Mailinglist Archive: opensuse-factory (269 mails)

< Previous Next >
[opensuse-factory] openSUSE reproducible builds status 2019-09
Hi,

last month's status:
https://lists.opensuse.org/opensuse-factory/2019-08/msg00186.html

Last months' reproducible builds project updates (including my work):
https://reproducible-builds.org/reports/2019-08/

Builds this month were much cleaner again with several long-standing
build issues fixed.
This month, I also pushed https://github.com/python/cpython/pull/12341
for inclusion in our python and python3 packages to fix a whole class of
reproducibility issues with python C bindings.
Tested it with python-ephem - the latest (and last) newcomer in Factory
to be affected.

Another toolchain fix that helps several others to become reproducible
is for OpenStack packages:
https://build.opensuse.org/request/show/732328



I uploaded https://rb.zq1.de/compare.factory-20190926/ today
and rbstats are:
total-packages: 12270
build-tried: 12260
build-failed: 115
build-n-a: 125
build-succeeded: 12020
build-official-failed+na: 220
build-compare-failed: 408
build-compare-succeeded: 11612

verify-failed: 588
verified-semi-reproducible: 11396
verified-bit-identical: 0

bit-by-bit-identical: 11201
not-bit-by-bit-identical: 814


https://rb.zq1.de/compare.factory-20190926/graph.png
shows the change over time

https://rb.zq1.de/compare.factory-20190926/unreproduciblerings.txt
lists very unreproducible core packages (bootstrap+DVD)

Of the badly unreproducible packages,
3 were in ring0
49 were in ring1

That makes it 52/2901 => 1.79 %
which is below the overall average of
408/12020 => 3.39 %

814/12020 => 6.77 % of packages are not perfectly reproducible




Nearly fixed core packages:
MozillaThunderbird
https://build.opensuse.org/request/show/732106
kernel-vanilla
https://lists.opensuse.org/opensuse-kernel/2019-08/msg00000.html


I also found it is possible to build a reproducible Firefox, but only
when disabling PGO - and that was only possible when disabling LTO at
the same time (build errors otherwise):
https://build.opensuse.org/request/show/733089
So this is one of the places where one would have to make a trade-off
between performance of the binary and reproducibility.
I heard, upstream solved this by publishing the profiling data for other
people to use as input to their builds. Not such a nice solution either.

Then there is still one rpm update missing to use -flto=auto
As can be seen in
https://build.opensuse.org/package/revisions/openSUSE:Factory/rpm it was
briefly there and then reverted the next day, because it was bundled
with another (broken) change.


Since I am keeping a history of all Tumbleweed binaries (see
https://lizards.opensuse.org/2019/04/03/experimental-opensuse-mirror-via-ipfs/
), I also noticed that the normal Factory update grew from ~4GB to ~8GB
after LTO was enabled.
Still need to do a closer investigation on why that is.

Here is some raw data for x86 dirs with hardlinks:
opensuse/history$ du -smc *
90090 20190630
4270 20190701
4967 20190702
5225 20190703
4493 20190704
7526 20190708
72674 20190713 # LTO rebuild
18845 20190716
7373 20190718
17633 20190721
7812 20190723
16653 20190724
7137 20190726
16724 20190730
20694 20190805
7951 20190806
8047 20190809
8182 20190810
9065 20190814
19168 20190815
11267 20190820
8955 20190822
8326 20190823
7129 20190824
32824 20190828
9543 20190829
9532 20190902
7183 20190904
7836 20190905
8484 20190907
9266 20190909
11367 20190916
6468 20190917
10443 20190918
8100 20190920
7868 20190921
519103 total


Ciao
Bernhard M.
--
To unsubscribe, e-mail: opensuse-factory+unsubscribe@xxxxxxxxxxxx
To contact the owner, e-mail: opensuse-factory+owner@xxxxxxxxxxxx

< Previous Next >
Follow Ups