On 8/6/24 06:13, Eric Sunshine wrote:
On Mon, Aug 5, 2024 at 8:00 PM AbdAlRahman Gad <abdobngad@xxxxxxxxx> wrote:
Split "test-tool ... | sed" pipeline into two commands to avoid losing
exit status from test-tool.
Signed-off-by: AbdAlRahman Gad <abdobngad@xxxxxxxxx>
---
diff --git a/t/t7004-tag.sh b/t/t7004-tag.sh
@@ -97,7 +97,8 @@ test_expect_success 'creating a tag with --create-reflog should create reflog' '
- test-tool ref-store main for-each-reflog-ent refs/tags/tag_with_reflog1 | sed -e "s/^.* //" >actual &&
+ test-tool ref-store main for-each-reflog-ent refs/tags/tag_with_reflog1 >actual.body &&
+ sed -e "s/^.* //" actual.body >actual &&
It's not just `test_tool` we care about; we also (importantly) don't
want to see `git` itself upstream of a pipe, and there are many such
instances remaining in this script. Here are some common examples:
test $(git tag -l | wc -l) -eq 0 &&
git cat-file tag "$1" | sed -e "/BEGIN PGP/q"
git tag -l | grep "^tag-one-line" >actual &&
forged=$(git cat-file tag ... | sed -e ... | git mktag) &&
git tag -l --no-sort "foo*" | sort >actual &&
By the way, these days, rather than:
test $(git tag -l | wc -l) -eq 0 &&
we would say:
test_stdout_line_count = 0 git tag -l &&
which nicely avoids placing `git` upstream of a pipe.
Thanks for the review. Could this be done in a separate patch series? As
modifying a patch in the beginning of a patch series requires lots of
time to adapt the rest of the series to the change.
