On Fri, May 13, 2011 at 2:06 PM, Gordan Bobic <gordan(a)bobich.net> wrote:
Peter Robinson wrote:
>>> I must say that I like this idea A LOT! :D
>> I like this idea a lot too. It will speed up development. :)
>>
>> If the -main- development is done this way, then a final compile for
>> the release is done on actual hardware with the actual toolchain, it
>> would also meet the specs of the Fedora Project. This is less
>> frustrating to developers, and it should only require a 1 recompile
>> since the build issues should be fixed.
>
> Ultimately anything compiled in koji that is not a scratch build could
> conceivably become part of the final compose of the OS so there's no
> way of telling what is dev and what is final. In reality once things
> settle down all development should be done upstream in Fedora and
> koji-shadow then just follows mainline.
I didn't think this was a suggestion for the official build koji, was
it? The point is that it is very hard to meaningfully troubleshoot a
process that takes 3 days to fail on a 1.2GHz ARM, and then has to be
re-started from scratch. The incremental progress becomes too painfully
slow. But if we can get that 3 day process into a 3 hour process (e.g.
I've only got about 4GHz worth of ARM cores, but about 40GHz worth of
x86-64 cores), then the problem is at least transitioned from the realm
of unworkable into the realm of slow.
Well there's nothing stopping people from doing that themselves.
I've tried QEMU emulation on an IBM x3850x5 I have at work with 64 2.8
ghz threads and 256 GB of RAM. Its still slower than the real arm
boxes.
Once you know it builds and works, waiting for 3+ days so bad because
at
least you know something usable will come out of it.
Believe me I am painfully aware of that. I have got gcc to compile on
one of the ARM boxes in mock. Using koji on the same device it doesn't
compile and I have no idea why so its still not a guarantee it seems,
very unfortunately!
Peter