Re: patch for #14363

[Oliver Buchtala <oliver.buchtala@googlemail.com>]

On 06.08.2012 12:42, Phil Muldoon wrote:
> On 08/02/2012 05:21 PM, Oliver Buchtala wrote:
>>  From 05a921d423793db93dfead59c9fe12d342f55d75 Mon Sep 17 00:00:00 2001
>> From: Oliver Buchtala <oliver.buchtala@googlemail.com>
>> Date: Wed, 1 Aug 2012 21:59:43 +0200
>> Subject: [PATCH] Adapt py-prettyprint to inform pretty-printer about current
>>   recursion level.
>>
>> ---
>>   gdb/python/py-prettyprint.c  |    3 +++
>>   gdb/python/python-internal.h |    1 +
>>   gdb/python/python.c          |    2 ++
>>   3 files changed, 6 insertions(+)
> Thanks.
>
> Needs a GNU style ChangeLog, testcase and documentation
> additions.
>
> I am having trouble understanding what you are doing with "recurse"
> attribute.  From valprint.c, in the GDB sources (which calls
> apply_val_pretty_printer, among others):
>
>     "RECURSE indicates the amount of indentation to supply before
>     continuation lines; this amount is roughly twice the value of
>     RECURSE."
>
> So I am not sure that this is the value you want?  Anyway, I think you
> want the number of recursions for that printer, not the recurse value
> being supplied to apply_val_pretty_printer.  A testcase would help
> here to understand your intention.
>
>> diff --git a/gdb/python/py-prettyprint.c b/gdb/python/py-prettyprint.c
>> index 86d4f2c..d5c6b66 100644
>> --- a/gdb/python/py-prettyprint.c
>> +++ b/gdb/python/py-prettyprint.c
>> @@ -476,6 +476,9 @@ print_children (PyObject *printer, const char *hint,
>>     if (! PyObject_HasAttr (printer, gdbpy_children_cst))
>>       return;
>>   
>> +  if ( PyObject_HasAttr (printer, gdbpy_level_cst))
>> +    PyObject_SetAttr(printer, gdbpy_level_cst, PyInt_FromLong((long) recurse));
>> +
> There are a few issues with this patch hunk:
>
> * It is not clear in the documentation, but PyInt_FromLong can raise a
>    PyErr_NoMemory exception, so you need to NULL check the return and
>    dispatch the exception appropriately.
>
> * This leaks a reference from PyInt_FromLong.  PyObject_SetAttr
>    increments the reference count of the right side of "=" Python
>    equivalent (IE foo.bar = baz), which in this case is the integer. So
>    the transient reference from PyInt_FromLong will last forever.  You
>    you need to assign the results of PyInt_FromLong to a variable and
>    call Py_DECREF on it after the "PyObject_SetAttr" call.
>
> * PyObject_SetAttr can fail, so the code needs to deal with this
>    contingency.
>
> * Why PyInt_FromLong over PyLong_FromLong?
>
> Beyond the patch fixes, is it not possible to count the recursion
> level by some internal bookkeeping in the printer?  I have no problem
> with including the attribute if it is helpful, or clearer to solve the
> problem, but I am curious if you have tried the alternative, if
> possible?
>
> Cheers,
>
> Phil
Thanks,

I see there are lots of deficiencies with my patch. Sorry for that.
You are right in all your points.
I was not aware that recurse == indentation. (misnaming? ;) )

For the second proposal, I need a set data structure for longs.
Do you have such a data structure in gdb somewhere? Or which should I use?
Then I would offer a prototype implementation.

Regards,
Oliver