Re: Q: access to subprogram

[kenner@lab.ultra.nyu.edu (Richard Kenner)]

In article <m291cv99nr.fsf@devo.ridgecrest.ca.us> Ronald Cole <ronald@ridgecrest.ca.us> writes:
>Not true.  A single instance of GCC doesn't support multiple machine
>targets (other than the -m options for some processor families).  GCC
>has to be recompiled to support different machine descriptions (that's
>what the -b and -V options of gcc are for).  I consider this as having
>multiple backends. 

I don't think that's what Robert meant.  The distinction he was
trying to make is that GCC has only one piece that actually generates
code (what is normally called the "back end") and that the config files
modify what it does.  In most compiler technologies, the config files
say how to generate code, while in GCC (with some small number of
exceptions) they describe the machine.

>Many years ago, when asked about why he didn't
>make GCC read loadable md files, Stallman answered that it would hurt
>performance (both for the compiler and for the generated code)

Yes, having loadable MD files would seriously hurt compiler
performance, but is at least something that could be contemplated.  By
contrast, trying to load the tm.h file at runtime seems hopeless.

>that was around the time when Stallman said that
>writing what we now know as the bfd libraries would be "hard").

Yes, that's the original of the unoffical acronym of "BDF".  However,
note that this library is indeed far from trivial.