There is a problem when the material datablock is removed. Looking for a solution.

@Julian Eisel (Severin) How can I clean the pointer in the modifiers when the material is removed from object materials? Now, if you remove a material and then try to update the prop, crash..also the prop value did not change when the material is removed and this is wrong.

I Have tested using row.prop(md, "material", text="") instead of row.prop_search(md, "material", gpd, "materials", text=""') and get the same crash (also prop has the problem it shows all materials, not only objet materials as prop_search does).

As a note, I have a fix ready to make the search menu behave as before with string properties for ID names (which is discouraged because of possible name duplicates). That's a very simple one and can go into 2.83, so this at least behaves as before. I also have an alternative one to handle this better in the UI code (P1380), so that the data-name is set for string properties, not the UI name. But it's not save for 2.83.

For 2.90 we can then do this change here, addressing the remaining issue with linked materials.

There are at least two obvious huge things missing here: handling of those pointers in read code, and in lib_query.c

Scratch that, was obviously not well awake this morning, those are handled by the generic ID foreach callbacks... Others remarks remain valid though.

@Julian Eisel (Severin) If we all agree to not move this to 2.83, you could move the search fix to make it works now, and later I can move to 2.90 this fix and in this way we have time to check carefully all. I think is too risky add a change like this few days of release date.

Really, with your temp fix in prop_search, 99% of the problems will be solved, except linked materials, but this is not a big issue and always it's better that having this important bug open.

Thanks @Bastien Montagne (mont29) for your review...your help has saved me a lot of code.