step into next source line (that belongs to me)

step into next source line (that belongs to me)

[Christian Parpart]

[-- Attachment #1: Type: text/plain, Size: 868 bytes --]

Hi all,

When you need to trace a c++ constructor with multiple inheritages, you maybe 
want to 'step in' to a few of them, but not all.
What I am missing is a "step in"-alike that does only step to the very next 
line that is owned by my very own source code. Everything else is not of my 
interest and just blowes you up your mind when debugging and trying to 
understand.

The code to be "my very own" could be defined as all source files under a 
specified root directory. every debugging step that can be assigned to a file 
inside this (or one of these) root directories should be used as a step. 
every other lines (e.g. caused via big header files with lots of 
inlines/templates/etc.) should be stepped over and though ignored unless 
explicitely stated.

Is there a way to do such things using gdb?

Thanks in advance,
Christian Parpart.

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: step into next source line (that belongs to me)

[Daniel Jacobowitz]

On Tue, Dec 26, 2006 at 01:19:00PM +0100, Christian Parpart wrote:
> Is there a way to do such things using gdb?

Not yet, no.  But it could probably be added using your definition of
interesting files.

-- 
Daniel Jacobowitz
CodeSourcery

Re: step into next source line (that belongs to me)

[Christian Parpart]

[-- Attachment #1: Type: text/plain, Size: 340 bytes --]

On Tuesday 26 December 2006 16:19, Daniel Jacobowitz wrote:
> On Tue, Dec 26, 2006 at 01:19:00PM +0100, Christian Parpart wrote:
> > Is there a way to do such things using gdb?
>
> Not yet, no.  But it could probably be added using your definition of
> interesting files.

please CC me when implemented though - I'm pleased to beta-test ;)

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: step into next source line (that belongs to me)

[Robert Dewar]

Christian Parpart wrote:
> On Tuesday 26 December 2006 16:19, Daniel Jacobowitz wrote:
>> On Tue, Dec 26, 2006 at 01:19:00PM +0100, Christian Parpart wrote:
>>> Is there a way to do such things using gdb?
>> Not yet, no.  But it could probably be added using your definition of
>> interesting files.
> 
> please CC me when implemented though - I'm pleased to beta-test ;)

Why don't you investigate and propose a patch. I doubt anyone is about
to implement this unless you do it, or hire someone to do it! That's
how gdb development works!

Re: step into next source line (that belongs to me)

[Jim Blandy]

Daniel Jacobowitz <drow@false.org> writes:
> On Tue, Dec 26, 2006 at 01:19:00PM +0100, Christian Parpart wrote:
>> Is there a way to do such things using gdb?
>
> Not yet, no.  But it could probably be added using your definition of
> interesting files.

From time to time people complain here about GDB stepping into system
library functions, and it's often because they've installed debugging
info for those libraries and found GDB doesn't silently 'next' over
them even when given the 'step' command, as it used to.  So GDB's
fallback behavior for dealing with undebuggable functions ends up
inadvertently providing a feature people want.

Perhaps each objfile (executable or shared object) could have a "step
status", specifying how step should behave with respect to
inter-object calls into that objfile.  All objfiles would start out in
the "ask" state, where step would ask the user whether they're going
to be interested in that objfile.  The answer would put the objfile in
the "step-past" or "step-into" status.

If control ever stopped in an objfile (breakpoint hit; segfault), GDB
could automatically mark it "step-into"; obviously, it's interesting
now.

The status would need to be saved somehow so that settings for shared
libraries wouldn't get lost each time the objfile was re-loaded.

There could be commands to mark object files manually.  They could use
globbing characters.

Re: step into next source line (that belongs to me)

[Vladimir Prus]

Daniel Jacobowitz wrote:

> On Tue, Dec 26, 2006 at 01:19:00PM +0100, Christian Parpart wrote:
>> Is there a way to do such things using gdb?
> 
> Not yet, no.  But it could probably be added using your definition of
> interesting files.

FWIW, it is not obvious this should be in GDB. The frontend might be
in a better position to conveniently specify which files to ignore,
and all that's needed is emitting "finish + step" when
entering "non-interesting" function. I planned to implement
some "auto-finish" functionality in KDevelop for quite some time.

- Volodya

Re: step into next source line (that belongs to me)

[Bob Rossi]

On Tue, Dec 26, 2006 at 11:20:41PM +0300, Vladimir Prus wrote:
> Daniel Jacobowitz wrote:
> 
> > On Tue, Dec 26, 2006 at 01:19:00PM +0100, Christian Parpart wrote:
> >> Is there a way to do such things using gdb?
> > 
> > Not yet, no.  But it could probably be added using your definition of
> > interesting files.
> 
> FWIW, it is not obvious this should be in GDB. The frontend might be
> in a better position to conveniently specify which files to ignore,
> and all that's needed is emitting "finish + step" when
> entering "non-interesting" function. I planned to implement
> some "auto-finish" functionality in KDevelop for quite some time.

Well, the CLI is considered to be a front end, IMO.

Bob Rossi

Re: step into next source line (that belongs to me)

[Vladimir Prus]

On Tuesday 26 December 2006 23:22, Bob Rossi wrote:
> On Tue, Dec 26, 2006 at 11:20:41PM +0300, Vladimir Prus wrote:
> > Daniel Jacobowitz wrote:
> > 
> > > On Tue, Dec 26, 2006 at 01:19:00PM +0100, Christian Parpart wrote:
> > >> Is there a way to do such things using gdb?
> > > 
> > > Not yet, no.  But it could probably be added using your definition of
> > > interesting files.
> > 
> > FWIW, it is not obvious this should be in GDB. The frontend might be
> > in a better position to conveniently specify which files to ignore,
> > and all that's needed is emitting "finish + step" when
> > entering "non-interesting" function. I planned to implement
> > some "auto-finish" functionality in KDevelop for quite some time.
> 
> Well, the CLI is considered to be a front end, IMO.

Ok, then correction: "The frontend" above should be replaced with "A GUI frontend".

- Volodya

Re: step into next source line (that belongs to me)

[Daniel Jacobowitz]

On Tue, Dec 26, 2006 at 11:20:41PM +0300, Vladimir Prus wrote:
> FWIW, it is not obvious this should be in GDB. The frontend might be
> in a better position to conveniently specify which files to ignore,
> and all that's needed is emitting "finish + step" when
> entering "non-interesting" function. I planned to implement
> some "auto-finish" functionality in KDevelop for quite some time.

I think, rather, it should be in GDB but many frontends won't choose
to use it.  In Eclipse, you'd want to step into source files associated
with this project, probably, and offer check boxes for other Eclipse
projects it depends on to step into those too, but not step into system
headers.  But the CLI ought to be able to at least do it by directory.

-- 
Daniel Jacobowitz
CodeSourcery

Re: step into next source line (that belongs to me)

[Eli Zaretskii]

> Date: Tue, 26 Dec 2006 17:34:50 -0500
> From: Daniel Jacobowitz <drow@false.org>
> Cc: gdb@sources.redhat.com
> 
> On Tue, Dec 26, 2006 at 11:20:41PM +0300, Vladimir Prus wrote:
> > FWIW, it is not obvious this should be in GDB. The frontend might be
> > in a better position to conveniently specify which files to ignore,
> > and all that's needed is emitting "finish + step" when
> > entering "non-interesting" function. I planned to implement
> > some "auto-finish" functionality in KDevelop for quite some time.
> 
> I think, rather, it should be in GDB but many frontends won't choose
> to use it.

Then we should implement it as an optional feature that can be turned
off.