Hi, and thanks for letting me know that pristine-tar failed. It's known to work on about 99.5% of gz files, so I suppose this makes you the unlucky fraction.
I've collected all the files you submitted into a testsuite (available
in the 'testsuite' branch of pristine-tar's git repo), and have done
some preliminary analysis.
The most common problem are tarballs that were produced on some sort of
Microsoft system (windows/dos), and that zgz's existing windows support quirks
don't handle. These include SFML, libmail-sender-perl, tinydent, hires, and
Config::IniHash. These all seem to have been created by compressors that use a
different algorythm than any zgz contains; zgz output does not approximate any
of them.
gnome-rdp is weird, it was apparently made by gzip, but is missing the OS
flag, and zgz cannot approximate it.
php-geoip can nearly be replicated by zgz in buggy-bsd quirk mode.
The header is not 100% right:
-00000000 1f 8b 08 00 00 00 00 00 00 03 ec 3d 7b 7f da 48 |...........={..H|
+00000000 1f 8b 08 00 00 00 00 00 00 03 ed 3d 7b 7f da 48 |...........={..H|
And there's a small variation near the end, where 00 00 ff ff 03 00
was inserted:
-00001ef0 a2 7f ff 07 00 00 ff ff 03 00 84 50 ed 40 00 bc |.............@..|
-00001f00 00 00 |..|
-00001f02
+00001ef0 a2 7f ff 07 84 50 ed 40 00 bc 00 00 |.......@....|
+00001efc
Finally, xsupplicant contains some additional padding NULs at the end but
can otherwise be reproduced.
The xsupplicant and php-geoip cases could be nicely handled by pristine-gz
deltas including a small binary diff.
If any of you are in contact with the creators of any of these and
could ask them what compression programs they used, that could potentially
be very helpful.
--
see shy jo
signature.asc
Description: Digital signature
