Info call makes unnecessary long listing for "directories"

[shcheklein]

When path is a "directory" (prefix) we might end up going into full listing operation to just get an info for a single path. We call ls and we don't pass any limits (and it seems GCSFS doesn't support max results atm?).

        out = await self._ls(path, **kwargs)  # Here we can go into a long listing since we don't limit it
        out0 = [o for o in out if o["name"].rstrip("/") == path]
        if out0:
            # exact hit
            return out0[0]
        elif out:
            # other stuff - must be a directory

Is it intentional for some reason?

On s3 and Azure, in similar place we list to check if at least a single key exists (we don't check for an exact hit).

Can we drop the exact hit check and pass max results into ls?

[martindurant]

If I understand, you want there to be a max results argument to ls OR that there be a first-pass exact hit check?

On s3 and Azure, in similar place we list to check if at least a single key exists (we don't check for an exact hit).

It is totally reasonable that all three should have the same logic here, so if you can phrase this in code as a PR, I'd be happy to see it.

[shcheklein]

thanks @martindurant !

I think s3 / Azure logic is just to have a check that at least some objects exist under the prefix. They don't do exact hit check AFAIU, and they don't list everything.

So, the proposed solution is to:

• remove the exact check (not sure why it was needed tbh?)
• add max results supports to list as minimum as possible (get_info should be a constant in terms of API calls - ideally a single call)

so if you can phrase this in code as a PR, I'd be happy to see it.

I'll try to get there. Intention of this issue was to ask if there were some reasons for the existing logic that I'm not aware / can't see.