I work on an application which uses both OpenGL and OpenCL, and relies a bit on interop between them.
The worst part of it is: when application starts, it checks for GL/CL extension and runs a simple diagnostic to verify that it indeed works properly. Right now this totally breaks replaying for me, even though I'm not interested in debugging any parts which actually use OpenCL.

It is clear that apitrace does not create OpenCL context and ignores all OpenCL calls, so any data getting from CL into GL is wrong. But the problem is that one interop function is in OpenGL: it is glCreateSyncFromCLeventARB from extension GL_ARB_cl_event. Apitrace honestly records this call to the trace, but fully skips it during replay. As the result, replay lacks a sync object which is used in subsequent GL calls that apitrace does replay... and crashes.

I know this is not something supported by apitrace, but maybe it's possible to avoid this crash with a simple fix?

Some ideas:

1. Replay glCreateSyncFromCLeventARB, but call glFenceSync instead of original function. That would create a sync object which can be waited for and deleted. I think it should be fine in any GL calls that application can use with it. I have a working prototype of this.
2. In all calls accepting sync object, skip the call if sync object does not exist. It's a matter of checking the result of std::map access. But I'm afraid the number of such calls can be pretty large.
3. Blacklist GL_ARB_cl_event during recording. If an application checks for availability of this extension, then it will skip the code path which uses it. If it does not... that's its own problem then.