Re: Warn on naked cross-tool names, add macros to detect target tools

[Ralf Corsepius]

On Tue, 2004-03-16 at 23:11, Kevin Ryde wrote:
> "Kevin P. Fleming" <address@hidden> writes:
> >
> > building on i686-pc-linux-gnu for i386-linux-uclibc
> 
> Yep, in fact I would guess a "downgraded" build like that is probably
> more common than an actual cross compile.
What you call "downgraded build", I'd call a "broken build".
And what makes you think that it is "more common"?

In the example above, you are building a package with a compiler not
having been compiled against the libc+OSes you are targeting. Thereby
you are using a compiler making bogus assumptions on your host.
This only works in cases, you don't trigger an incompatibility between
the libc's+OSes being involved.

i686-pc-linux-gnu  and i386-linux-uclibc might be similar enough to let
you get away this way in some cases, but in general, your consideration
does not apply.

>   That sort of thing is
> wanted all the time when making binary packages like for debian that
> are to run on the least common denominator in a chip family.
"Chip family" (CPU) is something very different than the libc. There you
often can get away using a compatible CPU, if libc+OS are the same or at
least very close.

Furthermore, you should consider that a target-tuple only describes a
"base family" of systems. It still can support several flavors of CPUs,
you can optimize for by using CFLAGS etc.

Ralf