debugging - Why does my Eclipse project have phantom debugger breakpoints?

Question

I've got a small project which, when run in the Eclipse debugger, always seems to stop in FileInputStream.class line 106, where files are opened. There are no breakpoints set, but Eclipse behaves exactly as if I have a breakpoint here. If I clear all breakpoints, it still happens.

I have a second much-larger project in the same Eclipse workspace which does not suffer from this problem.

I just moved the smaller project off of my old Linux machine, where I developed it in Europa Eclipse and had this problem, onto my new Windows machine, where I continue to see the problem in Ganymede Eclipse. The problem persists across operating systems and across Eclipse versions, yet apparently not across projects. I don't get it! I grepped through every file in this project's directory and couldn't find anything that might be a file somehow directing Eclipse to stop in FileInputStream.

Further info: the apparent breakpoint is actually not for line 106 of FileInputStream; it appears to be an Exception breakpoint for FileNotFoundException, being thrown from native code called from that line in FileInputStream. But again, I don't appear to have any breakpoints set at all. Are Exception breakpoints defined somewhere else?

Answer 1

Did you try to un-select

Window > Preferences > Java > Debug : Suspend execution on uncaught exceptions

? (as mentioned in this thread, for instance)

Why does Eclipse work that way?

It goes back to 2002, when the breakpoint object hierarchy has been stripped down.

In order to set a breakpoint, with the old API, a client required Java Model Objects - such as IType, IField, etc.
With the new API, all that is required by the debug model is type names, field names, etc.

This allows clients to set breakpoints when Java Model Objects are not available.
Clients now specify the resource to associate a breakpoint with (before we constrained it to the associated Java Model resources).

Breakpoints can now also be "hidden". That is, they need not be registered with the breakpoint manager.
Breakpoints can also be selectively persisted (markers only allowed all/none of a marker type to be persisted).
This makes the debug model more flexible, and gives clients more building blocks.

This has also simplified some part of our Java debug implementation - for example, the feature "suspend on any uncaught exception", simply sets a breakpoint for the type named "java.lang.Throwable", rather than a specific IType in a specific project.
The breakpoint is not registered with the breakpoint manager (i.e. hidden) - it is only known and used by one client.
Another example is the "run to line breakpoint". The IJavaRunToLineBreakpoint has been removed, as its special functionality is no longer required. Now, the Java debug ui simply creates a "line breakpoint" that is hidden, non persisted, and has a hit count of 1. This is an example of providing building blocks to clients.