I think I found the problem... It's a linker (ld) bug !
This week, I spent most of my free time migrating my main system to a newer Linux distribution (I was still using Mandriva 2010.2, which I will keep a while still, as a release-viewer build system, and I'm now migrating to Rosa 2012 Marathon which is very similar to Mandriva 2010.2 but will be supported/maintained till 2017 while Mandriva (and not only 2010.2) is now history).
When I compiled the viewer under this new(er) distro, I found two problems. One was with gcc (v4.6.0), which is buggy and causes OpenJPEG to decode textures in a very wonky way (like if they had "seams"); the solution was just to compile gcc 4.6.4 (which I already had compiled for use with my old distro) and update to it on the new distro.
The second issue seems to be the same as yours: The red tracker arrow and the red beacon are not showing ! In fact, they are "there" (and you should still be able to see the black shadow "behind" the invisible red beacon), but drawn with a black transparent colour (LLColor4(0,0,0,0)). I tracked the issue to be a bad initialization of the gTrackColor global variable which is not properly set on viewer startup (it stays at 0 (black with 100% transparency) instead of being initialized to (opaque) red). That variable is initialized in the llui library, in llui.cpp:
The code is perfectly legit, and should work (like it does when compiled under Windows or on Mandriva 2010.2 and older).
Now, the interesting thing is that only the global variables initialized with LLColor4::red got badly set on viewer startup when compiled on Rosa 2012... Replacing the code with: | | | | Code: LLColor4 gTrackColor = LLColor4(1.f, 0.f, 0.f, 1.f); // This is LLColor4::red as well
LLColor4 gDisabledTrackColor = LLColor4(0.5f, 0.f, 0.f, 1.f);
LLColor4 gLandmarkMapColor = LLColor4(1.f, 0.f, 0.f, 1.f);
LLColor4 gLocationMapColor = LLColor4(0.f, 0.72f, 1.f, 1.f);
| | | | |
From there, only three possible components could be the cause of this issue: cpp (the C pre-processor), gcc/g++ (the compiler), or ld (the linker). Since I was using the same compiler (and its associated pre-processor) in both cases, it could only be the linker, and replacing Rosa's with Mandriva's indeed also solved the issue !
To summarize:
- Bugs are everywhere, and more often in others' code than in mine...
- While one can do educated guesses about where bugs do exist, it's still risky to bet on it (I lost my bet: it was not a MacOS-X graphics driver bug... for once).
- The work around is easy and will be part of next releases.
- If you can see (part of) the beacon when enabling the "Cheesy beacon" feature (in Advanced -> Rendering), but not the red part, then it is highly probable that the binary you are using is affected by the same bug and was compiled on a system with a buggy ld, i.e GNU ld v2.21.53.0.2 (and perhaps others; ld v2.20.51.0.7, on the other hand, works fine in this respect).
|
