HTTPFileSystem isdir downloads the whole file

[mxmlnkn]

I need to implement the FUSE getattr (stat) callback. I.e., I need to get at least the file type and size, and possibly name for a given path.

I am failing to do this with the HTTP filesystem implementation because:

• info(path) always returns the file information for the HTML file, i.e., the file type is also always a file. This is already inconsistent to all other fsspec implementations. The same for isfile, which always returns true.
• isdir(path) hangs and when looking at my local HTTP server log or at my network bandwidth when testing with an external server, I see that this call downloads the whole file. This means that currently an ls -la will download all files in the given folder...

Test to reproduce:

import pprint
import time
import fsspec

prefix="https://ash-speed.hetzner.com/"

def timedCall(f, *args):    
    t0 = time.time()
    result = f(*args)
    t1 = time.time()
    print(f"{f} took {t1 - t0:.3f} s")
    pprint.pprint(result)
    print()

f = fsspec.open(prefix)

print(f"# Testing {prefix}\n")
timedCall(f.fs.exists, prefix)
timedCall(f.fs.listdir, prefix)
timedCall(f.fs.info, prefix)
timedCall(f.fs.isfile, prefix)
timedCall(f.fs.isdir, prefix)

path = prefix + "100MB.bin"
print(f"# Testing {path}\n")
timedCall(f.fs.exists, path)
timedCall(f.fs.info, path)
timedCall(f.fs.isfile, path)
timedCall(f.fs.isdir, path)

Output:

# Testing https://ash-speed.hetzner.com/

<function HTTPFileSystem._exists at 0x7fb0be244700> took 0.362 s
True

<bound method AbstractFileSystem.listdir of <fsspec.implementations.http.HTTPFileSystem object at 0x7fb0beaca680>> took 0.110 s
[{'name': 'https://ash-speed.hetzner.com/10GB.bin',
  'size': None,
  'type': 'file'},
 {'name': 'https://ash-speed.hetzner.com/100MB.bin',
  'size': None,
  'type': 'file'},
 {'name': 'https://ash-speed.hetzner.com/1GB.bin',
  'size': None,
  'type': 'file'}]

<function HTTPFileSystem._info at 0x7fb0be244550> took 0.763 s
{'ETag': '"60f52d50-143"',
 'mimetype': 'text/html',
 'name': 'https://ash-speed.hetzner.com/',
 'size': 323,
 'type': 'file',
 'url': 'https://ash-speed.hetzner.com/'}

<function HTTPFileSystem._isfile at 0x7fb0be2445e0> took 0.108 s
True

<function HTTPFileSystem._isdir at 0x7fb0be244670> took 0.108 s
True

# Testing https://ash-speed.hetzner.com/100MB.bin

<function HTTPFileSystem._exists at 0x7fb0be244700> took 0.216 s
True

<function HTTPFileSystem._info at 0x7fb0be244550> took 1.098 s
{'ETag': '"60c9b8bd-6400000"',
 'mimetype': 'application/octet-stream',
 'name': 'https://ash-speed.hetzner.com/100MB.bin',
 'size': 104857600,
 'type': 'file',
 'url': 'https://ash-speed.hetzner.com/100MB.bin'}

<function HTTPFileSystem._isfile at 0x7fb0be2445e0> took 0.428 s
True

<function HTTPFileSystem._isdir at 0x7fb0be244670> took 38.450 s
False

Imho, isdir should be implemented via a listdir to the parent if there is no other way. I am also wondering what it does check. Is it simply doing a mimetype check whether it is HTML? If so, then the first 1000 or so bytes would suffice. But then, wouldn't it detect arbitrary HTML files inside a given "folder" wrongly as a folder?

My current workaround is to call info first and only call isdir if mimetype is text/html. This logic could also be implemented in HTTPFileSystem if there is no better way.

[martindurant]

The way that ls works for HTTP, is to download the URL/page, and look for links to URLs that look like children of the original page. This works well for "ftp-style" servers (like python -m http.server). That's the only way to know if a URL is a directory.

I guess, it would be reasonable to shortcut isdir to return False for ANY URL that isn't HTML?

[martindurant]

I suppose the shortcut would be in info or ls actually - would you like to have a go at coding that? Before calling .text here, checking the content-type of r and returning nothing for any request that isn't HTML should avoid the download.