From: Steve Kenton <skenton@ou...> - 2000-09-21 14:20:10
I look at the sim problem and concluded that
cygwin <include>/sys/socket.h had it wrong
but the comments in types.h made it pretty
clear that they know how hacked it is.
I guess the difference between signed and
unsigned on 32-bit numbers should not really
matter for packet length since the value is
supposed to always be non-negative, barring
the mythical "Godzilla-gram" packet >2GB.
However, an "approved" version of cygwin might
be a very good idea for a while if things like
this are going to be changing sign and gcc cares.
The HAVE_STRERROR problem is in support/cpp
I added #define HAVE_STRERROR to my copy of
config.h as a workaround test, but I think the
real problem is that config does a *load* of
tests and properly determines that strerror()
is present but there is nothing that uses
that information. The config.h file has an
include for i386 and i386/xm-linux hardwired
and neither of these define HAVE_STRERROR
although other system includes in i386 do.
I could not see where include.h or the system
includes in i386 were affected by the config
process in any way. Am I missing something?