Re: Objective-C gdb (was: Re: Problem with AppKit/Fonts?)

[Eric Dahlman]

On Monday, July 14, 2003, at 10:21 AM, Richard Frith-Macdonald wrote:

> On Sunday, July 13, 2003, at 05:41 AM, Adam Fedor wrote:
>
> > On Saturday, July 12, 2003, at 07:58 AM, David Ayers wrote:
> >
> > > Richard Frith-Macdonald wrote:
> > >
> > > > I use gdb on ix86 machines ... and I have no problem with the 
> > > > 'next' command.
> > > > Perhaps what's needed is a short test program to run the debugger 
> > > > on, and
> > > > an explanation of how to reproduce the problem with 'next'
> > >
> > > I havn't been able to reproduce this with "simple" -base tests, but 
> > > with ProjectCenter.
> > >
> > > I "break" at main, "next" all the way to NSApplicationMain, step 
> > > into  it and when trying to 'next' over the creation of the 
> > > autorelease pool, gdb continues (i.e. dosen't stop when the creation 
> > > returns).  Note that in main there is already a release pool created 
> > > and destroyed before calling NSApplicationMain.
> >
> > I get it now. It only seems to happen when using next within a method 
> > that is in a shared library. In fact, I've been able to replicate 
> > this on even C programs (see attached). Someone should tell me I'm 
> > wrong - I can't believe no one else has noticed this...
>
> I think it must be a relatively recent change ... it doesn't show up 
> superficially on tests of recent copies of the debugger, and mostly 
> I've been using older copies.
> My main development systems are using gdb built from cvs with your 
> patches on the 7th of March, and the problem
> does not show up in your test application with that version of gdb.

Just for historical reference I remember back when I was using DDD (as 
a front end to gdb) on Solaris around the turn of the century (Gee that 
makes me sound like a geezer) there was a similar problem where if you 
were stepping through code and returned form a function call in a 
library it would act like a cont(inue).  At the time I found out that 
this was a deficiency in GDB at the time (possibly related to elf...?) 
so I would usually set a breakpoint after the function call that was 
causing me the problem to get control again.

Now I looked for a google reference but I didn't get lucky although I 
did find some traffic on the NetBSD lists about a similar problem circa 
98.  I guess my point is that it is not surprising and it is not 
without precedent; make sure you have the newest GDB and report it as a 
bug if it happens.

-Eric