[cfe-commits] [PATCH 4/4] [clang.py] TranslationUnit API improvements

[Gregory Szorc]

On Fri, Apr 13, 2012 at 3:38 AM, Manuel Klimek <klimek at google.com> wrote:
> What's ERROR_OK needed for? It looks like it's not currently used; it
> actually looks like it's impossible to ever raise an exception with
> it...

That's true. I added it for parity with the C API. It can safely be removed.

> +    def __init__(self, ptr=None, filename=None, index=None):
> +        """Create a TranslationUnit instance.
> +
> +        Instances can be created in the following ways:
> +
> +          * By passing a pointer to a c_object_p instance via ptr. This is
> +            an internal mechanism and should not be used outside of this
> +            module.
>
> This interface seems strange - why have the mixture of 2 constructors in one?

Why not?

Unfortunately, I can't find any reputable style guidelines to defend
either perspective. The closest I have is
http://stackoverflow.com/questions/682504/what-is-a-clean-pythonic-way-to-have-multiple-constructors-in-python.
And, that seems to indicate a mix of an "overloaded" __init__ with
@classmethod is preferred. But, that's just one SO question.

Is this particular case, a TranslationUnit is ultimately instantiated
from a c_object_p "ptr." If we limited __init__ to a single
instantiation mode, we'd have to pass a c_object_p and since these are
internal to the module, __init__ wouldn't be an external API. In other
words, we'd be throwing __init__ away. Since Python programmers look
to __init__ first, I think this would be inconvenient. From an
external perspective, TranslationUnit still only has 1 instantiation
mode. If it had more, I'd definitely favor adding @classmethods to
cover each. I'm not against it today: I just see no reason for it.

Anyway, as I typed this, I realized that we need an additional
constructor mode: from source file (e.g. Index.parse). Let me code up
a new patch and we'll see what you think.