Re: Breakpoints in dynamically loaded objects

Re: Breakpoints in dynamically loaded objects

[Tom Tromey]

>>>>> "Jim" == Jim Blandy <jimb@zwingli.cygnus.com> writes:

Jim> However, that would mean that, if you're doing shared library
Jim> work and you really do rename a function that had a breakpoint
Jim> set on it, that old breakpoint will just sit around forever.  I'm
Jim> not sure what to do with that.

Nothing.  It doesn't matter.  Although I do sometimes leave gdb
running a long time (days), the overhead (actual and conceptual)
associated with a useless breakpoint is less than the agony of
constantly rediscovering that gdb has silently disabled breakpoints.

Try debugging a Java program that uses JNI some time.  It should be
easy, but it isn't.  You have to set a bp on main and then run before
you can set a breakpoint on any function the Java class library.  Then
you have to set a breakpoint on an internal function in libgcj,
continue, and then finish, before you can set a breakpoint on the JNI
function loaded from the shared library.

Then if you need to re-run, you get to go through this process again.
And, because of the need to "finish" on that second breakpoint before
enabling the JNI breakpoint, you can't automate it.

Tom

Re: Breakpoints in dynamically loaded objects

[Eli Zaretskii]

On 8 May 2000, Tom Tromey wrote:

> And, because of the need to "finish" on that second breakpoint before
> enabling the JNI breakpoint, you can't automate it.

Is that because GDB cannot run the debuggee from breakpoint commands?

If so, I suggested patches to allow that (a long time ago).  Michael 
Snyder asked for lots of testing before it could be accepted, so it got 
buried under the backlog.

Maybe we should revive that again.  I still have the patches, although by 
now they must have suffered from some bitrot.  But it shouldn't be hard 
to retrofit them, since the changes were quite simple (in a nutshell, GDB 
uses a bunch of static variables to control the debuggee, and running the 
inferior from a breakpoint command overwrites some of these variables).

Re: Breakpoints in dynamically loaded objects

[Eli Zaretskii]

On Wed, 5 Sep 2001, I wrote:

> On 8 May 2000, Tom Tromey wrote:
>    ^^^^^^^^^^
> > And, because of the need to "finish" on that second breakpoint before
> > enabling the JNI breakpoint, you can't automate it.

Sorry, I just now noticed that this is old mail.  Please disregard my 
response.

I noticed that quite a few messages from May 2000 were sent to the list.  
What happened?