I noticed a problem with Blender 2.4x and 2.5:
If you use the function “make parent” when the position of the parent isn’t (0,0,0), the coordinates of the child are wrong in the “Transform properties” dialog, or when you use childOB.LocX, …
Worse, the coordinates are wrong in the ipo curve too.
For en export script it’s very annoying because the animation of the child won’t be well exported in the case of the right window.
Are the problems, that you have noticed, similar to those I reported some days ago here (incl. via the two other older threads referred in there)? If “YES” - there may be a serious BUG in Blender!!! Indeed… :eek:
Regards,
If you were to come up with a simple test case that demonstrates the bug and posted it to the bug tracker it would have a significantly higher chance of getting fixed.
@ Uncle Entity - I think I demonstrated a rather weird behaviour in my three threads (quoted also here in this thread) and I mentioned that i dont know where to post the links to them to be analysed and - hopefully - bug get corrected. If you know the place, pls publish the link to it. Thanks in advance!
@ Uncle Entity - Thanks for your info! I reported my observations given in my three threads (see link in posting #2 above) following the first link of your list, section 2.49b, under bug report #23414…
If somebody observes the same as myself in 2.5x - please report in the 2.5x section cause my .blend file (see link in posting #2 above) is for 2.49b.
Regards,
hi all,
this behavior it is not an issue - it is a feature - as Ton said me on Blender Conference 2008 - supposedly for more comfortable rigging.
But you can get more intuitive behavior thanks well hidden and nowhere documented option: Shift+Ctrl+P which creates non-inverse parenting. 
I have added this to the manual page on BlenderWiki right now - look at: Manual/Modeling/Objects/Groups_and_Parenting
Thank you very much Migius!
It would have been strange if it had been a bug that the developers didn’t detect it.
I think the non-inverse parenting should be in the menu. It looks like a secret feature.
Else, I found a solution for exporting properly the animations:
For each frame_number used:
Set the Blender frame number to frame_number
Read the position of the object with ob.matrixLocal and make conversions with toEuler()...
@ migius - Hum, well… if it was “well hidden and nowhere documented option” - this explains WHY nobody answered my questions in my three threads on that subject in this forum so far plus the current thread (of vincentweb). With your permission, I will post a link in my threads to your explanation here so that more people know about THIS option. I’d also ask you to take a look at the question I raise in my threads (my third thread which contains links to the other two is here) - HOW would be possible for a python script to distinguish between the “visual” and saved object’s location. i.e. how a script may “judge” which is the correct one? Under the term “correct location” I mean the location the object is at the moment of checking… Thanks in advance!
Regards,
i agree, it is strange they forgot to take it in.
Else, I found a solution for exporting properly the animations:
For each frame_number used: Set the Blender frame number to frame_number Read the position of the object with ob.matrixLocal and make conversions with toEuler()...
thank you for sharing the solution with us 
this needs deeper investigation… sorry, i am afraid i have no time for it now, i will put it on my todo-list next to armature-animation support for Collada scripts (in opposite to openCollada library)