I take it it works for you then? Its good to have confirmation it works for people other than me, seeing as my only feedback so far is master tanker saying it doesn’t work! (My testing is obviously limited however, so I can believe there are bugs there. But I need someone to tell me where they are first, preferably by sending me problematic .blend files…)
I’ld agree that the code is mostly a mess and needs some serious work, an inevitable consequence of its organic development I expect. However, seeing as Blender is changing its gui model with 2.5 doing that right now wouldn’t make much sense, at least in the GUI bits which are the most problematic. (The actual object export code isn’t so bad, if under-commented.) It does take quite a lot of reading to work out what is going on though - its certainly not new feature friendly.
As for your second suggestion, that causes problems if you have two groups with the same material name, though I can see its potential value, esp. for exporting invisible geometry for AI use etc. Keeping the number and appending the material name on the end would probably be best, as that would make it easy to search for geometry chunks by material name whilst keeping name uniqueness and the ability to search by position. You can’t switch making an object for each material off however - the egg file format requires that each object have only one material.
I’m afraid that your 3rd suggestion is unsolvable - that is an inevitable consequence of using floating point maths. Reality is that the slight error should not be visible to humans anyway, so its not really a problem. (Well, it can become an issue if tiling stuff as cracks can form momentarily, but then you should probably be stitching your tiles together for speed anyway, or if they are simple generating them in code.) You have to remember that once the egg file is loaded it goes through almost identical floating point transforms before it hits the screen anyway, so even if you did fix that problem for loading it would still re-emerge in panda.
The other thing it needs is a far deeper manual - some explicit tutorials on things like setting up normal mapping, making collision geometry, the stride bone and example code of how to use some of these features within panda would be more valuable I think than more new features.
However, right now the above is mostly academic - I have effectively forked the exporter and whilst I can keep my fork going would very much prefer to fold it back into the actual sourceforge repository. That is entirely dependent on psiberpunk really. And even if he gives me sourceforge write access my time is limited as well, though I should be able to get some of that done, but rate of progress would be slow.