I'm planning to do a partial mass-rebuild for Python 2.7.
This would cover all Python 2 users within the distribution, roughly 1000 src.rpms.
Some notes can be seen at: https://fedoraproject.org/wiki/Features/Python_2.7/MassRebuild
I hoped to start this tomorrow (2010-07-21) at 16:00 UTC, but it looks like I need to wait until at least after the outage that's due for that time [1]. Further, I'm still waiting to hear back to see if the build should be done with the "gold" linker, rather than "gnu-ld" [2]. I will delay the build until after a decision is made on this.
This may mean that I'll wait until 2010-07-22.
Some areas of uncertainty remain: - the script to do this, as written, assumes no build ordering is needed. I know that some ordering is needed, and plan to manually order the first four builds which I hope is the bulk of it, but it's not clear to me if more is needed; I'm running some tests to try to better determine this.
Hope this makes sense Dave
[1] https://fedorahosted.org/fedora-infrastructure/ticket/2280 [2] https://fedoraproject.org/wiki/Features/GoldLinkerDefault
On Tuesday, July 20, 2010 07:02:13 pm David Malcolm wrote:
I'm planning to do a partial mass-rebuild for Python 2.7.
This would cover all Python 2 users within the distribution, roughly 1000 src.rpms.
Some notes can be seen at: https://fedoraproject.org/wiki/Features/Python_2.7/MassRebuild
I hoped to start this tomorrow (2010-07-21) at 16:00 UTC, but it looks like I need to wait until at least after the outage that's due for that time [1]. Further, I'm still waiting to hear back to see if the build should be done with the "gold" linker, rather than "gnu-ld" [2]. I will delay the build until after a decision is made on this.
This may mean that I'll wait until 2010-07-22.
Some areas of uncertainty remain:
- the script to do this, as written, assumes no build ordering is
needed. I know that some ordering is needed, and plan to manually order the first four builds which I hope is the bulk of it, but it's not clear to me if more is needed; I'm running some tests to try to better determine this.
Hope this makes sense Dave
[1] https://fedorahosted.org/fedora-infrastructure/ticket/2280 [2] https://fedoraproject.org/wiki/Features/GoldLinkerDefault
Dave,
Please just do the building, we currently have 19 x86_64 builders. builder cpu time really is not an issue here. we can untag any wasteful rebuild so that it gets garbage collected. if we change the distro default linker and it needs a rebuild then it will be done then. at this point we are best served by just getting things done.
Dennis
On Tue, 2010-07-20 at 19:14 -0500, Dennis Gilmore wrote:
On Tuesday, July 20, 2010 07:02:13 pm David Malcolm wrote:
I'm planning to do a partial mass-rebuild for Python 2.7.
This would cover all Python 2 users within the distribution, roughly 1000 src.rpms.
Some notes can be seen at: https://fedoraproject.org/wiki/Features/Python_2.7/MassRebuild
I hoped to start this tomorrow (2010-07-21) at 16:00 UTC, but it looks like I need to wait until at least after the outage that's due for that time [1]. Further, I'm still waiting to hear back to see if the build should be done with the "gold" linker, rather than "gnu-ld" [2]. I will delay the build until after a decision is made on this.
This may mean that I'll wait until 2010-07-22.
Some areas of uncertainty remain:
- the script to do this, as written, assumes no build ordering is
needed. I know that some ordering is needed, and plan to manually order the first four builds which I hope is the bulk of it, but it's not clear to me if more is needed; I'm running some tests to try to better determine this.
Hope this makes sense Dave
[1] https://fedorahosted.org/fedora-infrastructure/ticket/2280 [2] https://fedoraproject.org/wiki/Features/GoldLinkerDefault
Dave,
Please just do the building, we currently have 19 x86_64 builders. builder cpu time really is not an issue here. we can untag any wasteful rebuild so that it gets garbage collected. if we change the distro default linker and it needs a rebuild then it will be done then. at this point we are best served by just getting things done.
Dennis
[CCing Jeff Law for "gold" input (feature owner)]
I was asked at the FESCo meeting to wait for the python 2.7 rebuild until a decision is made on using gold or not, and how to turn it on in the buildroots: http://meetbot.fedoraproject.org/fedora-meeting/2010-07-20/fesco.2010-07-20-...
On rereading the logs it's not at all clear who was tasked with chasing up the "gold" feature (hence CCing Jeff Law).
My feelings here are that getting things done is good, and that builds are cheaper than contributor time, hence multiple rebuilds serving different purposes aren't a serious waste compared to risks of schedule slips. Plus I'd prefer to isolate possible breakage from python 2.6 -> 2.7 from possible breakage due to ld->gold: changing a single variable at a time is good for my sanity!
Dave
On Wednesday, July 21, 2010 11:15:43 am David Malcolm wrote:
On Tue, 2010-07-20 at 19:14 -0500, Dennis Gilmore wrote:
On Tuesday, July 20, 2010 07:02:13 pm David Malcolm wrote:
I'm planning to do a partial mass-rebuild for Python 2.7.
This would cover all Python 2 users within the distribution, roughly 1000 src.rpms.
