Some expected failures of git-http-backend leave running its children
(receive-pack or upload-pack) which still hold opened descriptors
to act.err and with some probability they live long enough to write
their failure messages after next test has already truncated
the files. This causes occasional failures of the test script.
Avoid the issue by unlinking the older files before writing to them.
Reported-by: Carlo Arenas <carenas@xxxxxxxxx>
Helped-by: Carlo Arenas <carenas@xxxxxxxxx>
Signed-off-by: Max Kirillov <max@xxxxxxxxxx>
---
Thanks for the analysis. I seem to have guessed the reason.
This patch should prevent it.
I think the tests should somehow make sure there are no such late
processes. I can see 2 options:
* somehow find out in the tests all children and wait for them. I have no idea how.
* make http-backend close handle to its child and wait for it to exit before dying.
This would not prevent childrenc in general, because http-backend may be killed,
but not in our expected failure cases
Actually, don't the children receive some SIGHUP? Maybe thy should. However, it
would still take some time for them to handle it, so it does not fully solve the issue
t/t5562-http-backend-content-length.sh | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/t/t5562-http-backend-content-length.sh b/t/t5562-http-backend-content-length.sh
index 90d890d02f..bb53f82c0c 100755
--- a/t/t5562-http-backend-content-length.sh
+++ b/t/t5562-http-backend-content-length.sh
@@ -25,6 +25,8 @@ test_http_env() {
handler_type="$1"
request_body="$2"
shift
+ (rm -f act.out || true) &&
+ (rm -f act.err || true) &&
env \
CONTENT_TYPE="application/x-git-$handler_type-pack-request" \
QUERY_STRING="/repo.git/git-$handler_type-pack" \
@@ -155,6 +157,8 @@ test_expect_success 'CONTENT_LENGTH overflow ssite_t' '
'
test_expect_success 'empty CONTENT_LENGTH' '
+ (rm -f act.out || true) &&
+ (rm -f act.err || true) &&
env \
QUERY_STRING="service=git-receive-pack" \
PATH_TRANSLATED="$PWD"/.git/info/refs \
--
2.19.0.1202.g68e1e8f04e
