Also complete commands (files in the env path) in REPL shell mode

[mariushoch]

Tested on Fedora only.

[tkelman]

I think this would be better as a keyword argument rather than positional

[mariushoch]

Addressed comments by @tkelman … thanks.

[mariushoch]

rebased

[yuyichao]

Assuming you are copying the block above, I think the path should be homedir().

You also don't need to do this in the home directory, #11440 need it because it is testing the ~ expansion. (Although a hidden and random directory might be better @blakejohnson ). You should use a tempdir for this PR.

[blakejohnson]

@yuyichao I was thinking I should at least revise the test I added to cleanup on failure.

[mariushoch]

Not using the homedir makes sense. Changed.

[mariushoch]

Any chance someone could have a look at this again?

[tkelman]

kind of feature-ish to merge this right now, would be better to wait until we've branched so any potential issues can be addressed with less time pressure

[StefanKarpinski]

Yeah, let's merge after 0.4 and then mark it as eligible for backporting since it's purely interactive.

[mariushoch]

Thanks for the heads up, that sounds good to me.

[tkelman]

Master's open again and we've branched, so I guess this is okay to merge and we can target to backport it for 0.4.1. Maybe prior to that, but would prefer this get a fair amount of testing on master first.

[tkelman]

Anyone familiar with the repl completions code have an opinion here?

[blakejohnson]

LGTM

[tkelman]

oh, I think isexecutable is likely to be deprecated soon

[mariushoch]

Can you (offhand) tell me what to use instead? If so, I'll amend this… otherwise, this can still be done later easily.

[mariushoch]

I looked for that and it is pending merge (#12819).

[tkelman]

isfile is probably the most appropriate choice for now

[mariushoch]

Not sure about that, there might be non executable files in PATH. In a perfect world, we would also filter based on whether the current user is able to execute the file in question (like bash does), but I don't see a nice way to do that (latest bash uses glibc's eaccess or access if available, otherwise it tries to figure on its own).
Implementing something like that might be something we want to do in a follow up.

[tkelman]

Completing on non-executable files isn't the end of the world. That function is about to be deprecated, so we can't be adding uses of it in base.

[mariushoch]

Fair enough, I'll amend this patch and add a note.

[tkelman]

I guess it could be filtered on file mode, but maybe leave that for a second round.

[mariushoch]

Amended to no longer use the soon to be deprecated isexecutable.

[mariushoch]

This breaks in case a user has something in their path which they can't actual read. #13959 will fix that.