#689 PureData writing beyond t_object struct on Windows.

[Stephen Sinclair]

This very simple external provides a responder to a single message, "test". The object's data structure extends t_object by a single integer, "pad". I set "pad" to zero immediately after pd_new(). Later, when it receives "test", it prints "pad" again. On Linux, the log contains the following:

hello_new, x->pad: 0
hello_test, x->pad: 0

However, on Windows, the log contains this:

hello_new, x->pad: 0
hello_test, x->pad: 0x1

This means somewhere between initializing my object and receiving "test", the t_object structure is written to beyond its length.

This was noticed because in a real object I had a pointer in that location, which was invalidated by this effect, and dereferencing it caused PureData to crash. My temporary solution is to put a padding int there, but it would be nice to find the problem. I haven't yet managed to compile Pd from scratch for Windows yet, so I can't debug it until I get that working, but I thought I should report it in case anyone knows more.

[Stephen Sinclair]

Simple "test" responder that demonstrates the problem.

[Stephen Sinclair]

Forgot to add that this behaviour was noticed on the most recent downloads of both Pd-extended and PureData vanilla.