Re: [Vala] embedding c code snippet in vala source.

[Yu Feng <rainwoodman gmail com>]

On Fri, 2008-08-22 at 10:44 +0200, Jürg Billeter wrote:
> Hi Yu,
> 
> On Fri, 2008-08-22 at 02:58 -0400, Yu Feng wrote:
> > Which do you prefer if you want to split a string? mysplit or split?
> > 
> >     [CCode (cname = "g_strmysplit", snippet = "gchar ** rt = g_strsplit(this,
> > delimiter, max_tokens); *result_length1 = rt?g_strv_length(rt):0; return rt;")]
> >     public string[] mysplit (string delimiter, int max_tokens = 0);
> > 
> >     [CCode (cname = "g_strsplit")]
> >     [NoArrayLength]
> >     public string[] split (string delimiter, int max_tokens = 0);
> 
> While mysplit is more comfortable to use right now as it correctly
> handles the null-terminated array, reading C code snippets embedded into
> a string in a .vapi file is not what I'm looking forward to.
> 
> Support for null-terminated arrays is planned, so the second split
> binding will work the same way without major additions in the .vapi file
> as soon as this is implemented.
> 
> As far as I understand it, you're proposing to not handle this at all in
> the compiler and just add a snippet attribute to any function that
> doesn't conform exactly to the default translation. 
Exactly. In my opinion, the less a compiler embeds for gluing, the more
it can binds to.

> While this would
> probably make it possible to drop some other attributes, reading and
> writing bindings like this would be a major pain.
I am not sure which is more pain. Begging hard for vala to change as one
wishes or to write some little workaround as one wishes? I would like to
emphasis again the snippets (or cexpressions, if snippet lead to a image
of copy-pasting code) are for rare cases where 
(1) no current vala quirks can work,
(2) no way to modify the library being bound.

> 
> I don't want to have dozens of g_strv_length calls in my .vapi file if I
> can fix this just once in the compiler. Also, this would require a lot
> more complicated metadata for generated bindings.
There won't be dozens. I understand you wish the vapi to be all built
from vapigen. But this is impossible in a near future, right? (At least
what we have seen so far is ABI for GLib2 is unsuitable)

> 
> Maybe I'm just missing an interesting use case, however, I currently
> fail to see where the snippet approach makes it more comfortable to
> write or use bindings. Can you provide examples where you think that it
> makes more sense to handle it in a snippet instead of the compiler code?
> I'm not convinced by the g_strv_length example as this would clutter the
> bindings too much, in my opinion.

Again it is for rare cases. I am thinking hard to make the snipples
working well, and I'll collect more cases later. I don't expect there to
be much though. Should it be much fewer than [NoArrayLength] stuff.

BTW: I found a way to kill the vast number static wrappers in DBus
dynamic methods. File a new bug?

Best,

Yu
> 
> Jürg
> 
>