Re: strsignal module

[Colin Watson]

On Tue, Jan 15, 2008 at 12:28:55AM +0100, Bruno Haible wrote:
> You fixed the first portability problem: the function is missing on some
> platforms. But there is another one.
> 
> I want to use the replacement in GNU clisp, so I did "man strsignal", and it
> says:
> 
>        "On some systems (but not on Linux), a NULL pointer may be returned
>         instead for an invalid signal number."
> 
> And it's even worse than that: for negative or out-of-range signal numbers,
> 
>   * Solaris strsignal returns NULL,
>   * AIX 5.1 strsignal returns (char*)(-1) !
> 
> As a user of this function, I don't want to special case these return values.
> The function should always return a reasonable pointer.

Thanks for pointing this out.

> So I propose to change the module so that, on these two platforms, it
> overrides the system function. Do you agree? May I do this? Do you want to
> do it?

I think the attached patch is right (I tested it with LIBS=-lstrsignal
after installing a bogus /usr/lib/libstrsignal.a with a strsignal
function that always returns NULL), but would appreciate review from
those more fluent in Autoconf than I am.

Is it sufficient to test a single value? It's been some time since I had
convenient access to either Solaris or AIX systems.

Cheers,

-- 
Colin Watson                                       address@hidden

gnulib.strsignal-working.diff
Description: Text Data