Hello Jelle. Jelle van der Waa wrote in <[email protected]>: |For Arch Linux we strive to make every package reproducible and we |noticed that s-nail is currently not reproducible due to the recording |of MAKEFLAGS in the s-nail binary by the VAL_BUILD_REST define. | |.obj/mk-config.h:#define VAL_BUILD_REST " -j100 --jobserver-auth=3,4" | |Would s-nail be open to removing the recording of MAKEFLAGS in the |resulting binary. The test result of building s-nail twice with a |different configured MAKEFLAGS can be viewed here: | |https://tests.reproducible-builds.org/archlinux/core/s-nail/s-nail-14.9.\ |15-2-x86_64.pkg.tar.xz.html | |More information about reproducible builds. https://reproducible-builds.\ |org/
Well. Hm. We are reproducable for many years! For me this smells like Debian messed around (again) with the definition of what a reproducible build actually is, or how up-to-date their builders are. I had a short run on the Debian list last year on that already [1]. [1] https://lists.debian.org/debian-devel/2018/03/msg00036.html Looking at the Debian tracker, i see that the reproducibility suns disappeared on all tested platforms, but for the unstable branch only. To me this is again a misconfiguration of the test builders. I think baking MAKEFLAGS into the binary is just as proper as baking in compiler command line flags, they just did not think about it. The make command line is used for configuration options, just as are ./configure arguments for other packages. While some of them are deducable by other means, some may not, or be only very hard to collect whereas now a user could simply mail bugs in conjunction with the output of "mailx -v -Xversion -Xx". And really, the Debian tests fail because they same to pass paths like test1/ and test2/ (or so) into our configuration -- via make command lines? How can that be right? |Thanks in advance, I refrain from doing that, because our build system allows passing prefilled INCS= and LIBS= variables etc, and we also allow predefined CONFIG= to be passed, things which are _only_ tracked via the make(1) command line, and which thus enters MAKEFLAGS. I think reproducible-builds.org should treat make(1) flags just as they do with other flags, and like i said in above thread you can close the issue that is my rant, but will not get away from the fact that you cannot expect exactly identical binary outcome on two different build hosts, unless the actual build environment is the same to the detail. Unless the social pressure becomes unbearable to me this is a bug on the Debian side. Ciao, and a nice Sunday i wish! --steffen | |Der Kragenbaer, The moon bear, |der holt sich munter he cheerfully and one by one |einen nach dem anderen runter wa.ks himself off |(By Robert Gernhardt)
