On 05/02/2015 10:40 AM, Peter Eisentraut wrote:
> On 5/2/15 10:20 AM, Andrew Dunstan wrote:
>> Putting the CORE directory at the end of PG_CPPFLAGS doesn't work. Using
>> the override to put it at the end of CPPFLAGS does work.
> It's strange that the include directory order should matter. What is
> the explanation for that? I don't see this documented anywhere. I
> don't see any evidence in the buildfarm that this makes a difference.
>
> The intent of my changes was to unbreak other platforms that had been
> broken by these Windows-related changes (which did succeed).
>
>
What failures on other platforms? There's not a single buildfarm failure
I can see attributable to this. All the recent failures are listed
here: <http://www.pgbuildfarm.org/cgi-bin/show_failures.pl>
The evidence in favor of my changes is in my testing. When I do it your
way I get errors, when I do it my way I don't.
Note that the override is EXACTLY what is done unconditionally in the
plperl GNUmakefile:
override CPPFLAGS := -I. -I$(srcdir) $(CPPFLAGS)
-I$(perl_archlibexp)/CORE
So, contrary to your claims, you have not made this makefile more like
plperl's - you've made it less like plperl's, and detrimentally so.
cheers
andrew
