Re: [Boost-build] V2/1.34: next steps

From: Vladimir Prus (ghost_at_[hidden])
Date: 2006-02-14 03:56:58

On Monday 13 February 2006 14:35, Jim Douglas wrote:
> Vladimir Prus wrote:
> > On Monday 13 February 2006 13:45, Jim Douglas wrote:
> >>Vladimir Prus wrote:
> >
> > If you already have initial qcc.jam for V2, can you try to follow the
> > steps outlined in:
> >
> >   https://zigzag.cs.msu.su/boost.build/wiki/TestingNewToolset
>
> Hmm. I think I would prefer to receive your stamp of approval before I
> commence testing :-)

Ok. I've looked at qcc.jam and it looks very much like gcc.jam. Given that 
gcc.jam is pretty solid, I'd expect little problems except where QNX's gcc 
differs from stock gcc.

However, maybe just inheriting from the gcc toolset would be better? We can 
always remove the inheritance if it provides problematic for future 
extensions, but if we start with a verbatim copy of gcc.jam it's very likely 
that changes in gcc.jam will never make it into qcc.jam, even if applicable.

> >>qcc supports:
> >>
> >>- two compilers GNU gcc & Intel, with versions of both.
> >>- two C++ standard laibraries GNU & Dinkumware with sub-variations
> >>- self hosted (native) compilation on QNX
> >>- cross compilation on Windows & Linux

Does that require some deep support from build system?

> >>- mutilple hardware targets: MIPS, PowerPC, SH-4, ARM, StrongArm, Intel®
> >>XScaleâ„¢ Microarchitecture, and x86

Is that anyting more than an extra command line option?

> >>One further requirement is that in order to run the regression tests in
> >>a cross-development environment, bbv2 should support some form of remote
> >>execution. (But that's for later...)
> >>
> >>It has been suggested by Reese Dunn (see thread started 3/11/2005) that
> >>the bjam command line options could look something like:
> >>
> >>     $ bjam qcc-3.3 architecture=power stdlib=dinkum
> >>     $ bjam qcc-2.95 architecture=x86 stdlib=default
> >>
> >>I have already ported a very crude version of qcc.jam for bbv2 that
> >>works, but it is far from complete/perfect. However, we do not have to
> >>implement all the features for v1.34. The "first cut" file is attached.
> >
> > Yes, I think all we need to to be able to run the regression tests with
> > V2, with the same confication as in V1. Advanced features can be
> > implemented later.
>
> Overall, if we can achieve the same capabilities as bbv1 for v1.34 I
> will be happy - anything extra will be a bonus. My main concern is that
> we start off down a road that will allow us to develop the potential
> offered by QNX. IMHO a quick port of v1 would not give us that flexibility.
>
> What is the best way forward?

The only big problem seems to be using both gcc and intel as backend. However, 
we've agreed to postpone that for later. So, I'd think the best idea now 
would be to inherit qcc.jam from gcc.jam, and then start testing it.

- Volodya