Re: __STRICT_ANSI__ "fixes" on STDC_0_IN_SYSTEM_HEADERS (solaris) targets

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

 >  > However I believe since fixincludes moved to the top level
 >  > directory we're no longer looking in the target headers and getting
 >  > that definition and thus the __STRICT_ANSI__ changes are always
 >  > applied, even when they're not supposed to be.
 >  > Am I reading the situation correctly?
 > 
 > I don't know.  We only have a couple of Solaris platforms where I
 > work and they (mostly) aren't configured to build anything.  This
 > means it isn't easy for me to chase this down.  I would hazard a
 > guess, though, that the change is at least innocuous or there would
 > have been complaints by now.  :)

The problems fixed by STDC_0_IN_SYSTEM_HEADERS certainly are not
showstoppers.  The fact that fixincludes no longer honors it is not a
huge deal, but the bug will trigger under the right set of unique
circumstances.  It was apparently enough of a problem at one point to
create the STDC_0_IN_SYSTEM_HEADERS hack in the first place.

First of all the problem only occurs on those systems which set
STDC_0_IN_SYSTEM_HEADERS, that means solaris and interix according to
my grep results below.  Second you must be using gcc-4.0 or later,
namely those gcc versions where fixincludes was moved to the top
level.  Third, you must use gcc with -ansi.  Fourth, you must use one
of the problematic headers in your source code which was broken with
__STDC__==1.

   > config/interix.h:#define STDC_0_IN_SYSTEM_HEADERS 1
   > config/sol2.h:#define STDC_0_IN_SYSTEM_HEADERS 1

Over time, sun fixed some of these problems, so e.g. solaris10 will
see less breakage than older OS versions.  I don't know the situation
on interix.  I think it's been hidden so far because not every solaris
user has 4.x installed and later solaris versions minimize the
problem.  And of course not everyone uses -ansi or -std=c89, etc.

Recently I know that e.g. this affected Eric on 2.5.1.  It occurs with
pthread.h.  That's partly how I realized this problem existed: he was
helping me by testing my pthread testcase on various solaris versions,
and the testcase failed on 2.5.1 with -ansi.

 > Anyway, is there a way to
 > determine the setting for STDC_0_IN_SYSTEM_HEAD at run time, since
 > not for compile time?  That'd fix it.
 > Thanks - Bruce

The runtime of what, gcc or fixincludes?  Whatever solution we come up
with, I'd like to avoid duplicating setting STDC_0_IN_SYSTEM_HEADERS,
i.e. bad idea to do it once in gcc and once in fixincludes.  Better is
if we can include the target config file somehow.

		--Kaveh

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