[bitbake-devel] [PATCH 00/17][PULL] Hob: Bunch of bug fixes

[Richard Purdie]

On Thu, 2012-03-22 at 15:32 -0700, Joshua Lock wrote:
> On 21/03/12 18:21, Xu, Dongxiao wrote:
> > On Wed, 2012-03-21 at 17:59 -0700, Joshua Lock wrote:
> >>
> >> On 21/03/12 05:55, Dongxiao Xu wrote:
> >>> Hi Richard,
> >>>
> >>> This pull request is Hob related bug fixes. Please help to review and pull.
> >>>
> >>> Thanks,
> >>> Dongxiao
> >>>
> >>> The following changes since commit d595960fea0988df9004d927bc2ec3439540dd9c:
> >>>
> >>>     Hob: save CONF_VERSION and LCONF_VERSION into template (2012-03-20 14:39:45 +0000)
> >>>
> >>> are available in the git repository at:
> >>>     git://git.pokylinux.org/poky-contrib dxu4/hob-bugfix
> >>>     http://git.pokylinux.org/cgit.cgi/poky-contrib/log/?h=dxu4/hob-bugfix
> >>>
> >>> Dongxiao Xu (14):
> >>>     Hob: Still use initcmd != None as the if judgement
> >>>     Hob: Remove split model in GTK Hob
> >>>     cooker: terminate each process when quitting recipe parsing
> >>>     Hob: change package classes selection GUI
> >>>     Hob: Change template button name from "Template" to "Templates"
> >>
> >> These all look fine.
> >>
> >>>     command.py: Change parseConfigurationFiles API from async to sync
> >>>     process.py: Increase the timeout value for polling commands
> >>>     Hob: Change parseConfigurationFiles API usage
> >>
> >> Can you explain some of the reasoning behind this set of changes? If the
> >> command truly belongs as an asynchronous one I'm not sure why we need to
> >> increase the timeout duration specifically for it.
> >>
> >>>     Hob: Fix the workaround to get image types
> >
> > Hi Josh,
> >
> > Actually the above three patches are preparations for this commit: "Hob:
> > Fix the workaround to get image types".
> >
> > The background is that, we need to add extra inherits of
> > image_types.bbclass before parsing configuration files and getting
> > variables values. Therefore the modified code piece is something like:
> >
> > def get_parameters(self):
> >      self.init_cooker()
> >      self.set_extra_inherits("image_types")
> >      self.parse_config()
> >
> >      self.server.runCommand(["getVariable", "MACHINE"])
> >      ...
> >
> > We can see from above code is that, the parse_config locates between two
> > SYNC commands (set_extra_inherits and getVariable). Therefore we also
> > need the parse_config to be a SYNC command.
> >
> > As far as I know the parseConfigurationFiles API doesn't cost too much
> > time (should less than 1s), so that's the reason why I change it to SYNC
> > mode.
> >
> > The increase of time duration for parseConfigurationFiles is that, the
> > process server use pipe mode to pass object (function) from one to
> > another. The original 0.5 second timeout isn't enough for transferring
> > "parseConfigurationFiles" object, the phenomenon is that on certain
> > machines, the runCommand result is messed up with wrong values. This
> > issue doesn't exist with xmlrpc server.
> 
> OK, I understand what you're trying to achieve. I'm afraid I'm not 
> convinced that we need to make such an invasive change though. Any 
> reason we can't re-order the initialisation so that everything is done 
> after the async parseConfigurationFiles?
> 
> I'm nervous of us pushing the 0.5s->1s change this late in the cycle. 
> Have you any indication of whether it affects builds using knotty?
> 
> Just to be clear I think Richard needs to ack/nack this change to core 
> BitBake. The Hob changes related to it are fine by me.

Having a sync command blocking the system for over 1 second isn't right.
Sync commands are meant to be fast (say 0.01s max) and this obviously
isn't.

I'm rather concerned about this and it suggests something is wrong with
hob's init/state code to be honest.

My suggestion would be to add an extra parameter to self.parse_config()
so you can set other classes to inherit (and variables to set). You will
still have to then read MACHINE later but I think that is reasonable.

Cheers,

Richard