Solaris2.7 bootstrap failure, cc cannot compile cpplex.c

[Zack Weinberg]

On Mon, Jul 10, 2000 at 10:41:24AM -0400, Kaveh R. Ghazi wrote:
> Hi Zack,
> 
> 	When I compile cpplex.c on solaris2.7 using cc (WorkShop
> Compilers 4.2), I get:
> 
>  > cc -c -DIN_GCC -DSVR4 -g -DHAVE_CONFIG_H
>  > 	-I. -I../../egcs-CVS20000710/gcc
>  > 	-I../../egcs-CVS20000710/gcc/config
>  > 	-I../../egcs-CVS20000710/gcc/../include
>  > 	../../egcs-CVS20000710/gcc/cpplex.c
>  > "../../egcs-CVS20000710/gcc/cpplex.c", line 172: non-constant initializer: op "+"
...
> I believe the cause is the macro definition of T() in cpplex.c which
> is "U STRINGX(e) + 4".  This appears to be done to remove the CPP_
> prefix from the entries in TTYPE_TABLE.

Yup.

> I think if you instead create the entries in TTYPE_TABLE without
> CPP_ and prepend CPP_ using CONCAT2 when desired it will probably
> fix the problem.

There was some reason why that wasn't done in the first place.  I
probably didn't want to drag symcat.h into every user of cpplib.h.

> Or we could just take out the "+ 4" since its just for debugging
> output printing (right?)

It's also being used for error messages now - but only ICE messages, I
think.  (Grep for _cpp_spell_operator())  And this is purely cosmetic,
so... 

> Does either approach sound okay with you?

Let's take out the + 4.  Would you mind generating a patch to that
effect?  I have too many untested changes in my own tree right now,
and you can test the patch directly where I can't.  Consider it
pre-approved.

zw