Missing support for sparse tar archives

[qkaiser]

Unblob does not support sparse tar archive because of the way the tarfile library handles those archives.

Our assumption is that the end of the archive corresponds to the last member offset + last member size, like so:

try:
    tf = tarfile.TarFile(mode="r", fileobj=file)
except tarfile.TarError:
    return -1

last_member = None

try:
    for member in tf:
        last_member = member
except (tarfile.TarError, SeekError):
    # recover what's already been parsed
    pass

if last_member is None:
    return -1

last_file_size = round_up(last_member.size, BLOCK_SIZE)
end_of_last_tar_entry = last_member.offset_data + last_file_size

This assumption does not hold true with sparse archives since tarfile set the TarInfo size to the original file size within _proc_sparse:

def _proc_sparse(self, tarfile):
    """Process a GNU sparse header plus extra headers.
    """
    # We already collected some sparse structures in frombuf().
    structs, isextended, origsize = self._sparse_structs
    del self._sparse_structs

    # Collect sparse structures from extended header blocks.
    while isextended:
        buf = tarfile.fileobj.read(BLOCKSIZE)
        pos = 0
        for i in range(21):
            try:
                offset = nti(buf[pos:pos + 12])
                numbytes = nti(buf[pos + 12:pos + 24])
            except ValueError:
                break
            if offset and numbytes:
                structs.append((offset, numbytes))
            pos += 24
        isextended = bool(buf[504])
    self.sparse = structs

    self.offset_data = tarfile.fileobj.tell()
    tarfile.offset = self.offset_data + self._block(self.size)
    self.size = origsize # <---- here
    return self

We therefore need a way to know the sparse'd size if the archive is a sparse archive, not the original file size.

We can check if any member of the archive is sparse by using the issparse() function. I suppose supporting sparsed archive would mean browsing through every entry in the archive, not just the last member.

[qkaiser]

It can be handled by monkey patching _proc_sparse with a version that does not update self.size. Tested locally but probably ugly :)