create-spdx and sstate
Hi Joshua / all
We've been having an issue with the create-spdx class if we share sstate between two configurations - one where gcc-cross-<arch> has a dependency and one where it doesn't (specifically, one where the abicheck class in meta- binaryaudit is inherited and the other where it isn't; that influences DEPENDS). The result is that if you build the configuration with the dependency then the one where it doesn't (in separate build dirs with the same sstate cache), image_combine_spdx fails because it can't find the SPDX data file for the dependency as it was not built in the second configuration.
It seems that create-spdx looks at BB_TASKDEPDATA to get dependencies and then adds BB_TASKDEPDATA to vardepsexclude, thus the dependencies changing does not cause the task to be re-executed. However, I assume a variable dependency on BB_TASKDEPDATA might be impractical, thus why it was excluded in the first place. Do we instead add an explicit dependency on DEPENDS? I'm happy to come up with a patch if we can determine what the correct fix is.
(FWIW we're still using dunfell, but I don't see any changes in master that alter this particular behaviour.)
Thanks
Paul
Quoted 12 lines · Paul Eggleton
Hi Joshua / all
We've been having an issue with the create-spdx class if we share sstate between two configurations - one where gcc-cross-<arch> has a dependency and one where it doesn't (specifically, one where the abicheck class in meta- binaryaudit is inherited and the other where it isn't; that influences DEPENDS). The result is that if you build the configuration with the dependency then the one where it doesn't (in separate build dirs with the same sstate cache),
image_combine_spdxfails because it can't find the SPDX data file for the dependency as it was not built in the second configuration.
Quoted 6 lines
It seems that create-spdx looks at
BB_TASKDEPDATAto get dependencies and then addsBB_TASKDEPDATAto vardepsexclude, thus the dependencies changing does not cause the task to be re-executed. However, I assume a variable dependency onBB_TASKDEPDATAmight be impractical, thus why it was excluded in the first place. Do we instead add an explicit dependency on DEPENDS? I'm happy to come up with a patch if we can determine what the correct fix is.
I think that sounds correct. The class does do_create_spdx[deptask] = "do_create_spdx" to ensure it picks up all the dependencies from DEPENDS, it would make sense that it should also have DEPENDS as a dependent variable so it reruns.
It is a little weird that none of the tasks upstream of do_create_spdx changing from that class are causing do_create_spdx to be invalidated and re-run.... but maybe that class is doing something a little less normal
Quoted 8 lines
(FWIW we're still using dunfell, but I don't see any changes in master that alter this particular behaviour.)
Thanks
Paul