Re: egcs-19980508 warning patches part 1/5

[Jeffrey A Law <law@cygnus.com>]

  In message < 199805141603.MAA25681@caip.rutgers.edu >you write:
  > 	I agree they are unlikely to point to an actual bug.  That's
  > why I try to eliminate them, so they don't distract you and others
  > from more serious warnings.
Yup.  Understood.  I've made the same argument myself :-)

  > 	Well if I'm the one investing the effort... :-)
Yup.  Thus the patches went in :-)

  > 	Seriously, I think I explained above why I see this as useful.
  > But I can understand your point of view and I'm not going to oppose if
  > you don't want these fixed.
As I said, I'm still undecided at this time.  And (of course) I'm
willing to keep going if that's what folks think is best.  I've
heard from both sides equally so far.

  > So maybe we should add:
  > 
  > 	-Wno-sign-compare
  > 	-Wno-unused
This kinda what I've been thinking about.  We can turn them on (particularly
-Wunused) at times to search for dead functions/parameters.

  > 	-Wno-uninitialized  (This one appears to be too broken to be useful.)
Yea.  I looked into improving this, but we need another fairly
significant concept to be added to the compiler.  I suspect we'll
have it one day, but not soon.

I actually did go fix a few of these.  But they're time consuming
since I actually did verify that the variable would be initialized
in each case before I put in a default initialization.

Even though these are the hardest warnings to fix properly, they
are probably the most important ones left to go after since they're
the most likely to expose a real (and bloody obscure) problem.

  > 	On the flip side, I think there are some other warnings (not
  > enabled by -W -Wall) that we should _add_ to WARN_CFLAGS.
I've considered this too.

Particularly aggregate-return (which isn't supported in some older
compilers, or has semantics which are rather unfriendly).

I've been nailed by shadowing problems in the past, so it's a
warning I like. :-)

  > 	BTW, what's a good value for LEN in -Wid-clash-LEN?
No clue.

  > 	-Wstrict-prototypes
  > 	-Wmissing-prototypes
  > 	-Wmissing-declarations
  > 	-Wredundant-decls
  > 
  > These might be useful to turn on once, fix the problems and turn it
  > off again since I think many spurious warnings would result.
There's certainly lots of warnings when you turn them on.  But
the prototype stuff is critical for fixing problems when HOST_WIDE_INT
is a different size than "int" or "long".

jeff