Re: A macro to check for data type alignment

[Paul Eggert]

> From: Akim Demaille <address@hidden>
> Date: Wed, 13 Feb 2002 13:42:38 +0100
> 
> I'm somewhat against using AC_DEFINE to replace functions (but until
> we can provide more convenient, I'll just shut up 1:), and against
> AC_FUNC which replaces without saying.  Please, submit an
> AC_REPLACE_FUNC_GETTIMEOFDAY.
> 
> Since there appears to be some debate about it, I'd like Paul's
> insight on this function.

First, does anybody still port to ancient hosts that insist on
1-argument gettimeofday?  I last used one about 8 or 10 years ago.
Solaris 2.4 had an unprototyped gettimeofday precisely so that C code
worked either way.  (Or maybe you're worried about C++ code? that
wasn't standardized in Solaris until 2.5 came out.)

Second, there's lots of prior art here.  For example:

* Emacs defines GETTIMEOFDAY_ONE_ARGUMENT if gettimeofday cannot accept
  two arguments.

* 'make' just sees whether a 2-arg gettimeofday exists, and if so defines
  HAVE_GETTIMEOFDAY.

* textutils has a gettimeofday replacement that works around a Mac OS X
  bug where gettimeofday clobbers the static buffer that localtime uses
  for its return value.

If you want a fancy, general solution to this problem, I'd suggest
combining the approaches of 'make' and 'textutils'.  User code should
assume a standard gettimeofday with two arguments.  The textutils
method is reasonable; maybe it needs to be extended to support
one-argument gettimeofday, if that's still a real problem.

If it matters, modern software tends to prefer clock_gettime to
gettimeofday, since clock_gettime has higher resolution.