Re: [PATCH 0/1] gnu-config make sure we use perl-native

[Tom Rini <tom_rini@mentor.com>]

On 03/14/2011 05:55 AM, Richard Purdie wrote:
> On Mon, 2011-03-14 at 13:18 +0100, Koen Kooi wrote:
>> Op 14 mrt 2011, om 12:51 heeft Paul Eggleton het volgende geschreven:
>>
>>> On Saturday 12 March 2011 02:31:40 Khem Raj wrote:
>>>>> Back to the drawing board on why the miss-match is occuring in
>>>>> Config.pm then.
>>>>
>>>> somewhere there is /usr/bin/perl referred may be a cross recipe ?
>>>
>>> I've hit this issue this morning, and have filed a bug report:
>>>
>>>   http://bugzilla.pokylinux.org/show_bug.cgi?id=866
>>>
>>> Not sure if it helps but at least it should be easy to reproduce.
>>
>> I haven't checked if the OE .dev and OE core gnu-config are similar, but have a look at:
>>
>> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=be21179c5321bd0afb9221f020ac12ad75c86a3b
>> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=edcdefbf6e0675c1bcc1fc4f464f654223380e50
>
> There are a few related patches to this floating around and I know Tom
> did some work in OE.dev to make perl-native an early bootstrap
> dependency.
>
> I've been giving this some thought and given the various goals we have,
> I think the long term best solution is going to be to place perl into an
> isolated bin directory in sysroots, then for anything using really using
> perl-native, add that directory to PATH. This is similar to what we now
> do with the bootstrap toolchains in OE-Core. This way the dependencies
> are deterministic and either perl is or isn't required but we don't
> instantly need to build perl-native or run the risk of nasty loops in
> the early bootstrap. Anything requiring perl-native would add the
> directory to PATH, probably using a tiny class file.
>
> The above patches still would make sense of course.

So, as background, there's a few cases of problems we have to deal with. 
  The ones Koen pointed out are fall-out from switching to making 
perl-native be part of the early bootstrap.

I would encourage someone to look at solving this problem in another way 
but I ended up finding that honestly, getting perl-native to build 
before we did almost anything else wasn't as hard as I feared (based on 
a previous attempt at doing it) and unless we can get things to the 
point where only perl needs perl-native (which should possibly be 
perl-cross) you can still hit some nasty corner cases.  And yes, doing 
this would mean changing the libperl-foo-native things into installing 
with host cpan into the TMPDIR which may or may not be easier said than 
done.

-- 
Tom Rini
Mentor Graphics Corporation