Next over function and breakpoint setting

Next over function and breakpoint setting

[Orjan Friberg]

Out of curiosity I followed what happens when issuing a next command 
over a function call.  The reason I got curious was that I noticed 
(remote protocol) that GDB sets a breakpoint on the first instruction of 
the function being stepped over, and, when stopped there, sets another 
breakpoint at that function's return address to accomplish the "next".

Actually, maybe that's a misrepresentation: looking at the code, it 
seems GDB does a step, and if it *then* finds it ended up inside a 
function (step_into_function -> step_over_function) it decides to step 
out of it.

Looking at the code in infcmd.c and infrun.c I'm sure there's a good 
reason for not just setting the breakpoint on the address belonging to 
the next line of code, but I couldn't find any comment saying what that 
reason is.  Anyone care to enlighten me?

(I couldn't say if any of this was remote target/CRIS specific; 
apologies if it is and I missed it.)

-- 
Orjan Friberg
Axis Communications

Next over function and breakpoint setting

[Orjan Friberg]

Out of curiosity I followed what happens when issuing a next command 
over a function call.  The reason I got curious was that I noticed 
(remote protocol) that GDB sets a breakpoint on the first instruction of 
the function being stepped over, and, when stopped there, sets another 
breakpoint at that function's return address to accomplish the "next".

Actually, maybe that's a misrepresentation: looking at the code, it 
seems GDB does a step, and if it *then* finds it ended up inside a 
function (step_into_function -> step_over_function) it decides to step 
out of it.

Looking at the code in infcmd.c and infrun.c I'm sure there's a good 
reason for not just setting the breakpoint on the address belonging to 
the next line of code, but I couldn't find any comment saying what that 
reason is.  Anyone care to enlighten me?

(I couldn't say if any of this was remote target/CRIS specific; 
apologies if it is and I missed it.)

-- 
Orjan Friberg
Axis Communications

Re: Next over function and breakpoint setting

[Daniel Jacobowitz]

On Tue, Mar 16, 2004 at 04:42:30PM +0100, Orjan Friberg wrote:
> Out of curiosity I followed what happens when issuing a next command 
> over a function call.  The reason I got curious was that I noticed 
> (remote protocol) that GDB sets a breakpoint on the first instruction of 
> the function being stepped over, and, when stopped there, sets another 
> breakpoint at that function's return address to accomplish the "next".
> 
> Actually, maybe that's a misrepresentation: looking at the code, it 
> seems GDB does a step, and if it *then* finds it ended up inside a 
> function (step_into_function -> step_over_function) it decides to step 
> out of it.
> 
> Looking at the code in infcmd.c and infrun.c I'm sure there's a good 
> reason for not just setting the breakpoint on the address belonging to 
> the next line of code, but I couldn't find any comment saying what that 
> reason is.  Anyone care to enlighten me?
> 
> (I couldn't say if any of this was remote target/CRIS specific; 
> apologies if it is and I missed it.)

[Sorry, missed this the first time.]

"The next line of code" is a very iffy concept.  GDB doesn't analyze a
line to figure out what it does, or where it might transfer control
to... while possible, that would be a very different approach to
debugging.  So the only way we have to implement next is to step and
see where we end up - and if we don't like it, go until we're somewhere
else.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: Next over function and breakpoint setting

[Orjan Friberg]

Daniel Jacobowitz wrote:
> 
> "The next line of code" is a very iffy concept.  GDB doesn't analyze a
> line to figure out what it does, or where it might transfer control
> to... while possible, that would be a very different approach to
> debugging.  So the only way we have to implement next is to step and
> see where we end up - and if we don't like it, go until we're somewhere
> else.

I guess I wasn't really asking a question - I was just confused by the 
way it was implemented since it wasn't what I had expected.  Anyhow, 
thanks for clearing that up.

-- 
Orjan Friberg
Axis Communications

Re: Next over function and breakpoint setting

[Andrew Cagney]

> Daniel Jacobowitz wrote:
> 
>>
>> "The next line of code" is a very iffy concept.  GDB doesn't analyze a
>> line to figure out what it does, or where it might transfer control
>> to... while possible, that would be a very different approach to
>> debugging.  So the only way we have to implement next is to step and
>> see where we end up - and if we don't like it, go until we're somewhere
>> else.
> 
> 
> I guess I wasn't really asking a question - I was just confused by the way it was implemented since it wasn't what I had expected.  Anyhow, thanks for clearing that up.

BTW, have a look at the step out of range code.  The theory is that the 
target be presented with a range of addresses and it run until it leave 
that range (as is the case with a step or next).

The implementation breaks the theory -- the code didn't consider threads 
and multi-threaded interactions.  As far as I'm concerned, to fix the 
theory we need to disentangle threads.

Andrew