> When doing native bootstraps, we need to set --with-gnu-as=no because > binutils 'as' does not handle a flag that we pass to our own version > of 'as'.
That looks like a kludge to me. You can probably achieve the same effect with a script filtering out the unsupported flag. > 2- Eric, your patch essentially disables the --with-gnu-as= flag. > When doing a native bootstrap on the 4.7 branch, HAVE_GNU_AS is set to > 1, regardless of the value of --with-gnu-as. The problem is that your > new test to decide whether to use gas just uses the value set in > config.gcc. But that's correct: the final say over --with-gnu-as is in config.gcc. Without my patch, you have a disagreement between auto-host.h and config.gcc (hence tm.h ultimately) if the latter overrides -with-gnu-as, which can potentially lead to inconsistencies like in the linker case. > Eric, shouldn't the whole section testing for GNU as move after the > inclusion of config.gcc? I don't think so, that would change the semantics of --with-gnu-as. -- Eric Botcazou
