Re: [avr-gcc-list] Re: crosstool-NG

[Georg-Johann Lay]

David Brown schrieb:
> On 04/03/2011 06:20, Trevor Woerner wrote:
>
> > On Thu, Mar 3, 2011 at 11:38 PM, Omar 
> > Choudary<address@hidden>  wrote:
> >
> > > Trevor, your idea seems nice, although I would be even more 
> > > enthusiastic to put all the patches into the mainstream gcc and
> > > then just do a normal cross-build for the tools.
> >
> > Yes, ideally this would be the best solution.
> >
> > However, given the fact the patches exist but are still not yet
> > merged into the gcc project I simply assumed getting these patches
> > accepted by the gcc maintainers is not trivial.

One could use the GCC build machinery, anyway. Or is that too 
restricted? Up to now I have just experience with canadian crosses with 
--build=linux --host=mingw32 and --target=foo and that works out of the 
box. (After you got/built cross-build tools from linux to mingw32). For 
linux guys this is easier than doing native build under cygwin or 
whatever with build=host (actually I never tried that because the 
canonical solution works fine).

> It always takes time (and effort!) for patches to make their way into
> the main FSF gcc tree, and it takes time before these patches end up
> in released versions.  gcc has a fair amount of bureaucracy, and the
> tree supports a vast number of ports, hosts, and combinations.

You have to go through the bureaucracy only once (signing FSF/GPL 
stuff). As far as I can tell, getting patches into avr backend is no big 
deal, at least for straight forward stuff (I did'n yet try to get more 
elaborate stuff into avr). Even though GCC supports fair amount of 
whatever, avr development takes place in a small (with respect to rest 
of GCC) sandbox: the avr backend. Other parts of GCC are not affected by 
the work in that sandbox and thus is no issue for maintainers other than 
avr backend maintainers.

Johann