>>sounds pretty wierd...
>>Gut reactions without any clear reason would be to test if this happens only in the init() of a control already instatianted on the form (my take on your description) or if the same behavior is seen in a class [defined with the changed init] before it is put on the form. Once it's on the form it isn't a class anymore, it is a an object instance, the term is pseudo-subclassing (as you probably know) as can be seen from the way such a method is coded in a prg - not that this should make a difference to the method not working.
>>
>>The other thing I'd check is if the behaviour is the same if the this.addproperty is before the dodefault()
>
>It happened even without any dodefault(). I don't think I need the dodefault(), but I included it later, just in case.

Ok, scratch idea two - assuming you did not have NODEFAULT somewhere in the code run from that init <g>.

>
>I believe this will be difficult to reproduce. The form worked at first, and started to do strange things only after I changed code - completely unrelated, apparently - in some other part of the form.

Still, having a very tiny repro and some of the possible fixes might give you an idea what is happening. [As I said, I like such puzzles]
>
>Changing the AddProperty() method to the AddProperty() function solved the problem. Naomi pointed me to a thread that had a similar problems - completely different symptoms, but also caused by the AddProperty() method, also in VFP 9 / SP 2. And also solved by changing to the AddProperty() function.

Actually, the mentioned thread would jibe better than your problem with a SWAG coming from the part prototypal nature of vfp. In that thread (only skimmed it) there was not any attempt to first cache the .width to a local and then set the new property from that local - even if Dragan hinted this should be tried. Especially the post Fabio answered fits with that idea IMHO - the "this" needed for "width" might already be partialy corrupted for the special way some of the GUI properties (remember _assign/_Access irregularities and was there not a caveat on Eventbinding ?) work. .Width worked in the debugger, but not when the line was actually run.

This my a scientific helper method called "beating around all the bushes" ;-)
but with luck you get a few more points of differing behaviour
to formulate a hypothesis not mostly drawn from your subconcious ;-)
and tempting the crowd here with code ready to run sometimes works.

And yes, I used .New/AddObject when possible to get object cleaning
or base addproperty() function for the last years...

regards

thomas