RFC Improve native/cross reproducibility
Get series with b4
b4 am [email protected]Versions newest first · linked by author and title
- v4 RFC Latest2 patches 0 repliesNew
- v3 RFC4 patches 0 repliesSuperseded
- v2 RFC2 patches 6 repliesSuperseded
Earlier versions · 1
| Version | Posted | Patches | Replies | Status |
|---|---|---|---|---|
| v1 RFC | 9 patches | 6 replies | Superseded |
Changes in v4 from the cover letter
* Rebase on top of master
* Cleanup fixes for openssl/ncurses/perlcross/perl
* Go back to padding rpath and revert the python3-native/kernel fixes, since this feels like the least intrusive change, and from what I understand cmake also does a similar type of rpath padding in some situations
Patches in v4 No applies recorded yet
| # | Subject and files | Size | Replies | Applied |
|---|---|---|---|---|
| 1/2 | bitbake.conf: Pad rpath in native binariesmeta/ meta/ meta/ | +7 | 0 replies | |
| 2/2 | Improve native reproducibility in recipesmeta/ meta/ meta/ +6 more | +61 | 0 replies |
Reviews trailers per patch
No Reviewed-by, Acked-by or Tested-by trailers for v4 yet.
Cover letter [RFC PATCH v4 0/2] Improve native/cross reproducibility
This patch series is not intended for merge. I only send it out to highlight where the problems are and to get some discussion going on how/if we want to improve the situation.
This is a patch series that tries to improve the reproducibility of the native/cross binaries when building in different directories. Whether this has any real benefits I'm still not sure of, but I thought I'd share this final version of the patch series in case someone comes up with an idea. This has been tested on a Fedora 35 system which uses gcc 11.2.1 at the time of writing.
The RUNTIME hack is questionable, maybe there is a better way to enforce a fixed RUNTIME entry size during linking. The build system does not do any additional rpath manipulation for target binaries, so in those cases this is not a problem. I did ask for some guidance on the gcc-help mailing list [1] on how to reserve a specific RPATH size during linking, but it would seem that is not currently possible.
If in the end we do come up with a solution, then it should be tested on the autobuilders, since otherwise this is going to degrade overtime. It would then be important that the build paths are of significantly different lengths. TMPDIR=/tmp/sysrootA/ and TMPDIR=/tmp/sysrootB/ will *not* uncover all rpath problems.
Another thing that would be useful is if buildhistory could start monitoring the depsig.do_populate_sysroot files, since that would highlight changes in the actual file contents, which is currently not covered by buildhistory.
The end result of this patch series is that I can build python3-native in two different build paths, and the resuling sysroot-components/x86_64/ directories are identical, except for the 'fixmepath.cmd' files, which are not included in the hash equiv. comparisons. Even so, there remains a lot of native builds that are going to need to be fixed in similar ways as the ones in this patch series.
Changes in v2:
* Fixed building icedtea7-native/openjdk-8-native by prepending a '/' to the rpath padding
* Squashed all the recipe-fixes together in one commit for now
Changes in v3:
* Rebase on top of master
* Instead of adding a padding entry to rpath, set phony rpaths in bitbake.conf and add patches for unbreaking python3-native and kernel
* With the changes above we don't need to remove the build ID, so add it back
Changes in v4:
* Rebase on top of master
* Cleanup fixes for openssl/ncurses/perlcross/perl
* Go back to padding rpath and revert the python3-native/kernel fixes, since this feels like the least intrusive change, and from what I understand cmake also does a similar type of rpath padding in some situations
Previous versions of the patch series:
v1: https://lists.openembedded.org/g/openembedded-core/message/158867
v2: https://lists.openembedded.org/g/openembedded-core/message/158998
v3: https://lists.openembedded.org/g/openembedded-core/message/159181
[1] https://gcc.gnu.org/pipermail/gcc-help/2021-November/140920.html
Jacob Kroon (2):
bitbake.conf: Pad rpath in native binaries
Improve native reproducibility in recipes
meta/classes/chrpath.bbclass | 3 +++ meta/classes/sstate.bbclass | 1 + meta/conf/bitbake.conf | 3 +++ ...sysroot-and-debug-prefix-map-from-co.patch | 24 ++++++++++++------- .../openssl/openssl_3.0.0.bb | 12 ++++++---- meta/recipes-core/ncurses/ncurses.inc | 9 +++++-- .../util-linux/util-linux_2.37.2.bb | 2 +- .../libtool/libtool-native_2.4.6.bb | 1 + meta/recipes-devtools/perl/perl_5.34.0.bb | 3 +++ .../pkgconfig/pkgconfig_git.bb | 1 + .../python/python3/determinism.patch | 17 +++++++++++++ .../recipes-devtools/python/python3_3.10.1.bb | 8 +++++++ 12 files changed, 68 insertions(+), 16 deletions(-) create mode 100644 meta/recipes-devtools/python/python3/determinism.patch