Thanks for this PR @NiklasReisser! Just merged #81, is this still required?

Hi @casperdcl ,

I am on holiday without my laptop from work, so I can't test it.
But for my use case, I'd say yes, this is still required.

Here is what I want to do:
I have a bunch of virtual resource paths which look like filesystem paths, but actually aren't.

For example host/class/subclass/name.

Now I want to complete that peacemeal just like a filesystem path.
To make this work I need compopt -o nospace, otherwise readline adds a space after every part of the path.

I'd also be fine with completing the whole path at once, like any other comletion, but in this case the issue is that shtab enables -o filenames globally for all completions. This causes readline to only print the last bit of the path for the completion.
So if I type hos<tab>, and return the full path from my custom completion function, I get name (and only name) as possible completion.

Best regards,

Niklas

@casperdcl Can confirm now, I'd still need this for my usecase.

Cheers
Niklas

@casperdcl This issue seems to also prevent usage of bash completion builtins like _available_interfaces, which modify COMPREPLY directly.

There's some conflicts now though with the main branch.

Also, would making this line configurable be a simpler alternative?

Hi,

I've resolved the merge conflict.

Unfortunately, I think making the main "complete" configurable would only allow a global setting for all completions for a given program. My approach is more generic.

I admit I don't particularily like it due to the tmpfile. But it is the only approach I can think of without changing the API.
(With a breaking API-Change, the inner, user-defined function could set or extend COMPREPLY, instead of of printing to stdout.)

@0x2b3bfa0 - PTAL - worth resolving conflicts and merging this?

@omesser, this pull request uses mktemp in a hardly avoidable but barely convenient way; being unable to find solutions that don't involve creting a subshell, I still hesitate to merge it.