Skip to contentSkip to conversation

RFC: systemd version - release or git?

DISCUSSION Richard Purdie 4 participants 6 messages Feb 19, 2014 – Feb 20, 2014

lore
Share & tools
lore
#0 lore raw

As anyone looking at the autobuidler will know, we've been having a lot of issues with instability with systemd tests under qemu.

I believe these are real world issues, most likely races in systemd itself.

The last release of systemd, 208 was five months ago. There have been a ton of changes since including many bug fixes.

Fedora 20 uses systemd 208 (the last release) with 200+ patches on top.

I'm therefore wondering if we shouldn't bite the bullet and add a git recipe rather than trying to patch 208 into some form of life.

The proposal is therefore to add a more recent git recipe and make that the default.

I have such a patch in master-next. I've debugged the toolchain issues, missing kernel config and other issues with systemd master, it always now seems to pass the autobuilder tests. I need to check on one remaining automated QA test result to ensure its not a regression in systemd.

Any objections to defaulting to a git version?

Cheers,

Richard

#1 lore raw
Quoted 17 lines · Richard Purdie

As anyone looking at the autobuidler will know, we've been having a lot of issues with instability with systemd tests under qemu.

I believe these are real world issues, most likely races in systemd itself.

The last release of systemd, 208 was five months ago. There have been a ton of changes since including many bug fixes.

Fedora 20 uses systemd 208 (the last release) with 200+ patches on top.

I'm therefore wondering if we shouldn't bite the bullet and add a git recipe rather than trying to patch 208 into some form of life.

The proposal is therefore to add a more recent git recipe and make that the default.

Quoted 7 lines

I have such a patch in master-next. I've debugged the toolchain issues, missing kernel config and other issues with systemd master, it always now seems to pass the autobuilder tests. I need to check on one remaining automated QA test result to ensure its not a regression in systemd.

Any objections to defaulting to a git version?

I think we can wait few more days for new release:
https://www.mail-archive.com/[email protected]/msg16738.html

but otherwise I don't have strong objections for switching to git version temporary (if it really makes QA test results better).

Cheers,

Signature · 1 line
Martin 'JaMa' Jansa     jabber: [email protected]
#5 lore raw
In reply to Richard Purdie

Op 19 feb. 2014, om 11:54 heeft Richard Purdie <[email protected]> het volgende geschreven:

Quoted 2 lines

As anyone looking at the autobuidler will know, we've been having a lot of issues with instability with systemd tests under qemu.

A number of those were (are?) caused by kernel config issues, like CONFIG_FHANDLE missing, have those been addressed?

Quoted 14 lines

I believe these are real world issues, most likely races in systemd itself.

The last release of systemd, 208 was five months ago. There have been a ton of changes since including many bug fixes.

Fedora 20 uses systemd 208 (the last release) with 200+ patches on top.

I'm therefore wondering if we shouldn't bite the bullet and add a git recipe rather than trying to patch 208 into some form of life.

The proposal is therefore to add a more recent git recipe and make that the default.

Quoted 7 lines

I have such a patch in master-next. I've debugged the toolchain issues, missing kernel config and other issues with systemd master, it always now seems to pass the autobuilder tests. I need to check on one remaining automated QA test result to ensure its not a regression in systemd.

Any objections to defaulting to a git version?

In meta-oe we used to build systemd from a git tag instead of tarball release and ran into the usual gtk-doc and docbook problems. Can you build systemd git without docbook dtds installed on the buildhost?

If the docbook issues have been fixed I have no objection to adding a git recipe. I poked Lennart at FOSDEM about v209 and he said they need to fix some issues in kdbus and networkd first before making a release, "hopefully soon".

Speaking of networkd, any idea if our connman is recent enough to support networkd -> connman handover?

regards,

Koen

#2 lore raw
In reply to Martin Jansa
Quoted 5 lines · Martin Jansa
Quoted 1 line · Richard Purdie

Any objections to defaulting to a git version?

I think we can wait few more days for new release:
https://www.mail-archive.com/[email protected]/msg16738.html

I was just about to reply with the same e-mail link...

Quoted 2 lines

but otherwise I don't have strong objections for switching to git version temporary (if it really makes QA test results better).

Otherwise, I don't have any strong feelings against a git version. Not as long as it works as it should.

One objection might be that it's easier for someone to update it to a strange version in between the regular releases. As long as we try to avoid making a release where systemd is too far of from a regular release, that shouldn't be any issue, though.

Cheers,
Anders

Signature · 2 lines
Anders Darander
ChargeStorm AB / eStorm AB
#3 lore raw
Quoted 8 lines · Anders Darander
Quoted 5 lines · Martin Jansa
Quoted 1 line · Richard Purdie

Any objections to defaulting to a git version?

I think we can wait few more days for new release:
https://www.mail-archive.com/[email protected]/msg16738.html

I was just about to reply with the same e-mail link...

I asked about this a few days ago and scanned the archives and didn't find it. Typically, they sent it out after I did that :)

Quoted 10 lines
Quoted 2 lines

but otherwise I don't have strong objections for switching to git version temporary (if it really makes QA test results better).

Otherwise, I don't have any strong feelings against a git version. Not as long as it works as it should.

One objection might be that it's easier for someone to update it to a strange version in between the regular releases. As long as we try to avoid making a release where systemd is too far of from a regular release, that shouldn't be any issue, though.

I agree its worth waiting for the release if it really is than imminent. I'd previously not been able to find a commitment or schedule.

It might be useful to have a git based recipe anyway for debugging/testing...

Cheers,

Richard

#4 lore raw

Op 19 feb. 2014, om 16:47 heeft Richard Purdie <[email protected]> het volgende geschreven:

Quoted 5 lines · Anders Darander
Quoted 4 lines · Martin Jansa
Quoted 2 lines · Richard Purdie
Quoted 1 line

Any objections to defaulting to a git version?

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

I think we can wait few more days for new release:
https://www.mail-archive.com/[email protected]/msg16738.html

I was just about to reply with the same e-mail link...

I asked about this a few days ago and scanned the archives and didn't find it. Typically, they sent it out after I did that :)

And it has been released!

Quoted 27 lines
Quoted 11 lines
Quoted 3 lines

but otherwise I don't have strong objections for switching to git version temporary (if it really makes QA test results better).

Otherwise, I don't have any strong feelings against a git version. Not as long as it works as it should.

One objection might be that it's easier for someone to update it to a strange version in between the regular releases. As long as we try to avoid making a release where systemd is too far of from a regular release, that shouldn't be any issue, though.

I agree its worth waiting for the release if it really is than imminent. I'd previously not been able to find a commitment or schedule.

It might be useful to have a git based recipe anyway for debugging/testing...

Cheers,

Richard

_______________________________________________
Openembedded-core mailing list
[email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-core