bootstrap/7800: underscore.c:0: malformed option `-A system=posix'

[Zack Weinberg]

Wolfgang Bangerth <bangerth@dealii.org> wrote:
>   Zack posted this question a while ago:
>     
>      This error indicates that the bootstrap compiler is picking up the
>      specs file for the just-built compiler; naturally this cannot work.  I
>      am not sure how it could have happened -- please repeat the command I
>      quoted, with an additional -v option tacked on the end, and post the
>      output.

Wow, this is old.  I did track down one way this could happen, and
papered it over by removing underscore.c.
See http://gcc.gnu.org/ml/gcc-patches/2002-09/msg00271.html (ignore
the subsequent thread, which is mostly sound and fury about who owns
the c++filt executable).

I hesitate to call the problem resolved, however, because there is
nothing intrinsically *stopping* Make from generating $build/gcc/specs
before we are done using the bootstrap compiler.  I don't know why the
bootstrap compiler was reading $build/gcc/specs in the first place,
but my theory is it was a bug in gcc 2.95, which it is far too late to
fix now.

A larger question is why do we generate that file at all?  It is an
exact dump of the specs compiled into the driver; as such I see no
purpose for it other than to slow down the driver by making it read
the file on every invocation.  People who want to hand-hack the specs
file can always use -dumpspecs to get a copy first (of course, this
would need documentation, if we made that change).

zw