x/tools/gopls: various analyzers panic

[pjweinb]

With gopls at tip, and go either at 1.15.1 or at tip, internal/lsp/testdata/semantic/a.go causes several analyzers to panic.

a.go contains code that will not build, and some analyzers panic on a type assertion where the types look like []interface{}.(string).

gopls will run analyzers on code that does not build, so it can test proposed fixes,
that is, in cache/analysis.go:325:

		data.err = errors.Errorf("analysis skipped due to errors in package: %v", pkg.GetErrors())
		return data
}

the test cannot be replaced by if len(pgk.GetErrors()) > 0, because then the tests for proposed fixes fail.

Most of the tests for analyzers won't run on files that don't build (nor will the tests in internal/lsp/source) so one could argue that these analyzers, at least, aren't intended to work on such code. Choices seem to be:

• Put up with panic messages from gopls (bad user experience)
• Fix the analyzers one by one (whack-a-mole)
• For code that does not build, only run the analyzers that propose fixes (contributions welcome)
• Do a more careful job at analysis.go:325 in deciding whether to run analyzers (contributions welcome)
• Fix a.go so it doesn't trigger panics (push the problem off into the future)

[gopherbot]

Change https://golang.org/cl/273766 mentions this issue: go/analysis: avoid panic in ifaceassart

[gopherbot]

Change https://go.dev/cl/420538 mentions this issue: internal/lsp/cache: invalidate analysis results on packages.Load

[adonovan]

The analysis driver was rewritten in CL 443099 and related work, and we took a zero tolerance approach to analyzer panics, so I think this is fixed.