sstate mirroring between different host OSes requires pre-building pseudo-native
I run my CI autobuilders on Ubuntu 20.04 servers, which generate sstate that I mirror out to my development machine, which runs Ubuntu 20.10. That used to work fine - everything would setscene in as expected. For some time now, though, I see just about everything setscene in, and then everything also gets rebuilt from scratch again. I'm not sure if this dates back to November when the change to de-universalize pseudo-native happened, or with some more recent pseudo-related changes, as I only had time to investigate it this week.
If I bitbake pseudo-native first so it's at least in my local sstate-cache, then everything works as expected again. It's only if pseudo has to be built from scratch that I get the rebuild cascade. I thought this might be due to my distro setup, but it's easily reproducible with a stock poky setup.
Is this the expected behavior now?
Thanks,
-Matt
Quoted 9 lines · Matt Madison
I run my CI autobuilders on Ubuntu 20.04 servers, which generate sstate that I mirror out to my development machine, which runs Ubuntu 20.10. That used to work fine - everything would setscene in as expected. For some
time now, though, I see just about everything setscene in, and then everything also gets rebuilt from scratch
again. I'm not sure if this dates back to November when the change to de-universalize pseudo-native happened,
or with some more recent pseudo-related changes, as I only had time to investigate it this week.
Quoted 6 lines
If I bitbake pseudo-native first so it's at least in my local sstate-cache, then everything works as expected
again. It's only if pseudo has to be built from scratch that I get the rebuild cascade. I thought this might be due to my distro setup, but it's easily reproducible with a stock poky setup.Is this the expected behavior now?
Sadly this is expected, yes. It wasn't really intended but an unfortunately side effect. Ideally bitbake would delay the rest of sstate until pseudo-native is built but it never used to be able to do that. The code probably now can cope with it but it would need modifications to runqueue to make that happen.
Its something I'm aware of and would like to improve but the release needs priority right now.
The reason is do_package sstate must be extracted under pseudo-native. The irony is that sstate would likely be skipped anyway but it realises the "hard" dependency isn't present first :/.
Cheers,
Richard