Mail Archives: djgpp/2001/06/24/09:38:06
On Sun, 24 Jun 2001, Laurynas Biveinis wrote:
> > I agree that "make bootstrap" should work, but it involves additional
> > complications, so if you just want to build GCC with minimal fuss, you
> > may wish to avoid bootstraping.
>
> Eli, IMHO this is a wrong advise. The one and only supported way
> for ordinary user to build a native compile is `make bootstrap'. Plain
> make _might_ work, but if it doesn't, there's noone to blame. I understand
> that bootstrapping takes longer and uses more disk space, but it gets
> much more tested than ordinary `make'. So it's the way with minimal fuss.
> Please note that language frontends other than C are not written in portable
> C - they use GNU C, and might use features found only in the same version
> of compiler. In other words, building of GCC 3.0 with 2.95 or earlier
> might fail in C++ frontend and the like.
If the GCC's front-ends are written using C extensions that might not be
supported by older versions of GCC, then my advice is indeed wrong. But
why would GCC maintainers do such a grave mistake? It's perfectly clear
that strict ANSI C cannot be used, but using extensions supported only
by the latest version is a far cry from that. Doing so narrows the range
of possible systems which can build GCC, without any good reason, so it
doesn't sound like a good idea to me.
IIRC, a large portion of GCC bug reports was related to bootstrapping, at
least last time I checked.
And, if it's true that the normal build is less tested in the DJGPP
configuration, perhaps that ought to change. Unless GCC really uses the
latest extensions, there's no reason why a DJGPP user would need to
bootstrap.
- Raw text -