Reuse read interfaces across compiler requests in OCaml rewatch

[cknitt]

With in-process compilation, OCaml rewatch compiles many modules on each worker domain, but every module read again each interface it uses. For the large Stdlib interfaces (Stdlib_Array, Stdlib_String, …), reading and unmarshalling them is a sizable part of compiling a small module.

Bs_cmi_load can now keep the interfaces it has read, per domain. Rewatch turns this on; the standalone bsc doesn't.

• Per domain, not shared. Copying a signature (Subst) temporarily marks the source types in place, so domains sharing one cached value crash. Within a domain, each copy restores its marks before the next request.
• Validated by file metadata. An entry is used only while the file's device, inode, size and modification time (from fstat on the descriptor being read) are unchanged.
• Recently modified files are always read. Published .cmi files are rewritten in place, so on a file system with a coarse clock, a same-size rewrite within one tick would keep all of that metadata. Like Git with racily clean files, an interface modified in the last two seconds is never cached or reused.
• Bounded. A domain keeps at most 16 MB of interface files (the runtime's are about 1.5 MB) and starts over past that, so long watch sessions don't grow without bound.

Outputs (JS and everything in lib/) are byte-identical to master at 1 and 8 domains. Peak memory is about the same or lower.

Clean builds on a Linux VM:

project	domains	master	this PR
tests/tests (without string_unicode_test)	1	1.27 s	1.20 s
	8	0.373 s	0.356 s
synthetic, 1501 modules using Stdlib	1	3.87 s	3.17 s
	8	0.913 s	0.696 s

The synthetic project is closer to a real application, whose modules use the standard library throughout.

New test: cmi_read_cache in ounit_compiler_request_tests.ml checks that a just-written interface isn't kept, that an older one is, and that a same-size rewrite with a different modification time is read again.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 95.00000% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.03%. Comparing base (dcc7e35) to head (77df19a).

Files with missing lines	Patch %	Lines
compiler/core/bs_cmi_load.ml	90.47%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #8874      +/-   ##
==========================================
+ Coverage   81.02%   81.03%   +0.01%     
==========================================
  Files         453      453              
  Lines       63030    63069      +39     
==========================================
+ Hits        51070    51108      +38     
- Misses      11960    11961       +1     

Files with missing lines	Coverage Δ
tests/ounit_tests/ounit_compiler_request_tests.ml	97.72% <100.00%> (+0.04%)	⬆️
compiler/core/bs_cmi_load.ml	84.21% <90.47%> (+6.43%)	⬆️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[pkg-pr-new]

Open in StackBlitz

rescript

npm i https://pkg.pr.new/rescript@8874

@rescript/belt

npm i https://pkg.pr.new/@rescript/belt@8874

@rescript/darwin-arm64

npm i https://pkg.pr.new/@rescript/darwin-arm64@8874

@rescript/darwin-x64

npm i https://pkg.pr.new/@rescript/darwin-x64@8874

@rescript/linux-arm64

npm i https://pkg.pr.new/@rescript/linux-arm64@8874

@rescript/linux-x64

npm i https://pkg.pr.new/@rescript/linux-x64@8874

@rescript/runtime

npm i https://pkg.pr.new/@rescript/runtime@8874

@rescript/win32-x64

npm i https://pkg.pr.new/@rescript/win32-x64@8874

commit: 77df19a

[github-actions]

Developer playground preview: https://rescript-lang.github.io/rescript/dev-playground/?version=pr-8874