patchperf: do allow `GIT_PERF_*` to be overridden again

From: Johannes Schindelin <johannes.schindelin@gmx.de>

A common way to run Git's performance benchmarks on repositories other than Git's own repository (which is not exactly large when compared to actually large repositories) is to run them like this:

	GIT_PERF_LARGE_REPO=/path/to/my/large/repo \
	./p1234-*.sh -ivx

Contrary to developers' common expectations, this failed to work when Git was built with a different `GIT_PERF_LARGE_REPO` value specified at build time: That build-time option would have been written to the `GIT-BUILD-OPTIONS` file, which in turn would have been sourced by `test-lib.sh`, which in turn would have been sourced by `perf-lib.sh`, which in turn would have been sourced by the perf test script, _overriding_ the environment variable specified in the way illustrated above.

Since perf tests are not run as part of the build, this most likely unintended behavior was not caught and certainly not fixed, as the `GIT_PERF_*` values would have been empty at build-time.

However, in 4638e8806e3a (Makefile: use common template for GIT-BUILD-OPTIONS, 2024-12-06), a subtle change of behavior was introduced: Whereas before, a couple of build-time options (the `GIT_PERF_*` ones included) were written to `GIT-BUILD-OPTIONS` only when their values were non-empty. With this commit, they are also written when they are empty.

The consequence is that above-mentioned way to run the perf tests will not only fail to pick up the desired `GIT_PERF_*` settings when they were specified differently while building Git, instead the desired settings will be only respected when specified _while building_ Git.

Let's work around the original issue, i.e. let `GIT_PERF_*` environment variables override what is recorded in `GIT-BUILD-OPTIONS`.

Note that this is just the tip of the iceberg, there are a couple of `GIT_TEST_*` options that may want a similar fix in `test-lib.sh`. Due to time constraints on my side, this here patch focuses exclusively on the `GIT_PERF_*` settings.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
---
    perf: do allow GIT_PERF_* to be overridden again
    
    This issue was noticed when working on large-scale issues.

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-1900%2Fdscho%2Fsupport-ad-hoc-git-perf-settings-again-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-1900/dscho/support-ad-hoc-git-perf-settings-again-v1
Pull-Request: https://github.com/gitgitgadget/git/pull/1900

 t/perf/perf-lib.sh | 12 ++++++++++++
 1 file changed, 12 insertions(+)

On 4/4/2025 6:56 AM, Johannes Schindelin via GitGitGadget wrote:

This issue also extends to other necessary variables such as GIT_PERF_REPEAT_COUNT.

I verified this fix in my local environment. Thanks so much for digging in and finding the solution here!

-Stolee

On Fri, Apr 04, 2025 at 10:56:07AM +0000, Johannes Schindelin via GitGitGadget wrote:

It doesn't look like Junio picked this up, so I wanted to chime in that this regression bit me today, too (specifically for GIT_PERF_LARGE_REPO, but also another variable which I'll detail in a moment).

This is also the same issue that hit the interop suite discussed in:

  https://lore.kernel.org/git/Z8IX2bMJe+V80idE@nand.local/

(there I was a bit dismissive of it, because I think GIT_*_MAKE_OPTS would usually be set at build time, but for these other variables, they're almost always going to come from the environment).

I like this direction. It not only fixes the regression, but makes things generally behave more as I'd expect them to.

I think it doesn't quite fix everything, though. I noticed that GIT_PERF_REPEAT_COUNT no longer correctly defaults to "3" when using the "run" script. And there are two problems here:

  1. The "run" script itself sources GIT-BUILD-OPTIONS, so it would need
     similar treatment.

  2. But even if we did, that, in my case I am not setting
     PERF_REPEAT_COUNT at all. So there is no local env variable that
     we'd use to take precedence. The problem is in the way the "run"
     script assigns defaults. If it sees an environment variable set
     (whether actually from the environment or from GIT-BUILD-OPTIONS)
     it accepts it as-is, rather than installing its fallback. So it
     will never use its default of "3", and instead retain the blank
     string (which, by a stroke of luck, does still cause it to run each
     trial at least once).

So I think we either need to rewrite the "run" script's fallback code, or teach the GIT-BUILD-OPTIONS writer to avoid mentioning unset variables (which is the real source of the problem in 4638e8806e3a).

So...

...yes, this is definitely the tip of the iceberg. I don't mind doing this patch as an incremental step forward (and because it is an improvement in behavior even if 4638e8806e3a were reverted). But the issue is far from solved overall.

The implementation here looks correct to me, including the quoting. The number of backslashes is a bit vomit-inducing, but I think unavoidable (we could put the sed command into single-quotes, but then you'd have to escape out of it to show the single-quotes you do need to mention).

-Peff

Jeff King <peff@peff.net> writes:

This was lost in the cracks. Thanks for bringing it back to our attention. I think what happened was that I saw whack-a-mole aspect of the root cause, which makes this "the tip of the iceberg", and felt it was more sensible to wait before a real solution, like ...

... this was raised. And then I completely forgot about the topic, as nothing happened since then.

I do not mind it as an incremental band-aid.

Thanks, all.

On Sun, Apr 20, 2025 at 02:12:34PM -0700, Junio C Hamano wrote:

Looks like this is in 'jch' now, but there's a mis-merge in e8cf2b99cd (Merge branch 'ps/meson-build-perf-bench' into jch, 2025-04-21).

The original patch in this thread did something like:

  git_perf_settings=$(...pull GIT_PERF_* from env...)
   . ../test-lib.sh
  eval "$git_perf_settings"

That is, we stash away the environment, then load test-lib.sh, which overwrites the environment, and then we restore (some of) the original values.

In that merge, the test-lib.sh inclusion is moved (and in fact is now accompanied by an explicit inclusion of GIT-BUILD-OPTIONS), and we now have:

   . "$GIT_BUILD_DIR"/GIT-BUILD-OPTIONS
   . "$GIT_SOURCE_DIR"/t/test-lib.sh
   git_perf_settings=$(...)
   eval "$git_perf_settings"

Which of course does nothing. We need to set $git_perf_settings before those other source lines (and the eval must remain after them).

-Peff

Jeff King <peff@peff.net> writes:

Right.  Thanks.