Re: Patch to enable -Werror [take 2]

["Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu>]

 > From: David Edelsohn <dje@watson.ibm.com>
 > 
 > 	We already have developers accidentally using functions not
 > present on all platforms (e.g., fseeko and ftello).  Just because someone
 > builds cleanly with -Werror on one system doesn't mean that errors won't
 > appear on other systems.

Certainly possible, but probably rare enough IMHO.


 > 	I already spend enough time tracking down real bootstrap failures
 > and regressions.  I do not want to spend more time tracking down warnings
 > converted to errors.  It is not a productive use of my time, but your
 > patch would require me to do that.

I'm willing to work out a compromise if you would too.

Here's my suggestion.  I could add a configure flag --disable-werror,
which you could use so that you'd never be completely stopped by a
warning turned into an error.  I would still have to insist that
-Werror be enabled by default or there's no firewall against general
regressions.

But in a the case of an AIX-specific regression you at least could
opt-out temporarily.  Does that satisfy your personal concern?

		--Kaveh

PS: I got your list of AIX warnings.  There are only a few minor host
related issues not covered by existing bypasses.  Prototypes for
strsignal, atoll and ldgetname, and there's the "uchar" conflict
between cpphash.h and sys/types.h.  These are all easy for you to fix
or add AIX-specific bypasses for in rs6000/t-aix43.

BTW if on AIX sizeof(char*) != sizeof(int) then the missing strsignal
prototype may cause an actual bug since it's being passed to a
variadic function.  Probably nothing to worry about in this case, but
I regularly come across real bugs in this process just like this one.

--
Kaveh R. Ghazi			ghazi@caip.rutgers.edu