This is yet another approach to f3a5a3c449cf004aee662966e6c8dd2ebe122c0f
which got merged in over the weekend. It is much simpler than what we've
got now with almost exactly the same performance characteristics. A great
combination of maintability and performance.
```console
$ hyperfine --parameter-list branch master,speed-up-loading-cask-and-formula-file-names_v4 --warmup 5 --setup 'git switch {branch}' 'brew cat gimp'
Benchmark 1: brew cat gimp (branch = master)
Time (mean ± σ): 1.459 s ± 0.012 s [User: 0.892 s, System: 0.536 s]
Range (min … max): 1.448 s … 1.483 s 10 runs
Benchmark 2: brew cat gimp (branch = speed-up-loading-cask-and-formula-file-names_v4)
Time (mean ± σ): 1.456 s ± 0.009 s [User: 0.889 s, System: 0.536 s]
Range (min … max): 1.445 s … 1.468 s 10 runs
Summary
brew cat gimp (branch = speed-up-loading-cask-and-formula-file-names_v4) ran
1.00 ± 0.01 times faster than brew cat gimp (branch = master)
```
Currently, ruby files that are not in the Cask directory are
considered to be formulae if a Formula or HomebrewFormula directory
doesn't exist which doesn't make sense. We know that these should
only be in a few directories so we can check for that explicitly.
Beyond that the `Tap#cask_file?` and `Tap.formula_file?` methods
were only used inside update-report so it doesn't make sense to
turn them into pathnames and expand things when we know that
each string will be a relative path from a tap that we can just
check with a regex.
This change will stop other tap changes like new commands or
changes to other directories like lib/ from showing up as new formulae.
I tried opening a PR for this a long time ago but I got busy
with other things and it got closed by the stale bot.
- https://github.com/Homebrew/brew/pull/15489
`brew test-bot` can occasionally spend a long time doing nothing when
testing dependents.[^1] This is because it takes a long time to work out
that there is nothing to do.
To fix this, let's adjust `brew determine-test-runners` to pass on only
the list of formulae for which there is work to be done so that
`brew test-bot` doesn't need to waste time working this out.
[^1]: For example, it spends about 15 minutes doing nothing at https://github.com/Homebrew/homebrew-core/actions/runs/10500178069/job/29133091332?pr=180185
These are slow because of some Pathname related reasons so it's been
changed up to use `Dir.glob` instead and that makes a large difference.
```
$ hyperfine --parameter-list branch master,speed-up-loading-cask-and-formula-file-names,speed-up-loading-cask-and-formula-file-names_v2 --warmup 5 --setup 'git switch {branch}' 'brew cat gimp'
Benchmark 1: brew cat gimp (branch = master)
Time (mean ± σ): 2.150 s ± 0.014 s [User: 1.369 s, System: 0.745 s]
Range (min … max): 2.126 s … 2.174 s 10 runs
Benchmark 2: brew cat gimp (branch = speed-up-loading-cask-and-formula-file-names)
Time (mean ± σ): 1.814 s ± 0.012 s [User: 1.068 s, System: 0.711 s]
Range (min … max): 1.797 s … 1.840 s 10 runs
Benchmark 3: brew cat gimp (branch = speed-up-loading-cask-and-formula-file-names_v2)
Time (mean ± σ): 1.489 s ± 0.009 s [User: 0.905 s, System: 0.550 s]
Range (min … max): 1.478 s … 1.503 s 10 runs
Summary
brew cat gimp (branch = speed-up-loading-cask-and-formula-file-names_v2) ran
1.22 ± 0.01 times faster than brew cat gimp (branch = speed-up-loading-cask-and-formula-file-names)
1.44 ± 0.01 times faster than brew cat gimp (branch = master)
```
A local `ruby` build failed while building extensions, with:
*** Following extensions are not compiled:
-test-/file:
Could not be configured. It will not be installed.
/tmp/ruby-20240824-893184-ehcnsa/ruby-3.3.4/lib/mkmf.rb:480: The compiler failed to generate an executable file.
You have to install development tools first.
Check ext/-test-/file/mkmf.log for more details.
-test-/symbol:
Could not be configured. It will not be installed.
/tmp/ruby-20240824-893184-ehcnsa/ruby-3.3.4/lib/mkmf.rb:480: The compiler failed to generate an executable file.
