[RFC] Do we care about binary compatibility of code produced by cross-compilers?

[Mark Mitchell]

Paolo Carlini wrote:

>> (a) determined by compile- and link-time tests, and/or
> 
> While we are at it, I think I need a clarification about link-time tests: in the current configure.ac there is this comment:
> 
>   # Only do link tests if native. Else, hardcode.
> 
> thus, I'm not sure to understand the status of those tests: are or not generally allowed also when cross-compiling?

Link time tests are conceptually fine.

However, because of the way some people build GCC (using a single tree 
with Newlib and such in it), I believe that some people dislike 
link-time tests.

The problem is that at libstdc++ build-time, you don't have enough of 
the rest of the bits (e.g., Newlib) built, or at least you don't have 
all the right magic flags to find them correctly for different 
multilibs, or something.  My feeling is that we should just drop this 
build methodology, but, of course, I don't use it so I'm biased. :-)

Anyhow, link-time tests are not a theoretical problem (i.e., don't lead 
to inherently lead to different cross vs. native results) but may be a 
problem in practice for some people.

I don't remember the exact details, so, unfortunately, that's not a 
terribly helpful answer. :-(

-- 
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713