Keep a running InstalledPackageIndex to skip expensive per-package ghc-pkg calls

[sheaf]

Building on the in-library refactor that landed in 6867dd5, this patch extracts computePackageInfoFromIndex from the Cabal computePackageInfo function, which allows cabal-install to use a running InstalledPackageIndex instead of having to repeatedly query ghc-pkg, once per package.

Several cabal-install functions now pass around a TVar InstalledPackageIndex which keeps track of the InstalledPackageIndex used for building a project. Each time a dependency gets registered, we update this TVar, and read from it to obtain an InstalledPackageIndex to pass to the Cabal configure function, skipping slow calls to ghc-pkg.

---

Template Α: This PR modifies behaviour or interface

• Patches conform to the coding conventions.
• Any changes that could be relevant to users have been recorded in the changelog.
• The documentation has been updated, if necessary.
• No manual QA notes necessary (no change to the command line interface).
• Testing approach: the existing tests continue to work even though we query ghc-pkg far less.

[sheaf]

This was a bit subtle to write, and getting it wrong means that we pass an incomplete or incorrect InstalledPackageIndex when configuring, which can lead to "Package not found" errors.

[ulysses4ever]

Since the title of this PR mentions replacing "expensive" ghc-pkg calls, I am curious if there is any fast, back-of-envelope timings of what sorts of gains one can expect from this change.

[sheaf]

Since the title of this PR mentions replacing "expensive" ghc-pkg calls, I am curious if there is any fast, back-of-envelope timings of what sorts of gains one can expect from this change.

I will get back to you with more precise measurements but for typical packages this can save over half of the time spent in configure. Taken with #11768 I am seeing typical configure times go from ~3s to ~0.5s.

[sheaf]

I have put up a spreadsheet here, compiling pandoc from scratch and using the --build-timings feature from #11769.

Base average time spent in configure: 2.9 seconds.
With this patch: it goes down to 1.7 seconds.
With #11768: it goes down to 1.2 seconds.
With both: it goes down to 0.15 seconds.

[Mikolaj]

Great job! Looks plausible to me.

[ulysses4ever]

The caching strategy is straightforward and looks good. The relation to some ghc-pkg calls is indirect (you can't see it in the diff), so it's unclear to me. I'd appreciate if it was clarified, to which end I leave a comment for the changelog below. Thanks!

[ulysses4ever]

"expensive ghc_pkg calls that we make when…"?

[philderbeast]

Could be done without result or return. Use of <&> instead of <$> would require another import.

Suggested change

	readCachedRegistration PackageFileMonitor{pkgFileMonitorReg} = do
	result <- checkFileMonitorChanged pkgFileMonitorReg "." ()
	return $ case result of
	MonitorUnchanged mIpi _ -> mIpi
	MonitorChanged _ -> Nothing
	readCachedRegistration PackageFileMonitor{pkgFileMonitorReg} =
	\case MonitorUnchanged mIpi _ -> mIpi; MonitorChanged _ -> Nothing
	<$> checkFileMonitorChanged pkgFileMonitorReg "." ()

[philderbeast]

Is this separator between imports and the rest of the module needed?