you think that it should have failed in this situation,
please commit or suggest an improvement.
Oh right MSAN is required for UMR detection:
$ ./configure CC=clang
$ make -j8 AM_CFLAGS='-Ulint -fsanitize=memory'
$ src/basenc --base32 -d /dev/null WARNING: MemorySanitizer:
use-of-uninitialized-value
#0 in decode_ctx_finalize /home/padraig/git/coreutils/src/basenc.c:358:14
#1 in do_decode /home/padraig/git/coreutils/src/basenc.c:1559:18
#2 in main /home/padraig/git/coreutils/src/basenc.c:1770:5
MSAN works in this case, but it doesn't look usable in general
as it requires all used libs to be built with MSAN enabled
or tracking is lost, resulting in false positives.
I.e. any calls to gettext, regex, gmp, ... will induce false positives.
So for UMR checking with the coreutils test suite,
coreutils/README-valgrind seems like the best solution currently.
Note one can perturb the stack with gcc and clang at least
with the -ftrivial-auto-var-init=pattern option.
I've confirmed that it triggers a segfault on my Linux system
with this code issue.
It's lightweight, so I've added it in my default dev environment,
but it might be worth enabling on CI systems too?
My defaut dev environment has these now:
export CFLAGS='-O2 -g -ftrivial-auto-var-init=pattern'
export MALLOC_CHECK_=3
# The following can cause performance issues
# so ensure unset before performance testing for example
# export MALLOC_PERTURB_=$(($RANDOM % 255 + 1))
cheers,
Padraig
