Bug report

Bug description:

Many common operations on tarfile.TarFile are orders-of-magnitude slower when you open it with mode r|* than with the default mode r:*. The better the archive's compression ratio, the bigger the speed difference.

Slow operations include: calling tf.list(); iterating TarInfos in the TarFile without extracting them; extracting with tf.extractall(); extracting with tf.extractfile(info) + shutil.copyfileobj; etc.
In contrast, reading a whole file into memory with extractfile() + f.read() is not affected.

Steps to reproduce

1. Create test tar files containing 3 files of size 100MB filled with the letter 'x':

   rm -rf data; mkdir data; for i in 1 2 3; do head -c100M /dev/zero | tr '\0' 'x' > data/$i.dat; done
   tar caf test.tar.gz data
   tar caf test.tar.xz data
   tar caf test.tar.bz2 data

   They are quite well compressed from the original size (300MB):

   -rw-r--r-- 1 tomi users 299K Jun 27 23:02 test.tar.gz
   -rw-r--r-- 1 tomi users  46K Jun 27 23:02 test.tar.xz
   -rw-r--r-- 1 tomi users  566 Jun 27 23:02 test.tar.bz2
   

2. Compare r:* versus r|*:

   import sys, tarfile
   with tarfile.open(sys.argv[1], sys.argv[2]) as tf: tf.list()

Expected results

Both modes should be fast and finish in a few seconds.

Actual results

Root cause

_Stream.dbuf can become very large. Slicing it with self.dbuf = t[size:] is accidentally quadratic™.

Mode r:* uses _compression.DecompressReader instead of tarfile._Stream, which calls self._decompressor.decompress() with a max_length parameter.

I might try to send a pull request if I figure it out.

CPython versions tested on:

3.10, 3.11, CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-121109: Fix performance of tarfile reading with "r|*" #121296
• [3.15] gh-121109: Fix performance of tarfile reading with "r|*" (GH-121296) #150604