Re-enable REPL completion for GAP records in Julia 1.10

[fingolfin]

The added annotation might also benefit performance of some code.

Thanks to @aviatesk who helpfully suggested this in this comment.

[codecov]

Codecov Report

Merging #914 (e059363) into master (3623cee) will not change coverage.
The diff coverage is 100.00%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #914   +/-   ##
=======================================
  Coverage   75.56%   75.56%           
=======================================
  Files          51       51           
  Lines        4170     4170           
=======================================
  Hits         3151     3151           
  Misses       1019     1019           

Files Changed	Coverage Δ
src/globals.jl	100.00% <ø> (ø)
src/ccalls.jl	99.17% <100.00%> (ø)

[fingolfin]

Weird. The CI on 1.10 keeps failing. But locally on my laptop with Julia 1.10.0-alpha1, the tests pass just fine.

[ThomasBreuer]

@fingolfin The same happens in my installation of 1.10.0-alpha1.
I get the error messages from the CI tests when I run the tests in the master branch, but with the changes from this pull request the errors disappear.

[fingolfin]

I figured out how to reproduce the errors locally: by enabling coverage, i.e., using Pkg; Pkg.test(; coverage = true).

Indeed, I can also reproduce the problem interactively if I start Julia with the option --code-coverage, then do using GAP and try to tab complete e.g. GAP.Globals.MTX.IsInd.

So why does code coverage break this? Huh.

[ThomasBreuer]

@fingolfin I can confirm your observation: When Julia 1.10 is started with --code-coverage, completion does not work in the session, but completion works when Julia 1.10 is started without --code-coverage.

Very strange.

[ThomasBreuer]

I have run the two Julia sessions (with and without --code-coverage) in parallel, and followed the code that is used for the input s = "GAP.Globals.MTX.IsI"; @which completions(s,lastindex(s)).

I get a difference at res = repl_eval_ex(ex, context_module) where ex = :(GAP.Globals.MTX) and context_module = Main.
Specifically, calling repl_eval_ex(:(GAP.Globals.MTX), Main) yields Any in one Julia session and a complicated object of type Core.Const in the other session.

(Inside REPLCompletions.repl_eval_ex, the first difference occurs in the result of REPLInterpreter(result), which looks rather cryptic.)

[fingolfin]

I wonder if code coverage collection interferes in some way with @assume_effects :effect_free :terminates_globally

[fingolfin]

OK this is a known thing, see JuliaLang/julia#49978

[fingolfin]

Based on that bug and the discussion on the Julia slack, I think the best solution for now is to just skip this test when code coverage tracking is on. It'll still work in REPL mode for 99.9% of users (I think people rarely turn on code coverage tracking for interactive use)