Is a target glibc really needed to build full crossgcc?

[Andrew STUBBS]

Marius Groeger wrote:
> Theoretically, possibly, you *could* get away with just the glibc 
> headers. AFAIK there are no target executable binaries generated when 
> providing the c++,objc,... compilers and libraries. However, I know only 
> one way to obtain these glibc headers, and that is to _build_
> glibc. :-)
> 
> Maybe you could experiment with glibc's configure and Makefile stuff, 
> possibly there even is a "make install-headers", but I doubt it is that 
> easy. Since glibc uses an elaborate directory overlay system to get to 
> the final set of .c and .h files for a specific platform, it is not very 
> easy to extract only the headers as it is with the Linux kernel, for 
> instance.

I did this ages ago. I can't remember what the exact problems were, but 
I think there was some trouble with the configure tests making wrong 
decisions/errors due to link failures. Certainly you don't actually need 
to pull anything in from the libraries.

Sigh, it used to be that you needed glibc headers/libraries for the C++ 
compiler, but the C compiler would build fine without either (at least 
for SH), so you had to build the compiler twice with different configs, 
but it was still simple. Building a compiler and library set from 
nothing but source has got a lot more complicated since then. Chicken 
and egg problems are always fun.

Luckily you only need the reproductive tract of the chicken (xgcc, cc1) 
in order to produce the egg (glibc headers) which can then produce a 
full (C only) chicken. xgcc and cc1 can be built without headers or 
libraries given the right make incantations. Don't even try to build C++ 
without a full glibc (ostrich egg?), it just isn't worth it once you 
have a C compiler for glibc.

Andrew

------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sources.redhat.com