| Message ID | 20230713110244.3169513-2-jose.quaresma@foundries.io |
|---|---|
| State | New |
| Headers | show |
| Series | [1/2] sstatesig: be more precise and show the full path in exceptions | expand |
On Thu, 2023-07-13 at 11:02 +0000, Jose Quaresma wrote: > sstate_report_unihash function from sstate.bbclass needs to > run with pseudo disbaled when running do_image_complete task. > This is because it have the deploy content with uid and gid > of the user that run bitbake on the build machine. > > By default the creation of sstate is skiped in the task do_image_complete > in the image.bbclass using SSTATE_SKIP_CREATION:task-image-complete=1 > so the issue is not triggered. > > There are some situation when is useful to store the image artifacts on the > sstate cache for future reuse and speedup the build without need to pruduce > the same image again. These cases are especially useful in the case of the > initramfs images used by the kernel. > > The issue can be reproduced in oe-core with: > echo "SSTATE_SKIP_CREATION:task-image-complete = '0'" >> meta/recipes-core/images/core-image-minimal-initramfs.bb > bitbake core-image-minimal-initramfs -C do_image_complete I can see why you've done this but I have a lot of worries about adding code for use cases we don't support. If we do this in too many places the code just becomes unmaintainable as we have no idea how it is being used. I'll merge patches which break things, then we'll get patches to fix that and break something else someone relies upon and things will turn into a mess. I really don't know what to do with such patches :( Cheers, Richard
Richard Purdie <richard.purdie@linuxfoundation.org> escreveu no dia quinta, 13/07/2023 à(s) 12:11: > On Thu, 2023-07-13 at 11:02 +0000, Jose Quaresma wrote: > > sstate_report_unihash function from sstate.bbclass needs to > > run with pseudo disbaled when running do_image_complete task. > > This is because it have the deploy content with uid and gid > > of the user that run bitbake on the build machine. > > > > By default the creation of sstate is skiped in the task do_image_complete > > in the image.bbclass using SSTATE_SKIP_CREATION:task-image-complete=1 > > so the issue is not triggered. > > > > There are some situation when is useful to store the image artifacts on > the > > sstate cache for future reuse and speedup the build without need to > pruduce > > the same image again. These cases are especially useful in the case of > the > > initramfs images used by the kernel. > > > > The issue can be reproduced in oe-core with: > > echo "SSTATE_SKIP_CREATION:task-image-complete = '0'" >> > meta/recipes-core/images/core-image-minimal-initramfs.bb > > bitbake core-image-minimal-initramfs -C do_image_complete > > I can see why you've done this but I have a lot of worries about adding > code for use cases we don't support. > > If we do this in too many places the code just becomes unmaintainable > as we have no idea how it is being used. I'll merge patches which break > things, then we'll get patches to fix that and break something else > someone relies upon and things will turn into a mess. > > I really don't know what to do with such patches :( > I can try another approaches to not change the default behavior: Inside the python sstate_report_unihash of the sstate bbclass I can get the SSTATE_SKIP_CREATION and if it is not set (disabled) and the task is do_image_complete we explicitly set the PSEUDO_DISABLED to run the report_unihash. Is something like that more acceptable? or is better to keep things like that in distro layers. Thanks for your time. Jose > Cheers, > > Richard > > > > >
diff --git a/meta/lib/oe/sstatesig.py b/meta/lib/oe/sstatesig.py index 4a886215b2..cbf551889c 100644 --- a/meta/lib/oe/sstatesig.py +++ b/meta/lib/oe/sstatesig.py @@ -503,7 +503,7 @@ def OEOuthashBasic(path, sigfile, task, d): corebase = d.getVar("COREBASE") tmpdir = d.getVar("TMPDIR") include_owners = os.environ.get('PSEUDO_DISABLED') == '0' - if "package_write_" in task or task == "package_qa": + if "package_write_" in task or task == "package_qa" or task == "image_complete": include_owners = False include_timestamps = False include_root = True
sstate_report_unihash function from sstate.bbclass needs to run with pseudo disbaled when running do_image_complete task. This is because it have the deploy content with uid and gid of the user that run bitbake on the build machine. By default the creation of sstate is skiped in the task do_image_complete in the image.bbclass using SSTATE_SKIP_CREATION:task-image-complete=1 so the issue is not triggered. There are some situation when is useful to store the image artifacts on the sstate cache for future reuse and speedup the build without need to pruduce the same image again. These cases are especially useful in the case of the initramfs images used by the kernel. The issue can be reproduced in oe-core with: echo "SSTATE_SKIP_CREATION:task-image-complete = '0'" >> meta/recipes-core/images/core-image-minimal-initramfs.bb bitbake core-image-minimal-initramfs -C do_image_complete Signed-off-by: Jose Quaresma <jose.quaresma@foundries.io> --- meta/lib/oe/sstatesig.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)