Allow control of --bounds-check in Pkg.test

[IanButterworth]

Forcing bounds checking to be on can severely slow down some test code, so it would be useful to be able to allow bounds checking to be the julia default automatic that observes in-code declarations (set here via check_bounds = nothing which omits the --check-bounds option entirely).

There may be some use cases where forcing it off entirely may be desired too (check_bounds = false), so all three are supported

[DilumAluthge]

Instead of adding another keyword argument to the Pkg.test function, could we do either of the following:

1. Just read the value of --bounds-check that was passed to the original (parent) Julia process.
2. Allow the user to pass --bounds-check to the julia_args keyword argument that we already have.

[IanButterworth]

Just read the value of --bounds-check that was passed to the original (parent) Julia process.

This would be a significant (breaking?) change, as the parent process is likely run by default with --check-bounds omitted, i.e. default checking that observes code declarations, not forced on like Pkg.test currently does

Allow the user to pass --bounds-check to the julia_args keyword argument that we already have.

Users can currently do this, but it's not possible to set it to the default, as that requires omitting --check-bounds entirely.

If Base was changed to accept --check-bounds=auto then the latter could work

[DilumAluthge]

If Base was changed to accept --check-bounds=auto then the latter could work

Regardless of what we end up doing in this PR, I do think it would be nice for Julia to accept --check-bounds=auto.

[IanButterworth]

Yeah and it would be nice to get that into 1.7

[fredrikekre]

Closing in favor of JuliaLang/julia#41551 then? In any case, if you need special behavior like this you can always start your own process from within the test/runtests.jl file. Many packages do that already for thigns like this.

[KristofferC]

Users can currently do this, but it's not possible to set it to the default, as that requires omitting --check-bounds entirely.

I don't understand why this is not possible. In case you want it as the default you just omit it in julia_args?

[IanButterworth]

By default I mean julia default, which respects @inbounds, which is a third state that requires omitting --check-bohnds entirely.

you can always start your own process from within the test/runtests.jl file

Yes but that makes managing test dependencies a pain, as there's no convenient way to merge a test Project and the parent in the same way that the sandbox does it, AFAIK

[fredrikekre]

Yes but that makes managing test dependencies a pain, as there's no convenient way to merge a test Project and the parent in the same way that the sandbox does it, AFAIK

You just launch julia with the env Pkg gave test/runtests.jl

[KristofferC]

By default I mean julia default, which respects @inbounds, which is a third state that requires omitting --check-bohnds entirely.

I understand that. But again, why isn't julia_args fine here, and just leave out the argument if you want the julia default?

[IanButterworth]

just leave out the argument if you want the julia default?

% julia --check-bounds
ERROR: option `--check-bounds` is missing an argument
% julia --project --check-bounds=
ERROR: julia: invalid argument to --check-bounds={yes|no} ()

or am I misunderstanding your suggestion?

You just launch julia with the env Pkg gave test/runtests.jl

The env Pkg gave test/runtests.jl is a merge of the two envs in a temp dir sandbox. How would I access that? or do you just mean launch with --project in /test?

[KristofferC]

or am I misunderstanding your suggestion?

Yes, I think so. Just don't add the whole --check-bounds bit at all.

[IanButterworth]

Are you factoring in that it's currently hardcoded before julia_args in the sandbox?

Pkg.jl/src/Operations.jl

Line 1374 in cb9441f

--check-bounds=yes

[KristofferC]

No, that would make it harder indeed 😆. Having an auto flag like in the Julia PR then seems like probably the best?

[IanButterworth]

Yeah. I'm happy for that PR to be the one