Re: [GSoC][PATCH v4 0/5] t: port reftable/tree_test.c to the unit testing framework

Chandra Pratap <chandrapratap3519@gmail.com> writes:

Consistently refer to an existing commit with --format=reference.

	$ git show -s --format=reference 8bf6fbd
	8bf6fbd00d (Merge branch 'js/doc-unit-tests', 2023-12-09)

I doubt that the reason why "unit-tests" written entirely in C was introduced is because we wanted to simplify testing and to reduce test run times to begin with. The traditional tests to observe the effect visible to end-users by actually running commands that would be run by end-users and unit tests serve two separate purposes. The latter does not "simplify", or "reduce time"---you cannot write a test "entirely in C" to make sure, say, that "git push --force" allows a non-fast-forward update to happen using unit-test framework. The unit-tests cannot replace end-to-end tests. They are complementary.

The statement may need to be rethought.

Or just stop at saying something like

    The reftable library comes with self tests, which are exercised
    as part of the usual end-to-end tests that are designed to
    observe the end-user visible effects of Git commands.  What it
    exercises, however, is a better match to the unit-testing
    framework, merged at 8bf6fbd0 (Merge branch 'js/doc-unit-tests',
    2023-12-09), that are designed to observe how low level
    implementation details, at the level of sequences of individual
    function calls, behave.

which already covers the next paragraph while at it.

By the way, did you rebase the patches? On which commit is this iteration based? Judging from the second patch, it seems to assume that Makefile does not yet have t-reftable-record in it, and it applies cleanly to 'master' before 9118e46e (Merge branch 'cp/unit-test-reftable-record', 2024-07-15). Newer 'master' has textual conflicts (nothing that I cannot resolve, but it shows that apparently that anything newer than 9118e46e are not commits that you developed this series on).

Has this series even been compile-tested? I do not think that even the unit test added by this series has been run (or compiled).

$ make -j32 t/unit-tests/t-reftable-tree.o
GIT_VERSION = 2.46.0.rc0.27.g0c2075a7c5
    * new build flags
    CC t/unit-tests/t-reftable-tree.o
In file included from t/unit-tests/t-reftable-tree.c:9:
t/unit-tests/t-reftable-tree.c: In function ‘t_tree_search’:
t/unit-tests/t-reftable-tree.c:45:44: error: ‘test_compare’ undeclared (first use in this function); did you mean ‘t_compare’?
   45 |         check(!tree_search(values, &root, &test_compare, 0));
      |                                            ^~~~~~~~~~~~
t/unit-tests/test-lib.h:75:45: note: in definition of macro ‘check’
   75 |         check_bool_loc(TEST_LOCATION(), #x, x)
      |                                             ^
t/unit-tests/t-reftable-tree.c:45:44: note: each undeclared identifier is reported only once for each function it appears in
   45 |         check(!tree_search(values, &root, &test_compare, 0));
      |                                            ^~~~~~~~~~~~
t/unit-tests/test-lib.h:75:45: note: in definition of macro ‘check’
   75 |         check_bool_loc(TEST_LOCATION(), #x, x)
      |                                             ^
make: *** [Makefile:2754: t/unit-tests/t-reftable-tree.o] Error 1

Let's concentrate on quality, not quantity; too many topics with the same prefix cp/unit-test* seem to be left unreviewed on the list.

In the meantime, I'll queue the following fix-up on top. In this codebase, it is preferred to write a pointer to a function whose name is "func" as just "func", not "&func".

 t/unit-tests/t-reftable-tree.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)