On Dec 15, 2009, at 10:49 PM, Luis Lavena wrote:

> On Wed, Dec 16, 2009 at 3:21 AM, Victor Costan <costan@gmail.com> wrote:

>> Is it feasible to make "FFI.add_sys_types" available on all platforms
>> if it's implemented at runtime?
>> 

> 
> You just made me discard my previous message. I was going to ask
> exactly the same thing.
> 
> I believe these values should be computed at build time and not at
> runtime. runtime calculation and shelling out is a hidden requirement
> and add complexity to debug it.
> 
> Even you put guards to avoid errors, expect bug reports about these
> sys types being missing from people not having the development tools
> for their platforms at hand.

Fully agree. My intention was not to mean that FFI.add_sys_types would begin parsing header files right then figure out the results. Only that the by hiding it behind a method, every implementation is free to make the best decision about how they want to import them.

I'd imagine that typically, ruby-ffi would calculate these values at build time and put them in either a .rb file that you include, or perhaps the FFI.add_sys_types method would be a C method that adds the types, thereby using the compiler itself to do most of the work.

Again, the method is a simple abstraction to allow each implementation the most flexibility with the least pain.

> 

>> I'm thinking ideally the ffi gem will come prebuilt, at least for OSX
>> and Windows, so users might not have Xcode / mingw / Vistual Studio.
>> To me, the biggest advantage of using ffi is not having to deal with
>> all the extension building stuff on a per-gem basis.
>> 

> 
> On this field, I'm working on adding an option to rake-compiler to use
> and abuse of RVM and Pik to build fat binaries on all the platforms,
> for anyone that cares.
> 
> Cheers,
> -- 
> Luis Lavena
> AREA 17
> -
> Perfection in design is achieved not when there is nothing more to add,
> but rather when there is nothing more to take away.
> Antoine de Saint-Exupéry
>