Robert Schwebel wrote:
> On Fri, Oct 20, 2006 at 06:30:05PM +0200, Jan Kiszka wrote:
>> What I mean with this are -maybe- only menu skins to customize PTXdist
>> in a way that users who don't know if they need "minimal logrotate" or
>> busybox with or without chroot are able to install target images from
>> pre-built binary packages. 
> 
> This is something which is on my radar for quite some time now, but I
> don't had the time to think about it in more detail.
> 
>> This would mean to reduce some menu options drastically, but also to
>> add new, special options ("install demo scripts+configs -> Erika robot
>> driving along the hall" ;)).
> 
> Note that customizing menues from a BSP is possible today.

Some pointers on this (for poor Christoph who has^H^H^Hwants to do the
work :))?

> 
>> And it would require to have those pre-built packages for the
>> installing host, not on the target. That's something I still don't
>> understand, if your ipkg is for installing packages from within the
>> running target (would sound like the SLIND approach) or if those
>> packages can be used by the host to prepare e.g. a CF-card image (like
>> we do here every day).
>>
>> Once this feature became real, our LiRE would be a very, very small
>> configuration + collection of in-house packages for PTXdist.
> 
> At the moment the ipkg packages are only binary packets of the files for
> the target side, no development files involved.

Ok. That would already be a problem for us, because our industry
partners and our students develop within a robotic framework that builds
against various special libs.

> 
> What I'd like to see in PTXdist is something which is known under the
> label PTXDIST_PATH within Pengutronix. The idea would be to have a
> PTXDIST_PATH environment variable or something like this, pointing to a
> location where pre-built binary packets (host+target part) are stored.
> 
> This would make it possible to do work in a "layered" way. For example,
> a common scenario in our customer projects is that 
> 
> a) Pengutronix develops a BSP for a module
> b) The module vendor adds his own stuff
> c) The end user adds his application.
> 
> So if you are for example c), it would mean that, if somewhere in
> PTXDIST_PATH there is a pre-built glibc packet found, it is simply
> installed instead of being built from the sources. Only the local
> applications, being added by the end user himself, are being built from
> the sources.

That sounds perfect.

> 
> I've once hacked a prototype, using one of these LD_PRELOAD tools that
> can be used to do on-the-fly packaging of "make install" stages. It
> turned out that this has some drawbacks:
> 
> - You don't catch if some upstream parts have to be recompiled due to
>   confiuration changes.
> 
> - It didn't work properly for all packets.
> 
> - I've no good idea how to solve the case where somebody wants to
>   recompile a packet by intention.
> 
> Seen from the configuration side, we could introduce [*] and [M] for the
> variants "recompile" vs. "take from packets".
> 
> It needs more thinking, so if you come up with a good idea I'm all ears.

I think we first need to analyse the PTXdist-way of building and
installing things more thoroughly. Also, this part is not top-most on
our list. The migration path includes an intermediate step to keep our
installation infrastructure over a PTXdist build system. Nevertheless,
we will always keep this goal in mind and may come back with more
concrete questions/ideas.

Jan

Attachment: signature.asc
Description: OpenPGP digital signature

-- 
ptxdist mailing list
[email protected]

Reply via email to