Much like zipfile.BadZipfile, we need a base custom exception for the 
gzip module. At least, catch gzip-related exceptions and throw a 
tarfile.TarError when used *via* tarfile.*.

See the following example (the exception escaped the "try... except 
tarfile.TarError: .. " block!):

***
  [...]
  File "/home/sridharr/as/pypm/src/pypm/common/compression.py", line 
199, in _ensure_read_write_access
    for tarinfo in tarfileobj.getmembers():
  File "/opt/ActivePython-2.6/lib/python2.6/tarfile.py", line 1791, in 
getmembers
    self._load()        # all members, we first have to
  File "/opt/ActivePython-2.6/lib/python2.6/tarfile.py", line 2352, in 
_load
    tarinfo = self.next()
  File "/opt/ActivePython-2.6/lib/python2.6/tarfile.py", line 2307, in 
next
    self.fileobj.seek(self.offset)
  File "/opt/ActivePython-2.6/lib/python2.6/gzip.py", line 382, in seek
    self.read(1024)
  File "/opt/ActivePython-2.6/lib/python2.6/gzip.py", line 219, in read
    self._read(readsize)
  File "/opt/ActivePython-2.6/lib/python2.6/gzip.py", line 284, in _read
    self._read_eof()
  File "/opt/ActivePython-2.6/lib/python2.6/gzip.py", line 304, in 
_read_eof
    hex(self.crc)))
IOError: CRC check failed 0x115929f0 != 0x9f074a38L

Unless something in the docs claims that there is/should be such a
thing, this is a feature request, not a bug ('behavior') report, and
only applicable to future x.y versions.

I have created a small patch, that introduces BadGzipFile exception. It is a subclass of IOError, so it would be backward compatible and will be still caught by old code, but this way is distinct from IOError.

Modified patch with test, that catches both BadGzipFile and IOError exceptions.

Bump! How about commiting this patch? Or maybe there is something missing? I'll be happy to fix it.

Since the patch makes BadGzipFile a subclass of IOError, it doesn't look unreasonable.
Some nits:
- a gzipped file is not an "archive"
- the unit tests should use either the "with" statement, or try/finally blocks to properly close the file even when the test fails
- you should probably explicitly test that BadGzipFile is a subclass of IOError

I'll be very happy to fix this after Friday. Thanks for your comments.

I had some time today, so I managed to fix the patch. I hope now everything is ok.

Bump! Antoine, do you think the patch is acceptable and can be committed now?

Your changes appear to address all three of Antoine's 'nits'.

Ezio has found a few other things to improve (follow the “review” link to the right of the patch link).

The attached patch follows Ezio's hints.

The new exception should also be documented, and a versionadded and Doc/whatsnew/3.4.rst entry added.

done!

Not all invalid gzip files raise BadGzipFile.  Some of them raises ZlibError.

Well, I specified the word ``files`` everywhere for that reason. Looking at Doc/library/zlib.rst I see:

  For reading and writing .gz files see the gzip module.

Also, I specified 'the gzip module' on the whatsnew section.
Is there anything more specific I could have done?

I added some comments on Rietveld.

read32() should raise BadGzipFile when less than 4 bytes read. Also you should introduce read8() function which raises BadGzipFile when less than 1 byte read.

See also issue4844 and issue14315 for zipfile.

I'd like to see this issue move forward, so I've created a PR.

New changeset cf599f6f6f1c392d8f12936982a370d533782195 by Serhiy Storchaka (Zackery Spytz) in branch 'master':
bpo-6584: Add a BadGzipFile exception to the gzip module. (GH-13022)
https://github.com/python/cpython/commit/cf599f6f6f1c392d8f12936982a370d533782195

messages: + msg341145
versions: + Python 3.8, - Python 3.4

messages: + msg175285

messages: + msg101187

nosy: + gruszczy
messages: + msg101185

keywords: + patch

messages: + msg91151

type: behavior -> enhancement