When trying to achieve binary reproducible packages, .ipk packages are not comparing favourably with neither RPM nor DEB packages. For core-image-minimal (see https://bugzilla.yoctoproject.org/show_bug.cgi?id=5866 ) we get something like this: core-image-minimal (with INHERIT +="reproducible") ================================================== IPK RPM DEB ---------------- Same: 0 530 3617 Diff: 3868 3338 251 Total: 3868 3868 3868
Gzip by default adds a timestamp to the archive. Modified opkg-build to supress timestamping the gzip archive. Also, added deterministic directory entries sorting, as well as support for SOURCE_DATE_EPOCH. With those fixes, opkg now looks more inline with deb and rpm: Comparing 2781 files in git/oe-core/build//tmp-glibc/deploy/ipk/ to test_oe/openembedded-core/build//tmp-glibc/deploy/ipk/ Same: 2713 Different: 68 Total: 2781
(In reply to comment #1) > Gzip by default adds a timestamp to the archive. Modified opkg-build to > supress timestamping the gzip archive. Also, added deterministic directory > entries sorting, as well as support for SOURCE_DATE_EPOCH. With those fixes, > opkg now looks more inline with deb and rpm: > > Comparing 2781 files in git/oe-core/build//tmp-glibc/deploy/ipk/ to > test_oe/openembedded-core/build//tmp-glibc/deploy/ipk/ > Same: 2713 > Different: 68 > Total: 2781 Great news, is there a patch in ML or upstream?
yep, I posted to the ML today [1]. Usually I leave it there for a few days before pulling into the trunk. Once I do that, I'll close this bug (usually I don't get a whole lot of feedback from the ML, but still like to leave the patch there for a few days). [1] https://groups.google.com/forum/#!topic/opkg-devel/sFGFB-W9xM8
I hade a look at the patch, I like the way you check if really gzip or pigz. I filed a bug complaining about gzip != pigz a while ago.
took me a few hours to realize that OE uses pgzip....and that -n on gzip in pgzip is -nT...really annoying, glad there is bug, the discrepancy makes no sense!
fixed on 682f8c5e35b8854a9bb858b8ee1714d27e0c00db
Thanks, however the fix still needs to percolate to opkg-utils recipe in yocto. AFAIK the recipe presently uses a tarball based on release ver 0.3.5. Are you planning a new release as well (i.e. 0.3.6) ? There seems to be quite a number of additional changes since 0.3.5 in master. In any case, the recipe needs to be modified, be it another version or simply git master. If there are no takers (designated maintainers) I don't mind doing it myself.
I am planning the 0.3.6 release of opkg & opkg-utils by mid February. With the release I usually post new recipes as well. Since we are so close to the release, is it OK to wait? if not, the fix can certainly be backported to 0.3.5.
(In reply to comment #8) > I am planning the 0.3.6 release of opkg & opkg-utils by mid February. With > the release I usually post new recipes as well. Since we are so close to the > release, is it OK to wait? if not, the fix can certainly be backported to > 0.3.5. Mid February sounds good. Looking forward to it.