Disable REPL hack in Julia 1.10

[fingolfin]

Resolves #864

@ThomasBreuer turns out this is much easier than we thought ;-)

[codecov]

Codecov Report

Merging #866 (5e22771) into master (9de7045) will increase coverage by 0.29%.
The diff coverage is 100.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #866      +/-   ##
==========================================
+ Coverage   75.34%   75.63%   +0.29%     
==========================================
  Files          50       51       +1     
  Lines        4104     4117      +13     
==========================================
+ Hits         3092     3114      +22     
+ Misses       1012     1003       -9     

Impacted Files	Coverage Δ
src/globals.jl	100.00% <100.00%> (ø)

... and 5 files with indirect coverage changes

[ThomasBreuer]

I cannot reproduce this.

I have tried the nightly builds julia-1dcac5eb40 (10 days old) and julia-4e782bf91a (of today) with the proposed change.
In both versions, entering GAP.Globals.MTX.Is followed by <TAB> does not yield any output.
With Julia 1.8.5, this input yields the expected 10 matches which I get also in the GAP session.

Why do I get a different behaviour? (Is it because of the operating system (ubuntu 20.04) or because of the terminal (xfce), or what?)

[fingolfin]

Argh. No, you are right, my mistake: I only tested GAP.Globals.IsP<TAB> as I forgot about the "nested" tab being the actual problem. Well, I also did run the test suite to verify it, knowing we had tests in it for the REPL completion... and the test suite passed!!!

... because we are not actually running the REPL completion tests 😠 so I just made PR #867 to fix that.

I'll resume looking for a proper solution. But in the meantime I will revamp this PR into simply disabling the REPL completion & tests for 1.10, to avoid the warnings when precompiling GAP.jl; we won't be worse off than we are this way.

[ThomasBreuer]

Thanks.

But in the meantime I will revamp this PR into simply disabling the REPL completion & tests for 1.10, to avoid the warnings when precompiling GAP.jl; we won't be worse off than we are this way.

The pull request does this already, except for the comment

In Julia >= 1.10, the hack is not necessary anymore.

[fingolfin]

Updated

[ThomasBreuer]

Looks good. Thanks.

[aviatesk]

You can add the following patch to get reasonable completions for cases like GAP.Globals.MTX.I|:

diff --git a/src/globals.jl b/src/globals.jl
index 8ee3805..0b7a3ac 100644
--- a/src/globals.jl
+++ b/src/globals.jl
@@ -37,7 +37,7 @@ Main
 """
 const Globals = GlobalsType()
 
-function getproperty(::GlobalsType, name::Symbol)
+Base.@assume_effects :effect_free :terminates_globally function getproperty(::GlobalsType, name::Symbol)
     v = _ValueGlobalVariable(name)
     v === C_NULL && error("GAP variable $name not bound")
     return _GAP_TO_JULIA(v)

I am not sure if the getproperty method holds the declared effects though (please assert it yourself if you use this patch). And if that holds, the effects declaration should probably be in _ValueGLobalVariable and _GAP_TO_JULIA since it allows a broader optimization possibilities.

[fingolfin]

OK, thanks for the hint! I am a bit wary of these annotations, after past experiments with @pure bit us. But these now new seem at least better documented, and after reading those comments, it seems we should indeed be able to use them here. Alas, they require Julia >= 1.8, but of course we can deal with that, too