| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
You could reproduce an error in this test with:
bitbake core-image-minimal
bitbake tzcode-native -c cleansstate
oe-selftest -r sstatetests.SStatePrintdiff.test_image_minimal_vs_base_do_configure
since tzcode-native isn't needed once tzdata is available and isn't rebuilt
by "bitbake core-image-minimal" if it is missig. tzdata is allarch so if tzdata is
built on an x86 host, a aarch64 build machine would never build tzcode-native with
this set of calls.
Add a --runall option to the initial bitbake invocation to force these things to be
created if they're missing.
This explains why some failures were occurring on the infrastructure. With that issue
fixed, drop the hash mode change since I believe this fixes that issue. That
restriction was hurting performance, this should allow sstate reuse for the test
and improve the speed of it.
(From OE-Core rev: 8f03ac39e7fe21f3d6eca35b12b203a73a15285d)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
|
'bitbake -S printdiff' is a useful diagnostic facility for finding out
why sstate is not being reused, but until now it had no tests that would
ensure it works. This commit adds three basic scenarios:
1. make a change in a really basic, common recipe that is at the very root
of dependency trees (quilt-native), and ensure that change is correctly discovered when
building an image.
2. make a change in gcc-source recipe, which is somewhat special
(operates in work-shared), and ensure that gcc-runtime builds track
that down as well.
3. make a change in base_do_configure() definition from base.bbclass,
which is not recipe-specific, but affects many basic recipes, and ensure that
is correctly reported as well.
The test itself actually runs twice:
- first against a fully populated build directory, where
the printdiff code is guaranteed to find the correct previous
stamp that can be compared with in a predictable manner.
- then in an empty build directory where the printdiff code
goes to look in the sstate cache, and so the existence of the
previous signature can be tested, but not the difference with it
(what the exact difference would be is unpredictable as the
sstate cache is indeed shared between many builds).
(From OE-Core rev: 7a7d76aa8a8d590ebc99156f9f4b9535cdf868c7)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
