Jeff King <peff@xxxxxxxx> writes: > On Wed, Oct 04, 2023 at 09:19:40AM -0500, Eric W. Biederman wrote: > >> What I see on a random failure looks like: >> >> > make -C t/ all >> > make[1]: Entering directory '/home/user/projects/git/git/t' >> > rm -f -r 'test-results' >> > GIT_TEST_EXT_CHAIN_LINT=0 && export GIT_TEST_EXT_CHAIN_LINT && make aggregate-results-and-cleanup >> > make[2]: Entering directory '/home/user/projects/git/git/t' >> > *** t0000-basic.sh *** >> > Segmentation fault >> > error: test_bool_env requires bool values both for $GIT_TEST_PASSING_SANITIZE_LEAK and for the default fallback >> >> Which doesn't sound like anything you have described so I am guessing it >> is something with my environment I need to track down. > > No, that seems different entirely. You'll have to figure out which > program is segfaulting and why (if you can see it in a script besides > t0000 you're probably better off, as that one is a maze of > tests-within-tests, since it is testing the test-harness itself). > > Although the "error" you see maybe implies that it is failing early on > in test-lib.sh, when we are calling "test-tool env-helper". If that is > segfaulting there is probably something very wrong with your build. Just to document what I am seeing it appears to be some odd interaction with address space randomization. If I run my make as: "setarch --addr-no-randomize make test" I don't see coredumps any more. Now to dig a deeper and see if I can figure out what about address space randomization is making things break. Eric
