Implement getJSSyntacticDiagnosticsForFile

[ahejlsberg]

In this PR:

• Port getJSSyntacticDiagnosticsForFile which generates diagnostics for TypeScript-only constructs in .js files.
• Remove extra blank lines printed between diagnostics in -pretty false mode.
• In language service, include semantic diagnostics even when there are syntax errors in a file (as was the case in Strada).
• Check JSDoc comments in source element order to avoid indeterminate order of enumerating JSDoc cache.

Note that getJSSyntacticDiagnosticsForFile is now called by ParseSourceFile such that it becomes part of the concurrent parsing of source files. In Strada it was deferred and called upon the first request to obtain syntactic diagnostics for a file. The function effectively constitutes an extra pass over every parsed .js file. Longer term we could consider integrating the logic with the parser itself.

This PR supersedes #1387.

Fixes #1386.

[Copilot]

Pull Request Overview

This PR implements getJSSyntacticDiagnosticsForFile to generate diagnostics for TypeScript-only constructs in JavaScript files. The implementation also removes extra blank lines in -pretty false mode and includes semantic diagnostics even when syntax errors exist.

• Adds new JS-specific syntactic error checking that identifies TypeScript constructs in JS files
• Cleans up diagnostic output formatting by removing unnecessary blank lines
• Improves language service behavior by including semantic diagnostics alongside syntax errors

[jakebailey]

Seems fine to me, though it does slow down parsing of JS files to do the second pass:

goos: linux
goarch: amd64
pkg: github.com/microsoft/typescript-go/internal/parser
cpu: Intel(R) Core(TM) i9-10900K CPU @ 3.70GHz
                               │   old.txt   │               new.txt               │
                               │   sec/op    │    sec/op     vs base               │
Parse/Herebyfile.mjs/tsc-20      941.2µ ± 1%   1032.6µ ± 1%  +9.71% (p=0.000 n=10)
Parse/Herebyfile.mjs/server-20   956.7µ ± 3%   1020.3µ ± 4%  +6.65% (p=0.000 n=10)
geomean                          948.9µ         1.026m       +8.17%

                               │   old.txt    │               new.txt               │
                               │     B/op     │     B/op      vs base               │
Parse/Herebyfile.mjs/tsc-20      461.8Ki ± 0%   462.0Ki ± 0%  +0.04% (p=0.000 n=10)
Parse/Herebyfile.mjs/server-20   461.8Ki ± 0%   462.0Ki ± 0%  +0.04% (p=0.000 n=10)
geomean                          461.8Ki        462.0Ki       +0.04%

                               │   old.txt   │              new.txt               │
                               │  allocs/op  │  allocs/op   vs base               │
Parse/Herebyfile.mjs/tsc-20      1.761k ± 0%   1.766k ± 0%  +0.28% (p=0.000 n=10)
Parse/Herebyfile.mjs/server-20   1.761k ± 0%   1.766k ± 0%  +0.28% (p=0.000 n=10)
geomean                          1.761k        1.766k       +0.28%

This is nothing "new", the old code had this problem.

I don't see a way around this unless we move this code into the parser itself (which, I was surprised to learn it wasn't).

[jakebailey]

I had thought this code was correct originally; I guess you're changing it to match that VS Code on the client side requested syntax and semantic in sequence? (I've been trying to find equivalent code but I think this is just somewhere that differs in general with Strada given the fine-grained-ness of tsserver.)

[ahejlsberg]

I simply noticed a behavioral difference, with semantic error red squiggles appearing and disappearing depending on whether there are syntax errors as you're typing. LSP doesn't distinguish between syntactic/semantic so we need to return all of them to get consistent behavior.

[jakebailey]

Seemingly a difference here.

[jakebailey]

Probably inconsequential, given it's a missing body error in JS. Which, sort of funny that we can detect that there's a syntax error for JS, as that implies that this check could be elsewhere.

[ahejlsberg]

We already have grammar checks for all modifiers, so the check for export in Strada is duplicative (as evident from the diagnostics). Also, it's not even correct, export is permitted in JS.

[jakebailey]

This error seems to still be missing? But, probably "good" because the old error was duplicated for some reason.

[ahejlsberg]

Another case of duplicate checks.