Geoff Levner wrote:
| Problem solved (no thanks to you guys!).
| The solution is to allocate Cocoa data structures on the AppKit
thread,
| exclusively, if they need to be deallocated later.
This probably isn't doing anything more than hiding the real problem...
| The AppKit thread allocates its own NSData, copies the PDF data into
| it, then signals the Swing thread that its copy of the data is no
| longer needed. And the AppKit thread releases its own data later.
Data-sharing bugs tend to be very timing dependent. A change as
radical as
you describe could easily reorder the thread execution enough to drive
the
data-trampling into hiding. For now.
| Is this a JVM bug? An AppKit bug?
Most likely, an application bug.
