"Raghul Nanth A via GitGitGadget" <gitgitgadget@xxxxxxxxx> writes:
> From: Raghul Nanth A <nanth.raghul@xxxxxxxxx>
>
> Add usage and performance tests for describe
>
> Performance metrics
> ...
The description is a bit skimpy. At least it should explain why
blindly flipping the "requires-full-index" bit off is all that is
necessary. I think in the review discussion on v1, Derrick gave some
explanation you can regurgitate and reuse.
> Signed-off-by: Raghul Nanth A <nanth.raghul@xxxxxxxxx>
> ...
> diff --git a/t/t6121-describe-sparse.sh b/t/t6121-describe-sparse.sh
> new file mode 100755
> index 00000000000..ce53603c387
> --- /dev/null
> +++ b/t/t6121-describe-sparse.sh
> @@ -0,0 +1,67 @@
> +#!/bin/sh
> +
> +test_description='git describe in sparse checked out trees'
> +
> +GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME=main
> +export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME
> +
> +. ./test-lib.sh
> +
> +check_describe () {
> + indir= &&
> +...
> + '
> +}
Having this almost identical helper copied from a near-by test
script means maintenance nightmare. People will forget to side port
to this copy any fixes they make to the other one.
I was hoping there would be a cleaner approach to reuse t6120, by
doing something similar to either how t8001-annotate shares blame
tests, or how t5559 takes advantage of t5551. One way might be ...
* prepare a prerequisite like so near the beginning of t6120
test_lazy_prereq WITH_SPARSE_INDEX '
test "$TEST_NAME" = t6121-describe
'
* add tests to be run with sparse-index enabled, but guarded with
some variable, e.g.
test_expect_success TESTING_SPARSE_INDEX 'a new test' '
... do sparse-index testing specific code ...
'
Such "sparse-index testing specific" code may include turning the
working tree the previous test already prepared into sparse
(i.e. additional "setup"), or running commands under
"ensure_not_expanded" (i.e. new tests).
* create t6121 that works similar to how t5559 takes advantage of
t5551, something like
#/bin/sh
. ./t6120-describe.sh
Ideally we should be able to do this without adding a new t6121, by
adding new tests that are specific to sparse-index at the end of
t6120, and avoid duplicated code.
Another possibility that may take the least amount of effort (but
may give us a lot less satisfactory outcome) may be to add a
lib-describe.sh library that is sourced from t6120 and t6121, and
move check_describe there. If check_describe has to behave slightly
differently, have a new conditional in the implementation so that
the caller can make a choice.
