Ignore load path caches generated by a different ruby version

[casperisfine]

Fix: #384

Until now it was assume that if the $LOAD_PATH was identical, then the content of the paths would be too.

However from one minor ruby version to another, the layout of the stdlib can change. So if the newer version is installed in the same place than the previous one, it might cause the load path cache to be invalid.

So we store RUBY_DESCRIPTION in the cache and make sure it matches before reusing the cache.

cc @eregon @noahgibbs if you mind giving this a quick look.

[eregon]

Should it warn if the load path cache can't be used due to different RUBY_DESCRIPTION?
Maybe only in $VERBOSE mode or so.

Having the cache per RUBY_REVISION might be nicer and more in line with the compile cache behavior, not sure how much work that is though 😅

In any case, RUBY_DESCRIPTION seems problematic. It does typically include a short git hash of RUBY_REVISION, so it's precise enough but it's too precise, e.g., using --yjit adds +YJIT in RUBY_DESCRIPTION.
So RUBY_REVISION seems better.

[casperisfine]

Should it warn if the load path cache can't be used due to different RUBY_DESCRIPTION?

I don't think that's particularly useful. Bootsnap regularly expect caches to be invalidated.

e.g., using --yjit adds +YJIT in

That's a good point. I'll use the same key than the compile cache.

[casperisfine]

Hum, I just remembered that what I liked too in RUBY_DESCRIPTION is that it contains the platform. E.g. if you try to use the same cache on your host and some kind of linux container and that somehow the ruby install path is the same (very unlikely, but might as well handle it). But I can use RUBY_REVISION + RUBY_PLATFORM

[casperisfine]

Done. @eregon if you want to give it another look.

[eregon]

Looks good, thanks for the fix.

[eregon]

Couple review questions:

• I don't see where the version is saved in the cache, just where it's read
• Is it OK to not use an existing cache from a previous Bootsnap version? Will it be overridden at the end of the run or so?

[casperisfine]

• I don't see where the version is saved in the cache, just where it's read

If the cache is missing or the version doesn't match, we initialize it with the version key. This cache is never really mutated, it's kind of generated all at once.

Is it OK to not use an existing cache from a previous Bootsnap version? Will it be overridden at the end of the run or so?

Yes it's fine. we could add the bootsnap version in there, but so far the cache format hasn't changed. In theory you never need to purge the bootsnap cache, bootsnap takes care of doing it itself.