Re: [RFA] Re: Python: add field access by name and standard python mapping methods to gdb.Type

[Doug Evans <dje@google.com>]

On Fri, Nov 4, 2011 at 1:40 PM, Paul Koning <paulkoning@comcast.net> wrote:
>
> On Nov 4, 2011, at 1:41 PM, Doug Evans wrote:
>
>> On Tue, Oct 4, 2011 at 8:56 AM, Paul Koning <paulkoning@comcast.net> wrote:
>>>
>>> On Oct 4, 2011, at 11:41 AM, Eli Zaretskii wrote:
>>>
>>>>> From: Paul Koning <paulkoning@comcast.net>
>>>>> Date: Tue, 4 Oct 2011 11:29:58 -0400
>>>>> Cc: Doug Evans <dje@google.com>, gdb-patches@sourceware.org
>>>>>
>>>>>
>>>>>> For future reference, there is a separate ChangeLog in doc.  Entries for
>>>>>> documentation have to go there.
>>>>>
>>>>> I overlooked that file.  Thanks for the reference.  Should I move the entry there?
>>>>
>>>> Yes, please.
>>>
>>> Done.
>>>
>>>>
>>>>>> Could you write a NEWS entry for this change?
>>>>>
>>>>> How about this?
>>>>
>>>> Fine with me, thanks.
>>>
>>> Committed.
>>
>> Ummm, hi.
>> I know I looked at the patch and approved it myself, but having played
>> with it for awhile I'm having second thoughts.
>> And before a release goes out I'd like to get this resolved.
>> If you want I'll do the work, or at least help however I can.
>>
>> One way to look at my reasoning is that a type "has a" field list but
>> it's not the case that a type "is a" field list.
>> And I'm uncomfortable with len(gdb.parse_and_eval("1").type) == 0.
>> IOW, len(gdb.Type of "int") is now 0.  I think it should flag an exception.
>>
>> OTOH, adding the new support to the result of gdb.Type.fields() is great.
>>
>> Anyone object to me changing things and moving the new iterator
>> support to gdb.Type.fields()?
>> Or do people disagree with my reasoning?
>> I haven't looked into what's involved.  At this point I just want to
>> get the user-visible semantics right.
>
> Part of my reasoning is to have gdb.Value and gdb.Type look alike.  gdb.Value always had field lookup by name, i.e., it behaves like a Python dictionary.

For structs and such sure, but not so for scalars.
Scalars shouldn't behave like a dictionary.

If we're going for consistency between gdb.Value and gdb.Type then for
structs and such we'll need len(gdb.Value) to return the number of
fields. [Right?]

Maybe I'd be happy if gdb.Type (and maybe gdb.Value) were simply more
rigorous in throwing exceptions for invalid cases.

> So I wanted to make the same apply to gdb.Type since the analogy seemed obvious.  And in both cases, I wanted the normal Python dict methods to be available.  (For gdb.Value, that's not submitted yet.)
>
> In my view, gdb.Type.fields remains as a backward compatibility synonym for gdb.Type.values (the standard dict method).
>
> I do agree that having len() return 0 instead of an error seems wrong.

That could be fixed by having typy_length throw an appropriate error
for scalars, etc.