Patch sysconfig data at install time

[charliermarsh]

Summary

This PR reimplements sysconfigpatcher in Rust and applies it to our Python installations at install-time, ensuring that the sysconfig data is more likely to be correct.

For now, we only rewrite prefixes (i.e., any path that starts with /install gets rewritten to the correct absolute path for the current machine).

Unlike sysconfigpatcher, this PR does not yet do any of the following:

• Patch pkginfo files.
• Change clang references to cc.

A few things that we should do as follow-ups, in my opinion:

1. Rewrite AR.
2. Remove -isysroot, which we already do for newer builds.

[charliermarsh]

I initially thought we could just parse this as JSON, but we only started to output sysconfig data as JSON in the most recent PBS release. So it won't apply to existing Pythons.

Instead, this is a small parser that just handles Python dictionaries with the assumption that keys are string and values are (strings or integers). Main complexity is that strings can be implicitly concatenated (and are in practice), but it's not too bad.

[zanieb]

Do we need to have @dhruvmanila review this? :)

[charliermarsh]

Yeah good idea. @dhruvmanila -- this is applied to a file like the following. We know that it's an object with string keys, and the values must be strings or integers (no f-strings, but they can be implicitly concatenated).

# system configuration generated and used by the sysconfig module
build_time_vars = {'ABIFLAGS': '',
 'AC_APPLE_UNIVERSAL_BUILD': 0,
 'AIX_BUILDDATE': 0,
 'AIX_GENUINE_CPLUSPLUS': 0,
 ...
 'XMLLIBSUBDIRS': 'xml xml/dom xml/etree xml/parsers xml/sax',
 'abs_builddir': '/private/var/folders/0w/4z5l9vds32nbkz7l22n8j6s80000gn/T/tmp5be2znu_/Python-3.10.16',
 'abs_srcdir': '/private/var/folders/0w/4z5l9vds32nbkz7l22n8j6s80000gn/T/tmp5be2znu_/Python-3.10.16',
 'datarootdir': '/install/share',
 'exec_prefix': '/install',
 'prefix': '/install',
 'srcdir': '.'}

[dhruvmanila]

The parser looks good. I think it just has a lot of lookaheads where we could just use bump but if this is not in a critical path (which it seems like so), then it shouldn't matter practically.

[charliermarsh]

This comes from Ruff (which ported it from rustc).

[charliermarsh]

\cc @bluss who I also chatted with on Discord

[dhruvmanila]

Do we need to always do the lookahead first and then bump the character? Typically, we avoid doing lookaheads and only do it when required. If we bump the character first, it always avoids bumping it for every branch in the loop.

This would look like:

loop {
	let Some(next) = cursor.bump() else {
		break;
	};

	match next {
		...
	}
}

For other parse methods, it would then need to accept the character that's bumped. For strings, you could then utilize the cursor.previous() method to add a debug assertion that the previous character was the same as the given quote character.

[charliermarsh]

I left it as-is but I'll play with changing it.

[dhruvmanila]

nit: we could utilize a match to avoid computing self.first for each branch

[konstin]

This looks a lot like https://github.com/typst/unscanny

[dhruvmanila]

That looks interesting!