Editing source while debugging

[Vladimir Prus]

On Sunday 31 December 2006 03:07, Cavalaria Christophe wrote:
> On Sunday 31 December 2006 00:27, Jens Dagerbo wrote:
> > On Saturday 30 December 2006 23:56, Vladimir Prus wrote:
> > > Hi,
> > > I start to wonder what's the right thing to do if a user modifies a file
> > > while debugging. In Java (Eclipse), the IDE tries to modify the running
> > > application with the new code, but it's not possible for C++/gcc/gdb.
> > >
> > > There's concrete problem. You have breakpoint set in KDevelop (and in
> > > gdb) and foo.cpp:10. You add a line on top of the file. Kate moves the
> > > breakpoint marker to foo.cpp:11. For the next debugger run, the gdb
> > > breakpoint will be set at foo.cpp:11, but what to do for *this* debug
> > > session.
> > >
> > > 	1. Moving gdb breakpoint to foo.cpp:11 is not a good idea. As line
> > > 	11 in old foo.cpp, that's compiled into the currently debugged
> > > application, is not the same line as line 11 in new foo.cpp.
> > > 	2. If we don't move gdb breakpoint, and gdb stops on it, it will report
> > > 	line 10, while the actual code being executed is now at line 11.
> > >
> > > It seems that after breakpoint is moved, debugging the application does
> > > not make sense any more. Should we perhaps mark moved breakpoints in some
> > > way, disable them, and enable only for the next debugger run?
> > >
> > > - Volodya
> >
> > I'm a fan of simple solutions that work.. either disallow editing while
> > debugging (not sure we can do that) or simply flush all breakpoints in a
> > file if the user starts editing it while the debugger is running.
> 
> Yuck, in my opinion, don't do that! There is that option in Delphi ( make code 
> read only during debugging ) and I've only used it for a few minutes before 
> disabling it. It's much better for the user to deal with somewhat incorrect 
> breakpoints than to prevent editing during debugging.
> 
> Simple solution : leave it as it is and add a "code changed, do you want to 
> rebuild it?" when the user tries to resume the program beeing debugged. This 
> should be enouth of a warning that future breakpoints might not be as 
> accurate as possible.

Then, we'd need a way to ask Kate not to move markers corresponding to breakpoints.
Otherwise, we'd have some complexity devoted to moving breakpoint, when in fact
we don't support this at all.

- Volodya