bug#20193: 25.0.50; declarative type specification for D-Bus args

[Michael Albinus]

Daiki Ueno <ueno@gnu.org> writes:

Hi,

>> Well, dbus-test.el passes now, thanks. But there are still some cases
>> I'm not so happy with:
>>
>> plist-get
>>  (dbus--test-create-message-with-args
>>   '(:array)
>>   '(:array :signature "u")
>>   :type '(:array :uint32)
>>   nil)
>>  :signature)
>>
>> |- "asauau"
>>
>> I would expect "asauaub"
>
> Right, thanks for pointing that.

Thinking about, it seems to be OK. Ah :type keyword must always be
followed by two other arguments, and so it takes nil as the empty array
values, instead of a boolean false. Whether we allow to be sloppy, and
to add a missing nil as empty value list at the end of the arguments, is
something to be decided (and documented).

>> Maybe you could be a little bit more verbose about, and adapt the code?
>> When there is a :type argument, there must *always* be two additional
>> arguments?
>
> That makes sense.

As said above.

> Moreover, I personally prefer not to mix implicit and explicit typing.
> So, I am currently thinking to collect type specifiers for all
> arguments as a list and put it in front of the actual arguments, like
> this:
>
> (dbus-message-internal ...
>   :timeout 100
>   :type '((:array :string) (:array :uint32) (:array :uint32) :boolean)
>   '("a") '(1) '(2) t)
>
> How does that sound?

I see two disadvantages:

- You couldn't use some functions any longer, like
  (dbus-string-to-byte-array "Hello world") mixed with :type prefixed
  arguments. Of course one could add a second function which returns
  just the value list, but is it really helpful? It makes everything
  more complex.

- You would loose the simplification of default types. A list is always
  an array of strings, a string is a string, a natural number is a
  uint32, and so on. You would be forced to write down the type
  explicitely for every argument.

And we could simply use signatures then. Something like

(dbus-message-internal ...
  :timeout 100
  :type "asauaub"
  '("a") '(1) '(2) t)

>> As you see, even I (who has tried to understand the new syntax) am a
>> little bit confused.
>
> Yes, I am realizing how helpful it is to write unit tests, to smoke out
> such pitfalls before landing the feature :-)

Yep. On my todo list is also to study the D-Bus tests at
<http://cgit.freedesktop.org/dbus/dbus-test/>. Maybe we could adapt
something from there.

> Regards,

Best regards, Michael.