This does the following.
1. Fixes checking if the CXX compiler works on platforms other than windows.
2. Turns the error when the CXX compiler is missing or doesn't work into a warning.
3. Adds HAVE_CC and HAVE_CXX.
4. Only adds CC and CXX to config.mk when HAVE_CC or HAVE_CXX are true.
5. Disables Qt companion, Vulkan, CXX_BUILD and NEED_CXX_LINKER if HAVE_CXX is false.
6. Explicitly errors when the CXX compiler is broken or missing and Qt or vulkan support is enabled.
7. No longer explicitly links with the CXX compiler on windows since this should no longer be needed.
This also adds the function `check_enabled` to `qb/qb.lib.sh` which
can be used to dynamically disable any libraries that require C++
support.
By default 'make install' will now install the following 2-3 files
to $(DESTDIR)$(SHARE_DIR)/doc/retroarch which by default is
/usr/local/share/doc/retroarch.
COPYING
README.md
and if media/assets exists it will install:
media/assets/COPYING to COPYING.assets
Users will be able to configure this install path with.
./configure --docdir=PATH
For example:
./configure --docdir=/usr/doc/RetroArch-1.7.0
This allows a user to use --datarootdir=PATH to configure the
share directory used for pixmaps, desktop files, man pages and assets.
By default this will be '/usr/local/share'. Assets and man page install
paths can still be configured with:
--with-man_dir=PATH
and
--with-assets_dir=PATH
Some operating systems like Haiku use unconventional install paths
and this should allow them to configure their build correctly.
Haiku really should use the configure flags provided, that is why they exists.
In this case.
./configure \
--global-config-dir=$PREFIX/settings
--with-assets_dir=$PREFIX/data
--with-mand_dir=$PREFIX/documentation/man
Changing the defaults just makes the script more complicated for no good reason.
The check_lib function does not seem able to detect vulkan on windows,
this replaces a hack in Makefile.common that does the same, but while allowing
C89_BUILD=1 and --disable-vulkan to work.
LD retroarch
obj-unix/./libretro-common/net/net_natt.o: In function `natt_open_port':
/RetroArch/libretro-common/net/net_natt.c:120: undefined reference to `UPNP_AddAnyPortMapping'
/RetroArch/libretro-common/net/net_natt.c:133: undefined reference to `UPNP_GetExternalIPAddress'
/RetroArch/libretro-common/net/net_natt.c:126: undefined reference to `UPNP_AddPortMapping'
obj-unix/./libretro-common/net/net_natt.o: In function `natt_init':
/RetroArch/libretro-common/net/net_natt.c:61: undefined reference to `upnpDiscover'
/RetroArch/libretro-common/net/net_natt.c:74: undefined reference to `miniwget'
/RetroArch/libretro-common/net/net_natt.c:77: undefined reference to `parserootdesc'
/RetroArch/libretro-common/net/net_natt.c:79: undefined reference to `GetUPNPUrls'
/RetroArch/libretro-common/net/net_natt.c:81: undefined reference to `freeUPNPDevlist'
collect2: error: ld returned 1 exit status
make: *** [Makefile:159: retroarch] Error 1
This is a proof of concept for the fallback path without pkg-config.
As jack is something often not installed by default it demonstrates
that when pkg-config exists and the dependency doesn't the check_val
function in qb/qb.libs.sh will redundantly check -ljack and as expected,
fail. It also shows that when pkg-config is not installed and jack is
enabled with --enable-jack it will bail and print a configure error.
This accomplishes two things for the fallback path without pkg-config.
1. If --disable-foo is passed to configure it will explicitly skip
check_val. This has the benefit of reducing checks in
qb/config.libs.sh which are easy to break due to human error.
2. When a fallback path exists and --enable-foo is passed to configure,
but fails due to the missing -lfoo check_val will now bail and print
a configure error. However --enable-foo will still be ignored if
there is no fallback path and pkg-config is not installed.
One issue with this is that if pkg-config is installed and the package
foo is not, it will still check if -lfoo works. As not all pkg-config
implemenations can be trusted to work even if they exist this seems
unavoidable.
This resolves an issue where the default freebsd compiler does not
include /usr/local/include which contains important headers such as
GL/gl.h and results in the check_header function failing to find them.
Unfortunately pkg-config/pkgconf will not work here for two reasons.
1. It does not seem able to actually check for gl.h which RetroArch
explicitly needs.
2. Not all systems have a pkg-config implementation so we will still
have to fall back to checking for gl.h...
Fixes https://github.com/libretro/RetroArch/issues/5958
RetroArch no longer uses submodules for various good reasons and now uses git subtrees instead.
As a result these files will always exist and these tests will always be true.
Recent Raspberry Pi firmware images have renamed the vendor graphics
libraries (brcmEGL, brcmGLESv2, brcmOpenVG) to distinguish from the
standard VC4 Mesa driver.
* When videocore is selected, first try to use pkgconfig for the
new library names (brcmEGL, brcmGLESv2, brcmOpenVG).
* Ensure that non-pkgconfig fallback detection also checks the new
library names.
This should resolve compile issues in stretch whilst maintaining
compatibility with jessie firmwares later than 1.20160921-1 (the
first package that includes the renamed libraries).
The PKG_CONFIG_PATH must be set in your build environment in order
for the pkgconfig to be utilized, which will be included in
the next firmware (1.20170811-2 or later).
Adding NAT traveral interface to libretro-common, with (currently) a
backend support MiniUPNPC. Sensible future backends would be libupnp and
a direct implementation of PCP/NAT-PMP.
One some systems (tested with Gnome 3 on Arch Linux) the current method
of using `xdg-screensaver` to suspend the screensaver does not work.
Instead, using DBus to issue an `Inhibit` request is recommended.
The request returns a cookie that needs to be re-used to un-inhibit the
screensaver later. Additionally if the DBus connection is closed the
current inhibition is discarded. Thus, the DBus connection needs to stay
connected for the duration of the screenshot inhibition.
The code is heavily inspired from the [SDL 2.x
code](http://hg.libsdl.org/SDL/file/default/src/core/linux/SDL_dbus.c#l172).
I didn't call the SDL 2 code though since this it to fix the issue with
the GL driver, and I assume one would want to have screensaver inhibited
even when SDL 2 is not available (but GL is).
I've set "WIP" because:
* I haven't done C in a long time so my code is probably not great
* There's a dependency on DBus which I don't know is acceptable or
not
* I've put my code where I could to check it works, but `x11_common` may
not be the best place
* The code need and "init" and "deinit" kind of method as it needs to
initialise the DBus connection, and on deinit close it properly. I've
used `x11_connect` and `x11_window_destroy` but they don't sound like
the best choices.
* I'm a bit unclear as to what happens when "suspend screensaver" is
ticked on/off in the menu. This doesn't seem to call
`x11_suspend_screensaver` everytime, so I'm not sure if there's a hook
somewhere (as disabling screensaver suspend in the menu should cause a
DBus unhinibit request to be sent).
* Should I just call the SDL 2.x code (meaning that the GL driver would
depend on SDL 2.x at runtime)?
So, first of all are you ok with the approach, and if yes I'd gladly get
feedback about the code, how to architecture it and the best place to
put it.
Thanks!
Valgrind requires shared objects to be kept open for meaningful
debug information. Add configure option to disable dlclose() in
dylib_close() for Valgrind support.
The driver is now superseded by the exynos video driver. Also limare
only works up to mali kernelspace driver version r3p2, which makes
the driver nonfunctional with the latest r4p0.
Brad Parker