You have to install development tools first.
Check ext/-test-/symbol/mkmf.log for more details.
[...]
`mkmf.log` indicated that the compiler shim failed to load `pathname`:
LD_LIBRARY_PATH=.:../../.. "gcc-13 -o conftest -I../../../.ext/include/aarch64-linux -I../../.././include -I../../.././ext/-test-/file -I/home/linuxbrew/.linuxbrew/opt/libyaml/include -I/home/linuxbrew/.linuxbrew/opt/openssl@3/include -O3 -fno-fast-math -ggdb3 -Wall -Wextra -Wdeprecated-declarations -Wdiv-by-zero -Wduplicated-cond -Wimplicit-function-declaration -Wimplicit-int -Wpointer-arith -Wwrite-strings -Wold-style-definition -Wimplicit-fallthrough=0 -Wmissing-noreturn -Wno-cast-function-type -Wno-constant-logical-operand -Wno-long-long -Wno-missing-field-initializers -Wno-overlength-strings -Wno-packed-bitfield-compat -Wno-parentheses-equality -Wno-self-assign -Wno-tautological-compare -Wno-unused-parameter -Wno-unused-value -Wsuggest-attribute=format -Wsuggest-attribute=noreturn -Wunused-variable -Wmisleading-indentation -Wundef -fPIC conftest.c -L. -L../../.. -L/home/linuxbrew/.linuxbrew/opt/libyaml/lib -Wl,-rpath,/home/linuxbrew/.linuxbrew/opt/libyaml/lib -L/home/linuxbrew/.linuxbrew/opt/openssl@3/lib -Wl,-rpath,/home/linuxbrew/.linuxbrew/opt/openssl@3/lib -L. -fstack-protector-strong -L/home/linuxbrew/.linuxbrew/opt/libyaml/lib -Wl,-rpath,/home/linuxbrew/.linuxbrew/opt/libyaml/lib -L/home/linuxbrew/.linuxbrew/opt/openssl@3/lib -Wl,-rpath,/home/linuxbrew/.linuxbrew/opt/openssl@3/lib -rdynamic -Wl,-export-dynamic -Wl,--no-as-needed -Wl,-rpath,/home/linuxbrew/.linuxbrew/Cellar/ruby/3.3.4/lib -L/home/linuxbrew/.linuxbrew/Cellar/ruby/3.3.4/lib -lruby-static -lz -lrt -lrt -ldl -lcrypt -lm -lpthread -lm -lpthread -lc"
/home/linuxbrew/.linuxbrew/Homebrew/Library/Homebrew/shims/linux/super/gcc-13:12:in `require': cannot load such file -- pathname (LoadError)
from /home/linuxbrew/.linuxbrew/Homebrew/Library/Homebrew/shims/linux/super/gcc-13:12:in `<main>'
I believe this was due to `../../..` in `LD_LIBRARY_PATH` containing
`libruby.so`, causing the Ruby script to load the being-built Ruby
library instead of the system/portable one.
This is most noticeable when you have a large local tap and it ends
up taking a nontrivial percentage of the overall run time for simple
commands like `brew info gimp`. The improvement here is to just
create paths later on in the loop which copies what we already do
for formulae.
See b1ddb05
```console
$ hyperfine --parameter-list branch master,speed-up-cask-files-by-name --warmup 5 --setup 'git switch {branch}' 'brew info gimp'
Benchmark 1: brew info gimp (branch = master)
Time (mean ± σ): 2.737 s ± 0.010 s [User: 1.958 s, System: 0.732 s]
Range (min … max): 2.720 s … 2.748 s 10 runs
Benchmark 2: brew info gimp (branch = speed-up-cask-files-by-name)
Time (mean ± σ): 2.597 s ± 0.017 s [User: 1.828 s, System: 0.724 s]
Range (min … max): 2.577 s … 2.624 s 10 runs
Summary
brew info gimp (branch = speed-up-cask-files-by-name) ran
1.05 ± 0.01 times faster than brew info gimp (branch = master)
```
This doesn't count PRs from forks, so it only works correctly for
maintainers (and only if they use non-fork branches).
Let's fix that, and:
- simplify the logic by using `paginate_graphql`
- return early if it's impossible for them to have too many open PRs
