Johannes Berg <johannes@xxxxxxxxxxxxxxxx> writes: > On Mon, 2021-08-23 at 09:19 -0400, Derrick Stolee wrote: >> >> We just add subshells this way: >> >> test_expect_success 'test name' ' >> prep_step && >> ( >> # now in a subshell >> cd wherever && >> do things >> # don't need to cd again >> ) && >> continue test >> ' > > Sure. I know how to do subshells :) > > My point was that inside the subshell you cannot do test_path_is_file > and similar, because the subshell didn't import the libs. Everything the call to prep_step, and anything that came before that call, did to the environment, like setting shell functions and variables, is visible inside the ( ... subshell ... ). > I was specifically asking Junio ;-) > > But realistically, if this is the requirement you want to impose, then > you _cannot_ test for the segfault within git's test suite. Your loss. With that "cannot", I think you are assuming too much. Because Git is a fairly long-lived project, we've had our share of cases where we needed to test for bugs that happen only when outside a repository. And we have facility for just that, it's called "nongit" test helper that comes from test-lib-functions.sh, which is dot-included already so your subshells get it for free. t5300-pack-object.sh, for example, wants to make sure that the "git index-pack --stdin" command invoked in a directory that is not a repository, but "git index-pack <packfile>" works outside a repository, and has two tests that uses the nongit helper. Thanks.