avoid isfile calls on find as much as possible

If `walk` is also returning the same file, it should already have
been included in the `out`. If `out` is not empty, we can
safely assume that it is either a directory or a file.

If it's empty, it could be that it is either an empty directory
or, the path is a file and was not included from the `walk` results.
This time we will make a `isfile` call. This is unfortunate but
better than before, where we were making an additional for each find.

This was noticed in DVC, which was making remote cache querying slower
when traversing through the caches (512 calls instead of 256 :( ).

Author: skshetry

fsspec/asyn.py

@@ -581,7 +581,7 @@ async def _find(self, path, maxdepth=None, withdirs=False, **kwargs):
             if withdirs:
                 files.update(dirs)
             out.update({info["name"]: info for name, info in files.items()})
-        if (await self._isfile(path)) and path not in out:
+        if not out and (await self._isfile(path)):
             # walk works on directories, but find should also return [path]
             # when path happens to be a file
             out[path] = {}

fsspec/spec.py

@@ -445,7 +445,7 @@ def find(self, path, maxdepth=None, withdirs=False, **kwargs):
             if withdirs:
                 files.update(dirs)
             out.update({info["name"]: info for name, info in files.items()})
-        if self.isfile(path) and path not in out:
+        if not out and self.isfile(path):
             # walk works on directories, but find should also return [path]
             # when path happens to be a file
             out[path] = {}