[PATCH 10/10] submodule: don't use a subprocess to invoke "submodule--helper"

In a preceding commit we created "builtin/submodule.c" and faithfully tried to reproduce every aspect of "git-submodule.sh", including its invocation of "git submodule--helper" as a sub-process.

Let's do away with the sub-process and invoke "cmd_submodule__helper()" directly. Eventually we'll want to do away with "builtin/submodule--helper.c" altogether, but let's not do that for now to avoid conflicts with other in-flight topics. Even without those conflicts the resulting diff would be large. We can leave that for a later cleanup.

This speeds up invocations of all "git submodule" commands, E.g. a trivial "foreach" command on git.git is around 1.50 times faster[1]. For more expensive commands this'll make less of a difference, as the fixed cost of invoking the sub-process will be amortized away.

	$ git hyperfine -L rev HEAD~1,HEAD -s 'make CFLAGS=-O3' './git submodule foreach "echo \$name"'
	Benchmark 1: ./git submodule foreach "echo \$name"' in 'HEAD~1
	  Time (mean ± σ):       9.7 ms ±   0.1 ms    [User: 7.6 ms, System: 2.1 ms]
	  Range (min … max):     9.4 ms …  10.2 ms    285 runs

	Benchmark 2: ./git submodule foreach "echo \$name"' in 'HEAD
	  Time (mean ± σ):       6.6 ms ±   0.1 ms    [User: 5.1 ms, System: 1.5 ms]
	  Range (min … max):     6.2 ms …   7.2 ms    414 runs

	Summary
	  './git submodule foreach "echo \$name"' in 'HEAD' ran
	    1.48 ± 0.04 times faster than './git submodule foreach "echo \$name"' in 'HEAD~1'

It's also worth noting that some users were using e.g. "git submodule--helper list" directly for performance reasons[2]. With 31955475d1c (submodule--helper: remove unused "list" helper, 2022-09-01) released with v2.38.0 the "list" command was no longer provided. Users who had to switch to "git submodule--helper foreach" were given a command that (on my system) is around 6.5x slower.

Now the "foreach" is around 0.10x slower (due to the slight shell overhead), with 31955475d1c reverted on top of this:

	$ hyperfine './git submodule--helper list' './git submodule foreach --quiet "echo \$name"' --warmup 10
	Benchmark 1: ./git submodule--helper list
	  Time (mean ± σ):       6.4 ms ±   0.1 ms    [User: 5.0 ms, System: 1.5 ms]
	  Range (min … max):     6.2 ms …   7.2 ms    427 runs

	Benchmark 2: ./git submodule foreach --quiet "echo \$name"
	  Time (mean ± σ):       7.0 ms ±   0.1 ms    [User: 4.8 ms, System: 2.3 ms]
	  Range (min … max):     6.8 ms …   7.4 ms    390 runs

	Summary
	  './git submodule--helper list' ran
	    1.10 ± 0.03 times faster than './git submodule foreach --quiet "echo \$name"'

I think it would make sense to implement a "--format" option for "git submodule foreach" to help anyone who cares about that remaining performance (and to improve the API, e.g. by supporting "-z"), but as far as performance goes this makes the runtime acceptable again.

The pattern in "cmd_submodule_builtin()" of saving "struct strvec" arguments to a "struct string_list" and free()-ing them after the "argv" has been modified by "cmd_submodule__helper()" is new, without it we'd get various already-passing tests failing under SANITIZE=leak.

1. Using the "git hyperfine" wrapper for "hyperfine":
   https://lore.kernel.org/git/211201.86r1aw9gbd.gmgdl@evledraar.gmail.com/
2. https://lore.kernel.org/git/87czatrpyb.fsf@bernoul.li/

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
---
 builtin/submodule.c | 41 +++++++++++++++++++++++++++++++++--------
 1 file changed, 33 insertions(+), 8 deletions(-)