urllib sets wrong Content-Length for pseudo files on Linux

Bug report

A value of the Content-length header returned by urllib.request.FileHandler.open_local_file may mismatch the length of data on Linux.

This happens when a file from a special file system (e.g., procfs or sysfs) is requested.

open_local_file relies on st_size; st_size is equal to zero for pseudo files on Linux.

cpython/Lib/urllib/request.py

Lines 1506 to 1511 in 8a0d9a6

    
           size = stats.st_size 
        
           modified = email.utils.formatdate(stats.st_mtime, usegmt=True) 
        
           mtype = mimetypes.guess_type(filename)[0] 
        
           headers = email.message_from_string( 
        
               'Content-type: %s\nContent-length: %d\nLast-modified: %s\n' % 
        
               (mtype or 'text/plain', size, modified))

Example

>>> import urllib.request
>>> url = "file:///proc/cpuinfo"
>>> handler = urllib.request.FileHandler()
>>> response = handler.file_open(urllib.request.Request(url))
>>> data = response.read()
>>> headers = response.info()
>>> assert int(headers["Content-length"]) == len(data)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AssertionError
>>> print(headers["Content-length"])
0
>>> print(len(data))
18294

Your environment

CPython versions tested on: 3.12.0 alpha 0
Operating system and architecture: Linux

pythongh-93296: test that urllib sets wrong `Content-Length` for pseu…

I'm not sure this really is a bug. What would you expect?

I'm not sure this really is a bug. What would you expect?

I would expect a value of Content-Length to be equal to the size of the HTTP message body regardless of the peculiarities of a file in this case.

curl returns the same result:

$ curl --dump-header /dev/stdout file:///proc/cpuinfo
Content-Length: 0
Accept-ranges: bytes
Last-Modified: Fri, 16 Feb 2024 14:54:57 GMT

processor       : 0
...

Header Content-Length: 0 followed by non-empty body.

AFAIK, a similar discrepancy can happen if the file changes while it's being prepared/sent.

urllib asks the system for size, mtime and contents, and gives them to you.
I don't think it's urllib's job to second-guess the system.

The situation is the same as when a HTTP server sends a mismatched Content-Length, except here it's the kernel, not a server. It's up to you to handle it.

IMO, this is not an issue. If there are no objections, I'll close in a month.

curl returns the same result:
$ curl --dump-header /dev/stdout file:///proc/cpuinfo
Content-Length: 0
Accept-ranges: bytes
Last-Modified: Fri, 16 Feb 2024 14:54:57 GMT

processor       : 0
...
Header Content-Length: 0 followed by non-empty body.

This is interesting, it used to return no content, but that was changed in 2016 curl/curl#681 (comment)

The situation is the same as when a HTTP server sends a mismatched Content-Length, except here it's the kernel, not a server. It's up to you to handle it.

IMO, this is not an issue. If there are no objections, I'll close in a month.

As a user I'd expect a server to handle known cases, but I don't insist on keeping the issue open.

Navigation Menu

Repository navigation

urllib sets wrong `Content-Length` for pseudo files on Linux #93296

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

	size = stats.st_size
	modified = email.utils.formatdate(stats.st_mtime, usegmt=True)
	mtype = mimetypes.guess_type(filename)[0]
	headers = email.message_from_string(
	'Content-type: %s\nContent-length: %d\nLast-modified: %s\n' %
	(mtype or 'text/plain', size, modified))

urllib sets wrong Content-Length for pseudo files on Linux #93296

Description

Activity

carlbordum commented on Jun 1, 2022

illia-v commented on Jun 1, 2022

serhiy-storchaka commented on Feb 21, 2024

encukou commented on Mar 26, 2024

illia-v commented on Mar 27, 2024

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions

urllib sets wrong `Content-Length` for pseudo files on Linux #93296