Re: patch: cp/search.c broken

[Jason Merrill <jason at redhat dot com>]

>>>>> "Aldy" == Aldy Hernandez <aldyh@cygnus.com> writes:

> On Mon, Nov 26, 2001 at 02:25:35AM +0000, Joseph S. Myers wrote:
>> On 25 Nov 2001, Aldy Hernandez wrote:
>> 
>> > search.c doesn't compile because there is an attempt to declare bk in
>> > the body of a function-- not after curly's.  this isn't c++ guys :)
>> 
>> It's perfectly valid C99.  I asked before without an answer, I'll ask
>> again: non-C front ends can require GCC, what GCC version can they
>> require?  That is, what native compiler version should people have
>> installed before building a cross compiler?

> i think the bottom line is k&r.  isn't it?

For the gcc/ directory, yes.  Not for frontends.

> just because it's valid c99 doesn't mean we should use it in gcc 
> source code.  we should probably stick to the lowest common (sane)
> denomitor (k&r?).

> i was bootstrapping with 2.95.3.  i don't know what the regression tester
> was using, but it died too.

Bootstrapping shouldn't build the C++ frontend with anything other than a
just-built compiler.

However, my preference would be to keep the frontends buildable with 2.95,
so that we don't always have to do a bootstrap in order to build a C++
compiler.  I only bootstrap to test backend changes.

Jason