split TypeChecker.fs

[dsyme]

This is cleanup to split TypeChecker.fs into two chunks of 12,000 and 6,000 lines.

The chunks are

• CheckExprs.fs - check syntactic types and expressions, and many aspects of members and recursion (since object expressions include members)

• CheckDecls.fs - check type declarations, modules, namespaces, signatures, files

This necessarily involves opening up some of the internals of TypeChecker.fs in the signature of CheckExprs.fsi but I think that's ok. More is opened up than I would like (and we may be able to reduce that in the future) but it's not too bad.

We can do further splitting at a later point, e.g. split out CheckComputationExprs.fs, CheckObjectExprs.fs, TcFileState.fs, TcEnv.fs and so on (if that proves the right split)

[forki]

but what will we use as benchmark after this? :P

[dsyme]

but what will we use as benchmark after this? :P

LOL I was just adjusting those benchmarks and thinking "oh wow, this will speed them up"

[dsyme]

BTW it feels like a good time to do this split, as AFAIK there are relatively few outstanding changes to TypeChecker.fs in proposed features or PRs

[dsyme]

Hmmm transient failure on Mac

2020-10-22T13:17:40.0576150Z [xUnit.net 00:00:26.55]     FSharp.Core.UnitTests.Control.MailboxProcessorType.Scan handles cancellation token [FAIL]
2020-10-22T13:17:40.0607290Z   X FSharp.Core.UnitTests.Control.MailboxProcessorType.Scan handles cancellation token [5s 156ms]
2020-10-22T13:17:40.0608500Z   Error Message:
2020-10-22T13:17:40.0609440Z    System.NullReferenceException : Object reference not set to an instance of an object.
2020-10-22T13:17:40.0610320Z   Stack Trace:
2020-10-22T13:17:40.0611230Z      at FSharp.Core.UnitTests.Control.MailboxProcessorType.Scan handles cancellation token() in /Users/runner/work/1/s/tests/FSharp.Core.UnitTests/FSharp.Core/Microsoft.FSharp.Control/MailboxProcessorType.fs:line 168

[dsyme]

@cartermp I'm loving the fact that the compiler guide is in the tree and we can search and replace to help update it

[KevinRansom]

Thanks Don, this is good.

[cartermp]

My approval is one of happiness and sadness.

Happiness that this is finally getting split up into some logical chunks!

Sadness that, like @forki, I no longer have this ridiculous amazing benchmark for various tooling experience smoke tests and benchmarks.

[baronfel]

I'm just amazed that now GitHub will actually render the content of the files, so we can deep-link directly to relevant code when showing folks around.

[dsyme]

Yes it's the end of an era. But at least there are still 230 character lines in the files to enjoy.

[MecuSorin]

I suggest a small optimization here:

let CheckDuplicates (idf: _ -> Ident) k elems = 
    let ids = elems |> List.mapi (fun i uc -> i, idf uc)
    ids
        |> List.iter (fun (i, id1) ->
            ids
                |> List.iter (fun (j, id2) ->
                    if j > i && id1.idText = id2.idText then 
                        errorR (Duplicate(k, id1.idText, id1.idRange))))
    elems

because is computing idf for all elements too many times n * (n + 1) instead of just n

[forki]

could you please send it as new PR. I'd also suggest you unroll it as nested for loops .

[MecuSorin]

The suggested code will look like this:

let CheckDuplicates (idf: _ -> Ident) k elems = 
    let ids = elems |> List.mapi (fun i uc -> i, idf uc)
    for (i, id1) in ids do
        for (j, id2) in ids do
            if j > i && id1.idText = id2.idText then 
                errorR (Duplicate(k, id1.idText, id1.idRange))
    elems

[MecuSorin]

@forki The PR is #10325