Le mar. 9 juin 2020 à 19:19, Marcel Taeumel <marcel.taeumel at hpi.de> a
écrit :

> Hi, there.
>
> I want to understand how data models for 64-bit influence the image side
> of Squeak FFI.
> https://en.wikipedia.org/wiki/64-bit_computing#64-bit_data_models
>
> I discovered that there are test functions in the SqueakFFIPrims plugin
> (sqFFITestFuncs.c) that expect actual "long" types. So, I assume that that
> is the "long" from C, not Squeak FFI, and thus different on 64-bit Windows
> vs. 64-bit Unix.
>
> ffiTest4LongSum
> ffiTestLongs8
> ffiTest8longSum
> ffiTestMixedDoublesAndLongs (!!!)
>
> Only the last one is actually called from FFI Tests, which could be worse.
> Because there is not yet a mechanism in place that considers the data model
> an 64-bit platforms.
>
> So, I think that an atomic type alias for such a C "long" argument could
> be defined like this:
>
> (wordSize = 4 or: [platformName = 'Win32'])
>    ifTrue: ['long' "meaning 'int' in the plugin]
>    ifFalse: ['longlong']
>
> BTW, I documented as I discovered in a serie of SO questions like

> Yet, I have no clue on how to name such a type alias. long_t ?
> native_long_t ? Squeak FFI's 'long' is already taken, and it is always 4
> byte.
>
> Me neither... c_long?

Would it make sense to also query the data model through a primitive and
> add that to FFIPlatformDescription? LLP64, LP64, ILP64, SILP64, ...
>
> YAGNI

Best,
> Marcel
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20200609/0be98abf/attachment.html>