[yocto] Change in override behavior?

[Paul Eggleton]

On Fri, 29 Apr 2016 12:00:58 Gary Thomas wrote:
> On 2016-04-29 11:51, Paul Eggleton wrote:
> > On Fri, 29 Apr 2016 09:24:23 Gary Thomas wrote:
> >> On 2016-04-29 09:16, Paul Eggleton wrote:
> >>> On Fri, 29 Apr 2016 08:06:16 Gary Thomas wrote:
> >>>> On 2016-04-29 07:08, Paul Eggleton wrote:
> >>>>> On Fri, 29 Apr 2016 06:24:55 Gary Thomas wrote:
> >>>>>> I used to be able to use something like this:
> >>>>>>       VARIABLE = "xyz"
> >>>>>> 
> >>>>>> and maybe in a .bbappend use
> >>>>>> 
> >>>>>>       VARIABLE_mymachine = "xyzzy"
> >>>>>> 
> >>>>>> where 'mymachine' is in OVERRIDES, which resulted in
> >>>>>> 
> >>>>>>       VARIABLE = "xyzzy"
> >>>>>> 
> >>>>>> This doesn't seem to work the same today, what I get is
> >>>>>> 
> >>>>>>       VARIABLE = "xyz"
> >>>>>>       VARIABLE_mymachine = "xyzzy"
> >>>>>> 
> >>>>>> Has this changed (recently)?  Why?  How to I make use of overrides?
> >>>>>> 
> >>>>>> Note: I'm using the latest Poky master (a9b503b26) updated 2016-04-28
> >>>>> 
> >>>>> When you say this is what you're seeing, how are you checking this?
> >>>>> bitbake -e ? Which part of that output are you looking at?
> >>>>> 
> >>>>> You *will* now see VARIABLE_mymachine in bitbake -e output even if
> >>>>> "mymachine" is in OVERRIDES - that's because the overridden values
> >>>>> stay
> >>>>> in the datastore instead of a "finalisation" step eliminating them, in
> >>>>> order to allow OVERRIDES to be changed dynamically (possibly multiple
> >>>>> times) and still get the datastore contents you expect. That change
> >>>>> was
> >>>>> made in the 2.0 release (jethro). However That should not affect the
> >>>>> final value of VARIABLE if "mymachine" is in OVERRIDES however. Are
> >>>>> you
> >>>>> sure you are seeing what you think you're seeing? If so, is the
> >>>>> bbappend
> >>>>> in fact being applied? Because if what you describe was really
> >>>>> happening
> >>>>> then pretty much the entire system would be broken. I just checked a
> >>>>> similar case here and I see the expected results.
> >>>> 
> >>>> Yes, I am using 'bitbake -e <recipe> | grep ^VARIABLE' to look at this.
> >>>> 
> >>>> In my main recipe, I have this:
> >>>>      SOME_OTHER_VARIABLE ?= "${MACHINE}"
> >>>> 
> >>>> and my .bbappend for that recipe has these lines:
> >>>>      SOME_OTHER_VARIABLE_my-target-A = "something-else1"
> >>>>      SOME_OTHER_VARIABLE_my-target-B = "something-else2"
> >>>> 
> >>>> and local.conf has
> >>>> 
> >>>>      MACHINE="my-target-B"
> >>>> 
> >>>> With Yocto 2.0 (Poky ceeb52a2544) I get this:
> >>>>      SOME_OTHER_VARIABLE = "something-else2"
> >>>>      SOME_OTHER_VARIABLE_my-target-A = "something-else1"
> >>>>      SOME_OTHER_VARIABLE_my-target-B = "something-else2"
> >>>> 
> >>>> With today's (Poky a9b503b26) I get this:
> >>>>      SOME_OTHER_VARIABLE = "my-target-B"
> >>>>      SOME_OTHER_VARIABLE_my-target-A = "something-else1"
> >>>>      SOME_OTHER_VARIABLE_my-target-B = "something-else2"
> >>>> 
> >>>> I also use this mechanism to set COMPATIBLE_MACHINE a lot and
> >>>> that doesn't seem to be working the same either.
> >>>> 
> >>>> Am I doing something incorrectly (and just got away with it for years)?
> >>>> 
> >>>> n.b. If you're interested, I can share the .bb/.bbappend files
> >>>> with you privately (I can't put them on the public list)
> >>> 
> >>> Ah, one thing... do your machine names have uppercase characters in them
> >>> (as your pseudo-examples above do)? If so that was a 2.1 change;
> >>> anything
> >>> that is to be used as an override must now be lower case (so DISTRO,
> >>> MACHINE, etc.)
> >> 
> >> Indeed, my machine names do have an upper case letter in them.
> >> Any suggestions for how to handle this (short of renaming everything
> >> which
> >> would be a HUGE chore)?
> > 
> > It's no longer supported I'm afraid, BitBake's datastore is now coded to
> > expect lowercase for all override values. You say it's a huge chore but
> > surely a grep -l | xargs -n1 sed -i would do it?
> 
> If I were to rename my target (MACHINE), I'd then have to go through
> and rename all of the directories in the recipes that hold machine
> customizations and there are quite a few of those.  Similarly I'd
> have to rework all the recipes that use the old names (with upper case
> letters) to the new ones.  Not impossible, but quite a lot of work.

That is unfortunate, but again should be pretty easy to automate (assuming 
they are local of course and not proliferated across many users' build 
machines).

> Short of actually renaming the target machines, I could add an additional
> override (in <MACHINE>.conf) that is lowercase only and update just the
> recipes to use that instead.
> 
> By experimentation, I have found that XYZ_append_<MACHINE> seems to do
> the correct thing, even when <MACHINE> has upper case letters.  Maybe this
> is an oversight that will be fixed?

I'm not sure - Richard?

Cheers,
Paul

-- 

Paul Eggleton
Intel Open Source Technology Centre