patch to suggest putc/fputs over printf("string") or printf("\n")

[Jeffrey A Law]

  In message < r2d84msuns.fsf@happy.cygnus.com >you write:
  > "Kaveh R. Ghazi" <ghazi@caip.rutgers.edu> writes:
  > OK, remember, this is for gcc.  We have special attributes for
  > functions like gettext.  Therefore
  > 
  > 	printf (gettext ("some string `%s'"), 10)
  > 
  > would lead to a warning.  The gettext call is transparent and I expect
  > a warning about using printf instead of fputs for
  > 
  > 	printf (gettext ("some constant string"))
  > 
  > But there are also situations where the replacement returned by a
  > gettext call with an argument without format specifiers contains
  > format specifiers.  A call could look like this:
  > 
  > 	printf (gettext ("no format here"), some_variable)
  > 
  > The return value of the gettext call could be "value %d".
And in that case we can not and must not optimize.

If the format specifier is not a compile time constant we can't optimize. 
What's so hard about that?


  > Now one can of course say this is not used.  I don't agree with this
  > since some of the gettext extensions which will come sometime will use
  > dirty tricks like this.  The other objections could be that the
  > gettext attribute should not be taken into account and therefore is
  > not replaced.  But this is completely against the purpose of this
  > attribute: if we user gets a warning s/he can look at the case and
  > decide whether the use of printf is necessary.  If the compiler simply
  > silently rejects this case and does not substitute fputs for printf
  > the programmer might never learn about the missed opportunity for an
  > optimization.
So, you're going to give up an optimization that applies to a large number
of programs because it does not apply to gettext.  That is absurd.  

  > A possibility is of course to perform automatic conversion in the case
  > where no such problem exists and warn in the other case (when using
  > gettext etc).  But this is IMO too inconsistent, better always go with
  > the warning and let the programmer decide.  In this situation it is
  > necessary, though, to have a way to shut up the compiler if the way
  > the code is written is correct.  Somebody said #pragma?
It's not inconsistent at all.  A compiler tries to optimzie as much as it
can, but when it is not safe it does not optimize.  There's no inconsistency
in doing that.

jeff