> Needs more work:
> 
> - copyright year 2004 in signull.c
> 
> - did you test it on ia64 and/or ppc?
>   can't quite tell from your wording

yep, and also tested on fc2 i386.

> - this code has ISO problems.

That's to put it mildly.  The trick it's pulling with function 
descriptors isn't valid ISO-C (but by knowing how descriptors work does 
lead to predictable behavior ... :-)

We're going to have to live with this - the testsuite needs to include 
deliberately erroneous programs.

>   % cat ptr.c
>   typedef long data_t;
>   typedef long code_t (void);
>   volatile data_t zero[10];
>   volatile code_t *desc = (void *) zero;
> 
>   % gcc-340 -Wall -S -pedantic -std=iso9899:1990 ptr.c
>   ptr.c:4: warning: ISO C forbids qualified function types
>   ptr.c:4: warning: ISO C forbids initialization between function pointer and `void *'
> 
>   What about:
> 
>     code_t * volatile desc = 0;

>   But that's semantically different from what you wrote.  Hmmm.  I don't
>   think it's possible in ISO C to initialize a pointer-to-code with the
>   address of a data object.  Any ideas?

This:

    code_t *volatile desc = (code_t *) (void *) zero;

does appear to fool the compiler.

see attched,
Andrew

> ===
> 
> 2004-06-15  Andrew Cagney  <cagney@gnu.org>
> 
> 	* gdb.base/signull.c (bowler): Replace data_pointer with data_read
> 	and data_write cases.  Add code_descriptor case.
> 	(zero, desc): New array and pointer.
> 	(data, code): Change to simple pointers.
> 	* gdb.base/signull.exp: Fix probe pattern matching a function
> 	descriptor SIGSEGV.  Replace data_pointer with data_read and
> 	data_write tests.
>