Hi,
The problem can be very easily reproduced. Just enable the
archiver.bbclass and build openjdk-8 and it will break. I'll explain
further down why.
I would really like to have some upstream solution, so openjdk-8 can be
built with the archiver.bbclass active, but so far I only have my hacky
patch[1].
[1]
https://github.com/RobertBerger/meta-java/commit/a97d27ea0d6701df222c79fb1ce983fc39308c0c
Here I try to explain the problem:
1) Without the archiver class:
in openjdk-8-common.inc:
1.1) do_unpack[postfuncs] += "do_unpack_extract_submodules"
1.2)
do_unpack_extract_submodules () {
cd "${S}"
# tar --transform
tar xjf ${WORKDIR}/${CORBA_FILE_LOCAL} --transform
"s,-${CORBA_CHANGESET},,g"
...
}
2) With the archiver class:
2.1) do_unpack_and_patch could be used instead of do_unpack directly:
2.1.1) It is always used for ARCHIVER_MODE[src] = "patched" and
ARCHIVER_MODE[src] = "configured".
2.1.2) For ARCHIVER_MODE[src] = "original" it is only enabled if
ARCHIVER_MODE[diff] = "1" is also enabled.
2.1.3) And for the new ARCHIVE_MODE[src] = "mirror" it is never enabled.
Thanks to Saur for the detailed explanation when and when not
do_unpack_and_patch is being called.
2.2) if do_unpack_and_patch() is used it sets WORKDIR itself to
${ARCHIVER_WORKDIR}, which defaults to ${WORKDIR}/archiver-work/
2.3) Which makes
do_unpack_extract_submodules () {
cd "${S}"
# tar --transform
tar xjf ${WORKDIR}/${CORBA_FILE_LOCAL} --transform
"s,-${CORBA_CHANGESET},,g"
...
}
fail, since it should be tar xjf ${WORKDIR}/../${CORBA_FILE_LOCAL}
--transform "s,-${CORBA_CHANGESET},,g"
...
due to the additional directory structure.
I would be happy to hear suggestions how to improve my hack.
If I could somehow programatically figure out within BitBake that
do_unpack_extract_submodules is called from do_unpack_and_patch ->
do_unpack I guess it would be better. Just don't know how this can be
done. Is there a BitBake callstack?
3) My (hacky) solution
[1]
https://github.com/RobertBerger/meta-java/commit/a97d27ea0d6701df222c79fb1ce983fc39308c0c
Regards,
Robert
--
_______________________________________________
Openembedded-devel mailing list
[email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-devel
