Message ID | 20230822203350.264625-1-paulo@myneves.com |
---|---|
State | Accepted, archived |
Commit | 5293a1b36eeb89f57577cb709ec7f293909039a1 |
Headers | show |
Series | [1/1] siggen.py: Improve taskhash reproducibility | expand |
Hello Paulo, On Tue, 22 Aug 2023 20:33:58 +0000 "Paulo Neves" <paulo@myneves.com> wrote: > file checksums are part of the data checksummed > to generate the task hash. The list of file checksums > was not ordered. > > In this commit we make sure the task hash checksum takes > a list of checksum data that is ordered by unique file name > thus guaranteeing reproducibility. > > Signed-off-by: Paulo Neves <paulo@myneves.com> Lots of errors like the following on the autobuilders are possibly caused by this patch: Hash for task dependency pseudo-native:do_populate_sysroot changed from f4435f759ac00c51c07eaa840d9d61249f38f82d39a8dbffa4e8f599d874a3c6 to 07a200198639768d86d5c2c654fcbd0d4ea23d333e735848ac22c12474db1c2d (and a lot more errors) Logs: https://autobuilder.yoctoproject.org/typhoon/#/builders/80/builds/5592/steps/14/logs/stdio https://autobuilder.yoctoproject.org/typhoon/#/builders/79/builds/5646/steps/14/logs/stdio https://autobuilder.yoctoproject.org/typhoon/#/builders/127/builds/1954/steps/15/logs/stdio Luca
On 24/08/2023 09:58, Luca Ceresoli wrote: > Hello Paulo, > > On Tue, 22 Aug 2023 20:33:58 +0000 > "Paulo Neves" <paulo@myneves.com> wrote: > >> file checksums are part of the data checksummed >> to generate the task hash. The list of file checksums >> was not ordered. >> >> In this commit we make sure the task hash checksum takes >> a list of checksum data that is ordered by unique file name >> thus guaranteeing reproducibility. >> >> Signed-off-by: Paulo Neves <paulo@myneves.com> > Lots of errors like the following on the autobuilders are possibly > caused by this patch: > > Hash for task dependency pseudo-native:do_populate_sysroot changed from f4435f759ac00c51c07eaa840d9d61249f38f82d39a8dbffa4e8f599d874a3c6 to 07a200198639768d86d5c2c654fcbd0d4ea23d333e735848ac22c12474db1c2d > (and a lot more errors) > > Logs: > > https://autobuilder.yoctoproject.org/typhoon/#/builders/80/builds/5592/steps/14/logs/stdio > https://autobuilder.yoctoproject.org/typhoon/#/builders/79/builds/5646/steps/14/logs/stdio > https://autobuilder.yoctoproject.org/typhoon/#/builders/127/builds/1954/steps/15/logs/stdio > > Luca > > -- > Luca Ceresoli, Bootlin > Embedded Linux and Kernel engineering > https://bootlin.com Thank you! I will have a look. Can you tell me how to run the test case which is showing this? I think this may be related to me having a different signature handler than the ones on the test.
On Thu, 2023-08-24 at 09:58 +0200, Luca Ceresoli via lists.openembedded.org wrote: > Hello Paulo, > > On Tue, 22 Aug 2023 20:33:58 +0000 > "Paulo Neves" <paulo@myneves.com> wrote: > > > file checksums are part of the data checksummed > > to generate the task hash. The list of file checksums > > was not ordered. > > > > In this commit we make sure the task hash checksum takes > > a list of checksum data that is ordered by unique file name > > thus guaranteeing reproducibility. > > > > Signed-off-by: Paulo Neves <paulo@myneves.com> > > Lots of errors like the following on the autobuilders are possibly > caused by this patch: > > Hash for task dependency pseudo-native:do_populate_sysroot changed from f4435f759ac00c51c07eaa840d9d61249f38f82d39a8dbffa4e8f599d874a3c6 to 07a200198639768d86d5c2c654fcbd0d4ea23d333e735848ac22c12474db1c2d > (and a lot more errors) > > Logs: > > https://autobuilder.yoctoproject.org/typhoon/#/builders/80/builds/5592/steps/14/logs/stdio > https://autobuilder.yoctoproject.org/typhoon/#/builders/79/builds/5646/steps/14/logs/stdio > https://autobuilder.yoctoproject.org/typhoon/#/builders/127/builds/1954/steps/15/logs/stdio FWIW I had this patch in master-next and didn't see those issues so it is possibly a different change... Cheers, Richard
diff --git a/lib/bb/siggen.py b/lib/bb/siggen.py index 879c136e1..b023b79ec 100644 --- a/lib/bb/siggen.py +++ b/lib/bb/siggen.py @@ -361,7 +361,7 @@ class SignatureGeneratorBasic(SignatureGenerator): for dep in sorted(self.runtaskdeps[tid]): data += self.get_unihash(dep[1]) - for (f, cs) in self.file_checksum_values[tid]: + for (f, cs) in sorted(self.file_checksum_values[tid], key=clean_checksum_file_path): if cs: if "/./" in f: data += "./" + f.split("/./")[1] @@ -426,7 +426,7 @@ class SignatureGeneratorBasic(SignatureGenerator): if runtime and tid in self.taskhash: data['runtaskdeps'] = [dep[0] for dep in sorted(self.runtaskdeps[tid])] data['file_checksum_values'] = [] - for f,cs in self.file_checksum_values[tid]: + for f,cs in sorted(self.file_checksum_values[tid], key=clean_checksum_file_path): if "/./" in f: data['file_checksum_values'].append(("./" + f.split("/./")[1], cs)) else: @@ -745,6 +745,12 @@ class SignatureGeneratorTestEquivHash(SignatureGeneratorUniHashMixIn, SignatureG self.server = data.getVar('BB_HASHSERVE') self.method = "sstate_output_hash" +def clean_checksum_file_path(file_checksum_tuple): + f, cs = file_checksum_tuple + if "/./" in f: + return "./" + f.split("/./")[1] + return f + def dump_this_task(outfile, d): import bb.parse mcfn = d.getVar("BB_FILENAME")
file checksums are part of the data checksummed to generate the task hash. The list of file checksums was not ordered. In this commit we make sure the task hash checksum takes a list of checksum data that is ordered by unique file name thus guaranteeing reproducibility. Signed-off-by: Paulo Neves <paulo@myneves.com> --- lib/bb/siggen.py | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-)