You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
iam working on llvm recipe #17509 which needs more ci time because it quite huge. I saw it really often that the v2 pipeline is broken before v1 is even done.
In such a case no ci report is generated, so its hard to tell what actually happens. I would guess that v1 need to be done within 12 hours.
build time is 600 min and there are large projects like boost/qt which are allowed to take 800 minutes
I'm not sure if now a report is generated in such case?
Because of this limitation I needed to reduce the ci enabled configurations.
Debug builds are disabled
Shared builds are disabled
Only latest patch version is enabled
Its a decision for this case but official recommendation would be nice?
For such cases it would also be handy to disable certain configurations just for the ci. E.g. either there is a config file for it or there is an attribute we can check. I don't like the basic idea of recipes building differently in ci but maybe there is a good way of achieving it.
If you are adding e.g. one more variant for compiler, libstd on any platform to the ci or if we would like to add 1-2 llvm versions no llvm merge request will pass the pipeline without breaking existing configurations. E.g. user would need to set conan_center_index_limits=False but this also means his configuration was tested and is no longer tested in ci. Because of this the quality assurance for the recipe is very much limited.
Thanks for your time.
Further details:
There are already issues for debug builds e.g. without tweaking debug builds on ELF platforms can take up to 26gb of ram, could bring this down to max 12gb without actually forcing everyone to build it with just 1 job, which would be painfully slow. A higher memory limit for specific recipes would also be very nice, because we had no idea what the memory limit is. Even used a dummy pr to profile the consumption in ci. Or at least document the limitations somewhere available. Debug passed but still disabled to reduce ci time.
Shared builds are disabled to just reduce the amount of configurations but shared builds are even stated by llvm project itself as "llvm devs only". Still would be nice to test it, because they can also break in upcoming pr.
Only latest patch is disabled, well currently it needs already around 18 hours for 4 versions, what happens if we set it to 24? So in such cases its fine to add all versions but if not the latest patch version raise UnsupportedConfiguration("llvm version is disabled for conan center index ci because [...] You can enable it with option conan_center_index_limits=False.") But even if I disabled them v1 didn't complete in time to run v2 pipeline. So overhead between each version is also taking a lot of ci time.
It seems that Windows / Mac testing needs way longer compared to Linux.
The text was updated successfully, but these errors were encountered:
What is your problem/feature request?
Hi,
iam working on llvm recipe #17509 which needs more ci time because it quite huge. I saw it really often that the v2 pipeline is broken before v1 is even done.
If you are adding e.g. one more variant for compiler, libstd on any platform to the ci or if we would like to add 1-2 llvm versions no llvm merge request will pass the pipeline without breaking existing configurations. E.g. user would need to set conan_center_index_limits=False but this also means his configuration was tested and is no longer tested in ci. Because of this the quality assurance for the recipe is very much limited.
Thanks for your time.
Further details:
raise UnsupportedConfiguration("llvm version is disabled for conan center index ci because [...] You can enable it with option conan_center_index_limits=False.")
But even if I disabled them v1 didn't complete in time to run v2 pipeline. So overhead between each version is also taking a lot of ci time.The text was updated successfully, but these errors were encountered: