Blender Git Loki

Temporarily let DirectX take priority over OpenGL

OpenGL doesn't work yet. Shouldn't be too difficult to get working, but
I don't have a working OpenXR runtime with OpenGL support to test here.

Merge branch 'master' into soc-2019-openxr

Fix some issues, improve error message

Fixes:
* Destruct surface when destroying session on error. Fixes null pointer
dereference when trying to draw the surface on next redraw.
* Fix trying to enable same extensions/API-layers multiple times due to
static array usage not being cleared after error.
* Null pointer dereference with OpenGL drawing

Fix crash on OpenXR error before context is created fully

Show useful error reports in case of errors.

E.g. with an active OpenXR runtime installed, but no HMD plugged in,
Blender will now show: "Failed to get device information. Is a device
plugged in?".
In case of such errors, the VR-session will be cleanly exited, with
no side effects to the rest of Blender (at least if there are no bugs).

The GHOST_Xr API now allows setting a custom error handling callback
which may cleanly destroy all GHOST_Xr data.

Safe and informative GHOST_Xr error handling

This wraps all functions that could fail into some proper (although
unfinished) error handling.

These were the requirements for the error handling strategy:
* If an error occurs, cleanly exit the VR session (or the context if the
error happend during its set up), causing no resource leaks or side
effects to the rest of Blender.
* Show a *useful* error message to the user.
* Don't impair readability of code too much with error handling.

After some back and forth I decided Ghost internal exceptions are the
best way to go about this. I get exceptions are a controversial feature,
IMHO that's because most people use them 'wrong' however. Here's the
rationale:
* Most alternatives require early exiting functions. This early exiting
has to be 'bubbled up' the call stack to the point that performs error
handling. So the code gets really impaired by error checking. Tried
this first and wasn't happy with it at all. Even if error handling is
wrapped into macros.
* All GHOST_Xr resources are managed via RAII. So stack unwinding will
cause them to be released cleanly whenever an exception is thrown.
* GHOST_Xr has a clear boundary (the Ghost C-API) with only a handful of
public functions. That is the only place we need to have try-catch
blocks at.
(Generally, try-catch blocks at kinda random places are a bad code
smell IMHO. Module boundaries are a valid place to put them.)
* Exceptions allow us to pass multiple bits of error information through
mulitple layers of the call stack. This information can also be made
specific with a useful error message.
As of now, they conain a user error message, the OpenXR error code (if
any), as well as the exact source code location the error was caught
at.

So if an error is caught inside GHOST_Xr code, an exception is thrown
with specific and hopefully useful information in it. In the Ghost C-API,
these exceptions are caught and passed on to a custom error handling
callback. This callback will be defined by the Blender window-manager
and output the error information via a usual user report popup (not done
yet). It can also ensure the entire session is shut down.

Note that the majority of errors OpenXR can return are for invalid API
usage. Assuming the API usage is valid though, most error messages
should never reach users and can be left a bit more vague. Maybe we can
add something like "This is probably a bug and should be reported" to
those.

Check runtime DirectX requirements for graphics binding too

Check runtime OpenGL requirements prior to creating graphics binding

DirectX will need these checks too.
Also fix warnings due to wrong printf formatting.

Merge branch 'master' into soc-2019-openxr

Disable debug extension on Windows

Unfortunately, enabling XR_EXT_debug_utils crashes instance creation
with the Windows Mixed Reality runtime. Maybe I'm doing something wrong,
for now, just disable it.

General minor fixes and cleanup

* Initialize all class member variables
* Add version to runtime name printing
* Separate functionality code from debug prints
* Improve code structure using Doxygen groups
* Make accessors const functions
* Add (Doxygen) comments
* Naming
* Reorder functions

Fix wrong return value causing OpenXR initialization to fail

Fix compile errors on windows and with bundled OpenXR SDK sources

Refactor GHOST_XrContext into class + interface

Makes GHOST_Xr much more consistent with the rest of GHOST. Basically I
added a GHOST_IXrContext interface which can be called by the GHOST
C-API. The internal GHOST_XrContext class implements this.
Outside of GHOST only the opaque GHOST_XrContextHandle is accessible.

Kept GHOST_Xr_intern.h and GHOST_Xr.cpp for now. I'll probably remove
them soon. There's not much reason for both of them to be there.

Use OpenXR extension for extra debug messages in debug mode

Enables (or tries to) the XR_EXT_debug_utils extension which allows
setting a custom callback for additional debug prints. Note that I
haven't been able to test this really, as the Monado runtime appears to
not have this fully implemented (had to force sending a custom message
to find that out...). Will test with the Windows MR runtime in a bit. We
can also improve message quality then (it can print the exact function
name the message was sent from, print additional custom labels which
could indicate session state, etc.).

Only try to enable OpenXR API validation layer in debug mode

Cleanup: Rename XR_DEBUG_BEGIN -> XR_DEBUG_ONLY_BEGIN

Of course also renames XR_DEBUG_END accordingly.

Only print info messages with --debug-xr enabled

Add --debug-xr commandline arg and pass on to GHOST XR contexts

Doesn't do anything yet.

Preparations to allow enabling OpenXR API validation layer

For this to work two environment variables have to be set:
XR_API_LAYER_PATH and LD_LIBRARY_PATH, both have to point to the API
layers built by the OpenXR SDK. It also requires you compile Blender
linked to your own SDK build (OPENXR_USE_BUNDLED_SRC disabled).
Further changes will make this set up unnecessary, so validation layers
can be enabled via some flag.