[BUG?] setting ulimit in test suite broken for me

Hi,

so I wanted to create a new test for large transactions, which should look like:

	run_with_limited_open_files () {
		(ulimit -n 64 && "$@")
	}

	test_lazy_prereq ULIMIT 'run_with_limited_open_files true'

	test_expect_success ULIMIT 'large transaction creating branches does not burst open file limit' '
	(
		echo $(ulimit -n)
		for i in $(seq 65)
		do
			echo "create refs/heads/$i HEAD"
		done >large_input &&
		git update-ref --stdin <large_input &&
		git rev-parse --verify -q refs/heads/65
	)
	'

Mind the "echo $(ulimit -n)" in there as a debugging output. So if I run the test with "-d -v" to actually see the debugging output, I see ulimit -n set to 32768 instead of the desired 64.

So I assumed I made a mistake and looked at other places with a similar pattern such as in t7004-tag.sh. Inserting a debug statement (as attached, so I don't need (to trust) -d and -v as parameters for the test) in there also doesn't report the desired stack size of 128, but rather the system default of 8192. Is that just my system or does anybody else also run into such a problem?

Running ulimit -{n,s} manually in a shell (/bin/sh) as well as bash does work as expected as it correctly reports with ulimit -a as well as breaking the tests I try to write.

Thanks, Stefan

Signed-off-by: Stefan Beller <sbeller@google.com>
---
 t/t7004-tag.sh | 1 +
 1 file changed, 1 insertion(+)