Some notes can be seen at: https://fedoraproject.org/wiki/Features/Python_2.7/MassRebuild
I hoped to start this tomorrow (2010-07-21) at 16:00 UTC, but it looks like I need to wait until at least after the outage that's due for that time [1]. Further, I'm still waiting to hear back to see if the build should be done with the "gold" linker, rather than "gnu-ld" [2]. I will delay the build until after a decision is made on this.
This may mean that I'll wait until 2010-07-22.
Some areas of uncertainty remain:
- the script to do this, as written, assumes no build ordering is
needed. I know that some ordering is needed, and plan to manually order the first four builds which I hope is the bulk of it, but it's not clear to me if more is needed; I'm running some tests to try to better determine this.
Hope this makes sense Dave
[1] https://fedorahosted.org/fedora-infrastructure/ticket/2280 [2] https://fedoraproject.org/wiki/Features/GoldLinkerDefault
Dave,
Please just do the building, we currently have 19 x86_64 builders. builder cpu time really is not an issue here. we can untag any wasteful rebuild so that it gets garbage collected. if we change the distro default linker and it needs a rebuild then it will be done then. at this point we are best served by just getting things done.
Dennis
[CCing Jeff Law for "gold" input (feature owner)]
I was asked at the FESCo meeting to wait for the python 2.7 rebuild until a decision is made on using gold or not, and how to turn it on in the buildroots: http://meetbot.fedoraproject.org/fedora-meeting/2010-07-20/fesco.2010-07-20 -19.30.log.html#l-429
On rereading the logs it's not at all clear who was tasked with chasing up the "gold" feature (hence CCing Jeff Law).
My feelings here are that getting things done is good, and that builds are cheaper than contributor time, hence multiple rebuilds serving different purposes aren't a serious waste compared to risks of schedule slips. Plus I'd prefer to isolate possible breakage from python 2.6 -> 2.7 from possible breakage due to ld->gold: changing a single variable at a time is good for my sanity!
Dave
indeed once cvs is updated and rebooted please proceed.
Dennis
On Tue, 2010-07-20 at 20:02 -0400, David Malcolm wrote:
I'm planning to do a partial mass-rebuild for Python 2.7.
This would cover all Python 2 users within the distribution, roughly 1000 src.rpms.
Some notes can be seen at: https://fedoraproject.org/wiki/Features/Python_2.7/MassRebuild
[snip]
Status: IN PROGRESS - IRC chat with nirik, pjones and others confirmed not to wait for gold - the infrastructure outage took somewhat longer that expected, so I didn't start until about 6pm EDT iirc. - The mass-rebuild script is running, and has done about 600 out of 1000 packages. All of this is into dist-f14-py27-rebuild, so it won't hit rawhide until rel-eng push it there. - We're at a roughly 50:50 success/fail rate - build ordering was much more important than I hoped; most of the failures in this run seem to be due to incomplete deps in root.log. I intend to retry these from the now-existing CVS tags, with a better build ordering (but I need to sleep first). - numpy is segfaulting during %check; am waiting on a gdb build to finish (linked against 2.7) before I debug; this blocks pygtk2 which blocks various things - I've applied some fixes directly to some packages to get things to build. I hope my changes are acceptable. - Sorry about all of the email notification spam some of you will have received
Notes: - There was a snag with python-nose: I needed to cherrypick 2.7 compat fixes from upstream: DONE - There are some loops in the dep graph: python-nose and python-jinja2 both BR: python-sphinx but are in the BR for that package. I've turned off doc generation for these for now; am waiting on repo to rebuild before I can build python-sphinx (deps appear to be ready).
[snip]
Hope this makes sense Dave
On Thu, 22 Jul 2010, David Malcolm wrote:
On Tue, 2010-07-20 at 20:02 -0400, David Malcolm wrote:
I'm planning to do a partial mass-rebuild for Python 2.7.
This would cover all Python 2 users within the distribution, roughly 1000 src.rpms.
Some notes can be seen at: https://fedoraproject.org/wiki/Features/Python_2.7/MassRebuild
[snip]
Status: IN PROGRESS
- IRC chat with nirik, pjones and others confirmed not to wait for gold
- the infrastructure outage took somewhat longer that expected, so I
didn't start until about 6pm EDT iirc.
- The mass-rebuild script is running, and has done about 600 out of 1000
packages. All of this is into dist-f14-py27-rebuild, so it won't hit rawhide until rel-eng push it there.
- We're at a roughly 50:50 success/fail rate
- build ordering was much more important than I hoped; most of the
failures in this run seem to be due to incomplete deps in root.log. I intend to retry these from the now-existing CVS tags, with a better build ordering (but I need to sleep first).
- numpy is segfaulting during %check; am waiting on a gdb build to
finish (linked against 2.7) before I debug; this blocks pygtk2 which blocks various things
- I've applied some fixes directly to some packages to get things to
build. I hope my changes are acceptable.
- Sorry about all of the email notification spam some of you will have
received
Notes:
- There was a snag with python-nose: I needed to cherrypick 2.7 compat
fixes from upstream: DONE
- There are some loops in the dep graph: python-nose and python-jinja2
both BR: python-sphinx but are in the BR for that package. I've turned off doc generation for these for now; am waiting on repo to rebuild before I can build python-sphinx (deps appear to be ready).
If you want some help open up an FES ticket, I'll see what we can do:
https://fedorahosted.org/fedora-engineering-services/
A list of failing builds would be good.
-Mike
infrastructure@lists.fedoraproject.org