Re: [ANN] Lua-SDL2 rc1 available for testing!

[Philipp Janda <siffiejoe@...>]

Am 26.04.2014 12:38 schröbte David Demelier:
> Hi there,

Hi!

> I'm very happy to announce the first release candidate of Lua-SDL2
> [1]. As I already announced a few
> months ago, Lua-SDL2 is a pure C binding against SDL 2.0.3.
>
> [...]
>
> * A few functions are not implemented, it's the case of Texture:lock
> for instance which is terribly
>    hard to implement from Lua because the C side expect to cast the
> pixel pointer to the underlyling
>    sized array which in Lua is impossible to do. Please provide me
> feedback on how you want it to
>    be implemented.

That depends on what you want the Lua code to do with a locked texture. 
For my own binding I considered using something along the lines of

    myTexture:with_lock( function( modifypixels )
      -- code using the modifypixels function ...
    end )

but in the end I simply didn't implement direct access to pixel data in 
my binding (it has a rather narrow focus). `with_lock` would `pcall` 
(not `pcallk`![*]) the given function and pass a closure that has access 
to the pixel data and the pitch as argument. The Lua code would only 
access the pixel data via that function. After the `pcall` returns the 
upvalues of `modifypixels` get nil'ed using `lua_setupvalue` (for 
safety), and `SDL_UnlockTexture` is called before `with_lock` returns ...

> * Feedback, provide me some enhancement you may find obvious or
> better. In the code or in the documentation :-).

I don't know if that's intended, but your binding is rather low-level: 
It can crash in certain situations (e.g. if a renderer is collected 
before its textures, or window before renderer, etc.), and it would be 
easy to call `SDL_Quit` automatically via garbage-collection so that the 
binding doesn't leak if there is a non-local exit (e.g. an error, or a 
yielded coroutine that gets collected). Another minor point: If 
`lua_newuserdata` ever raises an error you will leak the corresponding 
SDL object.

Aside from that it looks very good (and I probably would have used it 
instead of rolling my own if had known about it a week ago)!

> Enjoy this binding :-).
>
> Kind regards,
> David.

Philipp

  [*]: This is one of those rare cases where you *don't* want a 
yieldable `pcall` ...