1.9.0 fseek for zstandard file results in io.UnsupportedOperation: underlying stream is not seekable

[gcflymoto]

Can be easily reproduced

/nfs/site/home/<user>/pyenv3.11.1/bin/python3
Python 3.11.1 (main, Dec  6 2023, 09:00:14) [GCC 12.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import xopen, os
>>> f = xopen.xopen("file.zst")
>>> f.seek(0, os.SEEK_END)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/nfs/site/home/<user>/pyenv3.11.1/lib/python3.11/site-packages/xopen/__init__.py", line 476, in seek
    return self._file.seek(offset, whence)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
io.UnsupportedOperation: underlying stream is not seekable

[rhpvorderman]

Xopen never guarantees seekability. Xopen streams the zstd compressed data to the zstd application using pipes. Pipes are not seekable. Xopen also allows specifying stdin as input, which is also not seekable.

You can utilize threads=0 and make sure the zstandard package is installed. That way the file is always passed on to zstandard.open which should allow seeking.

If you are always opening zstandard files, it is almost always a better option to use the zstandard module itself. Unless you want the extra speed that xopen offers by piping it through the zstandard application, but in that case it is guaranteed not to be seekable. (There may be some exceptions on Windows IIRC).

[gcflymoto]

Hi. I tried it with xopen.xopen("file.zst", threads=0) with zstandard package is already installed

Zstandard package is installed
zstandard-0.22.0.dist-info/ zstandard/

but it has the same error. Is there any way to trace whether zstandard package was used?

[marcelm]

You can check the .buffer.raw attribute:

>>> f = xopen.xopen("tests/file.txt.zst", threads=0)
>>> f.buffer.raw
<zstd.ZstdDecompressionReader object at 0x7f7cee2a0f30>

It is possible to use f.buffer.raw.seek(100), but you can not seek backwards and I strongly suspect that seeking forwards is implemented by just decompressing the stream until the desired position is reached, that is, it is very inefficient compared to how seeking in an uncompressed file would work. (This is the same for the other supported compression formats.)

Maybe we could try to more consistently provide a forward-only seek function. I’ve previously not thought that this is a good idea because seek() to me implies that it is efficient, but I see now that GzipFile, BZ2File, ZstdDecompressionReader and IGzipFile all support seek()ing, so maybe we should ensure that this works with files opened through xopen as well.

Is seeking forwards what you need?

[gcflymoto]

Strangely I see this

print(f.buffer.raw)
AttributeError: 'zstd.ZstdDecompressionReader' object has no attribute 'raw'

Is seeking forwards what you need?

It's actually grabbing the end N number of lines like tail, but I can make it work with forward only seek

[marcelm]

Not sure what is going on, but it appears you need to use f.buffer.seek instead. But as Ruben said, you can also just use zstandard.open directly for now. This works for me:

python3 -c 'import zstandard; f=zstandard.open("tests/file.txt.zst"); f.seek(10)'

[gcflymoto]

Thank you for the workaround

[marcelm]

Let me re-open this until we’ve decided whether this is something we should fix.