Enable graph-based type checking, parallel optimizations and ILX gen in deterministic build

[majocha]

This as a follow up to #18998:

• adds tests for added dependency graph supported scenarios: script compilation mode and misordered .fs .fsi files
• updates naming and adds comments wrt to the above
• enables graph based type checking in deterministic builds
• enables parallel optimizations in deterministic builds
• enabled parallel ilx gen in deterministic builds

Necessary changes to enable determinism:

• sort things when writing PDB
• update NiceNameGenerator to produce stable names
• update generation of array literal types (multiple delayed gens would place then under "private implementation" concurrently, resulting in non-deterministic order)

There are no additional tests for determinism, it is currently tested only by running the test-determinism script locally and in the CI.

One issue needed investigation: type extensions with mismatched type argument names can break determinism:

// in one filetypeLexBuffer<'Char>// in another filetypeLexBuffer<'char>with

Either char or Char will end up in the IL.

This is related to #15287 and needs work: #19033

[github-actions]

❗ Release notes required

---

✅ Found changes and release notes in following paths:

Change path	Release notes path	Description
src/Compiler	docs/release-notes/.FSharp.Compiler.Service/11.0.0.md

[majocha]

Typar name not always picked up from the source (?)

But wth is going on here? Looks like a type parameter 'Char gets randomly converted to char.

[majocha]

Good thing is I got exactly the same diff locally as in the CI, so it is already deterministic in this regard 😁

[majocha]

Ok so it's not a type parameter but actual char type. It just randomly gets emitted either as Char or char.

Edit: Nope it was because of not unified type argument names between type declaration and extensions.
NOTE: It needs an actual separate fix.

[majocha]

Hallelujah.

Ilx gen is still disabled. Only optimization and graph checking works.

portable pdb test needs updating, too.

[majocha]

TODO:

fsharp/src/Compiler/CodeGen/IlxGen.fs

Line 2337 in 12efe3b

CompilerGeneratedName("T"+ string (newUnique ())+"_"+ string size +"Bytes")// Type names ending ...$T<unique>_37Bytes

and

fsharp/src/Compiler/CodeGen/IlxGen.fs

Line 2757 in 12efe3b

letilFieldName= CompilerGeneratedName("field"+ string (newUnique ()))

Apart from stabilizing these generated names, the order of of generated IL also needs to be deterministic. Specifically, in FCS.dll generated types in PrivateImplementationDetails are sometimes in different order.

Possibly relevant:

fsharp/src/Compiler/CodeGen/IlxGen.fs

Lines 2389 to 2394 in 12efe3b

	member_.GenerateRawDataValueType(cloc,size)=
	// Byte array literals require a ValueType of size the required number of bytes.
	// With fsi.exe, S.R.Emit TypeBuilder CreateType has restrictions when a ValueType VT is nested inside a type T, and T has a field of type VT.
	// To avoid this situation, these ValueTypes are generated under the private implementation rather than in the current cloc. [was bug 1532].
	letcloc= CompLocForPrivateImplementationDetails cloc
	rawDataValueTypeGenerator.Apply((cloc, size))

[T-Gro]

Waiting for #19062

[T-Gro]

One thing where I do not have high confidence in the automated test suite (==> will need more manual testing and dogfooding) is the debugger support via pdbs.