Problems reading stdin on Windows 11 Python 3.12

[peterjc]

This issue was identified on peterjc/thapbi-pict#627 where continuous integration tests using cutadapt to generate test inputs started to fail WITHOUT relevant code changes. i.e. Something changes in the AppVeyor Windows environment (unclear what) and/or the PyPI packages (nothing obvious).

I have since reduced this to a local test case under Windows 11, Python 3.12.4 (installed from python.org with the PATH option ticked), and cutadapt 4.9 (installed with pip install -U cutadapt):

C:\Users\xxx>python --version
Python 3.12.4

C:\Users\xxx>pip install -U cutadapt
Collecting cutadapt
  Downloading cutadapt-4.9-cp312-cp312-win_amd64.whl.metadata (3.5 kB)
Collecting dnaio>=1.2.0 (from cutadapt)
  Downloading dnaio-1.2.1-cp312-cp312-win_amd64.whl.metadata (3.6 kB)
Collecting xopen>=1.6.0 (from cutadapt)
  Downloading xopen-2.0.2-py3-none-any.whl.metadata (15 kB)
Collecting isal>=1.6.1 (from xopen>=1.6.0->cutadapt)
  Downloading isal-1.6.1-cp312-cp312-win_amd64.whl.metadata (10 kB)
Collecting zlib-ng>=0.4.1 (from xopen>=1.6.0->cutadapt)
  Downloading zlib_ng-0.4.3-cp312-cp312-win_amd64.whl.metadata (6.9 kB)
Downloading cutadapt-4.9-cp312-cp312-win_amd64.whl (231 kB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 231.1/231.1 kB 939.6 kB/s eta 0:00:00
Downloading dnaio-1.2.1-cp312-cp312-win_amd64.whl (86 kB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 86.2/86.2 kB 808.2 kB/s eta 0:00:00
Downloading xopen-2.0.2-py3-none-any.whl (17 kB)
Downloading isal-1.6.1-cp312-cp312-win_amd64.whl (201 kB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 201.5/201.5 kB 942.2 kB/s eta 0:00:00
Downloading zlib_ng-0.4.3-cp312-cp312-win_amd64.whl (88 kB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 88.5/88.5 kB 1.7 MB/s eta 0:00:00
Installing collected packages: zlib-ng, isal, xopen, dnaio, cutadapt
Successfully installed cutadapt-4.9 dnaio-1.2.1 isal-1.6.1 xopen-2.0.2 zlib-ng-0.4.3

C:\Users\xxx>cutadapt --version
4.9

This was then patched to include d9cf273 for a clearer error message.

Working example using a filename and one of the cutadapt test files (the primer isn't really appropriate for this dataset):

C:\Users\xxx\cutadapt\tests\data>cutadapt --quiet -g GAAGGTGAAGTCGTAACAAGG 454.fa | find ">" /C
59

Restructured to read from stdin:

C:\Users\xxx\Projects\cutadapt\tests\data>type 454.fa | cutadapt --quiet -g GAAGGTGAAGTCGTAACAAGG - | find ">" /C
Input file format not recognized. The file starts with b'TGTA', but files in supported formats start with '>' (FASTA), '@' (FASTQ) or 'BAM'
0

Inserting additional logging suggests function detect_file_format is called twice, the first time works and says FASTA format. The second time it is part way though the file, and fails.

This reminds me of #774, but is something Windows specific it seems.

[rhpvorderman]

does it work with the latest xopen 1.x version? 2.0.0 was a massive refactoring that enabled all sorts of cool functionality while also using less code, but as a consequence there were unforeseen bugs.

[peterjc]

I will try and check that next week (won't have access to a Windows machine over the weekend).

[peterjc]

No change with xopen 1.7.0, 1.8.0, or 1.9.0 - and my debugging to stderr still shows the detect_file_format function being called twice.

[marcelm]

I've now been able to reproduce the problem on Windows 10. I can also see where the function is called twice, will see what I can do.

[marcelm]

Interesting, it works on Python 3.10, 3.11 and 3.12.0, but fails on 3.12.4. Bisecting points to commit python/cpython@de347c0 (part of 3.12.3).