Hi,

This afects Cygwin badly, as it still uses stabs by default.
I've seen this problem here, but thought it was caused by a
few gcc changes I was making ...

Pierre Muller wrote:
>   As I was unable to understand the current implementation of
> the twos_complement_representation

I can't really see how it was supposed to work either.
Parsing 000000000 is broken currently.

> I rewrote it almost completely.

You are only counting the bits.  This function should
return the parsed number if it fits in a long:

/* If TWOS_COMPLEMENT_BITS is set to a strictly positive value and if
     the number is represented in an octal representation, assume that
     it is represented in a 2's complement representation with a size of
     TWOS_COMPLEMENT_BITS.

     If the number fits in a long, set *BITS to 0 and return the value.
     If not, set *BITS to be the number of bits in the number and return 0.

     If encounter garbage, set *BITS to -1 and return 0.  */


>   The code now checks that the most significant
> bit of the whole octal representation of the huge number
> that is being parsed is exactly at the bit position given by
> the twos_complement_bits parameter.
> 

>   The attached patch (against 6.6.90 source)
> fixes the problem that I describe in the previous
> email. I get no complaint for the 'unsigned long long' type
> compiled with '-gstabs+' option.
> 

Thanks for pointing in the right direction!

(Forgive me for counter patching, but I had started on this
  here too yesterday :( )

What about the attached?  Isn't it simpler?

We get passed the max number of bits the number we're parsing
can hold in TWOS_COMPLEMENT_BITS.

Just parse the number as unsigned, if it overflows, it doesn't
matter, we just account for the bits.
If it doesn't overflow (host has a big long),
	and TWOS_COMPLEMENT_BITS < sizeof (long) * HOST_CHAR_BIT
	and number is signed according to TWOS_COMPLEMENT_BITS - 1
	bit being set, sign extend the number into a long.

Just tested that it also fixes the problem.  I'll give it a
testsuite spin for C/C++ on Cygwin tomorrow, but the testcase
that suposedly tests this is in ADA, which I don't have a
setup for...

Cheers,
Pedro Alves