By using top_srcdir if available. Make's include is relative to
current dir, not the Makefile being processed. That makes it hard
to include Makefile's relatively.
When I designed this first I thought people can define make vars on the
command line to override default features. That works. However, it's
natural to want to define the list of features in the toplevel Makefile
and not on command line. So, we don't override user's var now. They
can set a make feature var to 0 to disable it.
The macros CAIRO_BEGIN_DECLS and CAIRO_END_DECLS are declared in two
places: config.h and cairo.h. On Win32 build there is no config.h. So,
we can't rely on pulling CAIRO_BEGIN_DECLS from there. Hence, we now:
* Not add those declarations to config.h anymore,
* Test that every cairo private header includes some other cairo
header before any other includes.
These two are fairly enough to ensure that cairo.h is seen from all
private headers. There's still the case of headers not including
any header file at all. I'll fix that later.
Previously I made the mutex implementation code to use Windows mutexes
conditional on HAVE_WINDOWS_H. This was suppposed to be a more wine-friendly
version of checking for a Windows platform. However, it also meant that
when compiling without config.h, one had to stuff a #define for HAVE_WINDOWS_H
somewhere. Traditionally it was put in cairo-features-win32.h, and that
practice was carried over even after my build system rewrite.
Somewhere along the road I also decided that the mutex implementation header
does not need to include anything other than config.h. So the hacked up
HAVE_WINDOWS_H in cairo-features-win32.h was not being seen by its lone
user.
We now condition on any of HAVE_WINDOWS_H and _MSC_VER. So the hack can be
removed altogether.
Carl had this confusing problem today making the snapshot, which was
"make release-publish" erring. This commit reworks the release-publish
machinery to enforce serialization of some of the tasks such that it
works with make -j. Don't expect any speedup though as "make distcheck"
is inherently sequential.
By introducing $(top_srcdir)/cairo-version.h I tried to make it
unnecessary to rebuild the entire library when the version changes.
That comes handy around releases and when bisecting.
However this was not working as the version info was also ending
up in config.h and hence forcing a full rebuild upon change.
I now undefine the PACKAGE_VERSION* macros that autoconf defines,
so config.h does not change when cairo-version.h changes.
--coverage is a synonym for "--fprofile-arcs --ftest-coverage" during
compilation *and* "--lgcov" during linking. One might think this would
be a perfect workaround for the broken debian libtool which stopped the
linker seeing -lgcov - but they strip CFLAGS as well. Oh well.
If the native windowing system is disable (e.g. --disable-xlib) then the
test suite fails to build since the vector converters typically depend
upon gdk-pixbuf-2.0, which in turn depends upon a native gdk which
requires cairo to be built with support for the native windowing system. A
mess that should be resolved by separating rsvg and poppler into core and
higher-level libraries, but which we can workaround by simply ignoring
errors from undefined functions at link time.
