Re: [python] Allow explicit locations in breakpoints.

[Simon Marchi <simon.marchi@ericsson.com>]

On 2017-08-23 02:30 PM, Phil Muldoon wrote:
> On 23/08/17 18:51, Keith Seitz wrote:
>> On 08/23/2017 06:58 AM, Phil Muldoon wrote:
>>
>>> diff --git a/gdb/python/py-breakpoint.c b/gdb/python/py-breakpoint.c
>>> index 6156eb6179..8431bed939 100644
>>> --- a/gdb/python/py-breakpoint.c
>>> +++ b/gdb/python/py-breakpoint.c
>>> @@ -681,7 +681,7 @@ bppy_init (PyObject *self, PyObject *args, PyObject *kwargs)
>>>  	case bp_breakpoint:
>>>  	  {
>>>  	    event_location_up location
>>> -	      = string_to_event_location_basic (&copy, current_language);
>>> +	      = string_to_event_location (&copy, current_language);
>>>  	    create_breakpoint (python_gdbarch,
>>>  			       location.get (), NULL, -1, NULL,
>>>  			       0,
>>
>> This binds python interfaces to the CLI, and I don't think we want
>> that. I would have expected (perhaps naively) to see explicit
>> locations supported using a more natural python convention, such as
>> using PyArg_ParseTupleAndKeywords.
> 
> My original implementation was that. I had function="foo" etc as
> parameters to the constructor. The problem is with the -line
> parameters and relative parameters. So line=+3 won't work in the
> Python parameter sense. So it would have to be line="+3" in the
> constructor. If keywords are all strings, I'm not sure I see the point
> of parsing them as keywords when you can just specify the whole string
> (i.e.  gdb.Breakpoint("-source=foo.c -line=28"). This is what we do,
> more less, with the current constructor: IE
> 
> foo = gdb.Breakpoint("bar.c:23")
> or
> foo = gdb.Breakpoint("functionName").
> 
> So this patch allows the addition of explicit location parsing in much
> the same vein as we handle regular breakpoints. (That is, with a
> CLI-like interface). The gdb.Breakpoint class, for better or for
> worse, is based pretty much on the interface to create_breakpoint. I'm
> not adverse to implementing keywords; it's a little extra string
> wrangling, but I'm not clear on the benefit? The gdb.Breakpoint class
> has always handed over the interpretation of breakpoints to GDB using
> the CLI like syntax. I can see why MI separates it out, and I take
> your point well, in a machine context.

I think for Python it would make sense to support the two paradigms.  If you
are writing a Python command that ends up installing a breakpoint, it would
be nice if you could directly pass what you received to the gdb.Breakpoint
constructor and have it parse it (including explicit locations).  For example,

  (gdb) special-break -file foo.c -line 17

But it would also be nice to have a keywords based API, for when the line/file/function
information is already split.  It would avoid having to build an explicit linespec
string just to have GDB parse it after.

In terms of API, I think the "spec" argument could be mutually exclusive with
the function/file/line/etc keywork arguments, which would be added.  An error
would be thrown if you try to use both ways at the same time.

About the line="+3" issue, because this is Python, the line keyword could
probably accept integers and strings.  And if it's a string, there could
be some validation on the format.

Simon