Re: [Pharo-dev] [Vm-dev] [squeak-dev] valid pragma syntax

On 12 Jan 2016, at 21:12, Nicolai Hess <nicolaihess@gmail.com> wrote:

2016-01-12 20:57 GMT+01:00 Esteban Lorenzano <estebanlm@gmail.com <mailto:estebanlm@gmail.com>>:

well… UnifiedFFI, UFFI or whatever it will be called already does that… as you know. Currently we are using “older” NB way: as a method call, then solving it in first call. This is nice because is compatible and easies some stuff. But final objective is to provide a pragma and solve it in compilation time.

Then we can 100% deprecate the invalid pragma old FFI uses today.

But how do we solve the issue?

Well, we should add an #visitFFIPragmaNode: who can deal with it, in the mean time :) Esteban

At least, that’s what we have talk :)

Esteban

...

On 12 Jan 2016, at 20:06, Nicolai Hess <nicolaihess@gmail.com <mailto:nicolaihess@gmail.com>> wrote:

2016-01-10 16:57 GMT+01:00 Eliot Miranda <eliot.miranda@gmail.com <mailto:eliot.miranda@gmail.com>>:

Hi Nicolai,

On Sun, Jan 10, 2016 at 3:16 AM, Nicolai Hess <nicolaihess@gmail.com <mailto:nicolaihess@gmail.com>> wrote: pharo fogbugz issue: 17359 MessageNotUnderstood: receiver of "keywords" is nil

is this valid pragma syntax:

<apicall: bool 'SetCursorPos' (long long) module: 'user32.dll'>

As others have said this is not a valid pragma. A valid pragma is a message expression (a message with no receiver) that has only literal arguments. There is one exception, to accept a variable name for the error code in a primitive invocation.

And as Levente has said it is a goal of the new FFI to get rid of the old syntax you show above.

Thanks Eliot.

I think I will wait what Esteban will say to this, he put those classes in the image, maybe he had a plan for the new ffi/apicall syntax

And as Igor Stasenko discovered /any/ C signature can be written as a literal Array, providing we allow underscores in Symbols (which is a preference). So for example, the heapsort declaration from the standard C library:

int heapsort(void *base, size_t nel, size_t width, int (*compar)(const void *, const void *))

could be written as a pragma via e.g.

<ffiCCall: #(int heapsort(void *base, size_t nel, size_t width, int (*compar)(const void *, const void *)))>

in which case it is equal to

<ffiCCall: #(#int #heapsort #( #void #* #base #, #size_t #nel #, #size_t #width #, #int #( #* #compar ) #( #const #void #*, #const #void #*)))>

which, while a massive hack, is pretty neat. The only difficulty in parsing the above s that void *, is #void #*, not #void #* #,.

So for me, the natural pragma syntax for the new FFI should be a single keyword starting with ffi that includes the language with which to parse the pragma's argument, followed by a literal Array containing the signature (rather than e.g. a String). I'd also provide a special-purpose pretty-printer for literal arrays that would be used to pretty-print the above, e.g. when decompiling.

This would easily allow extensions such as

<ffiCcall: #(wchar_t * wcschr(const wchar_t *s, wchar_t c)) arg1is: #UnicodeString arg2is: Character>

to guide a marshalling engine in providing automatic checks and conversions.

[Slightly OT] Note that I'm already using this syntax in Alien callbacks because one thing it does is to normalize the signature, eliminating the significance of most whitespace (the #*, above being one example where it fails). So a Callback for heapsort above looks like

voidstarvoidstarRetint: callbackContext sp: spAlien <signature: #(int (*)(const void *, const void *)) abi: #IA32> ^callbackContext wordResult: (block value: (Alien forPointer: (spAlien unsignedLongAt: 1)) value: (Alien forPointer: (spAlien unsignedLongAt: 5)))

or voidstarvoidstarRetintARM32: callbackContext sp: spAlien <signature: #(int (*)(const void *, const void *)) abi: #ARM32> ^callbackContext wordResult: (block value: (Alien forPointer: (spAlien registerAt: 1)) value: (Alien forPointer: (spAlien registerAt: 2)))

Hmmm, if Callback had subclasses for each ABI then the above would become

IA32Callback methods for signature voidstarvoidstarRetint: callbackContext sp: spAlien <signature: #(int (*)(const void *, const void *))> ^callbackContext wordResult: (block value: (Alien forPointer: (spAlien unsignedLongAt: 1)) value: (Alien forPointer: (spAlien unsignedLongAt: 5)))

ARM32Callback methods for signatures voidstarvoidstarRetint: callbackContext sp: spAlien <signature: #(int (*)(const void *, const void *))> ^callbackContext wordResult: (block value: (Alien forPointer: (spAlien registerAt: 1)) value: (Alien forPointer: (spAlien registerAt: 2)))

Much nicer. I shall make it so.

our (pharo) parser can not parse this (does not recognizes the "selector parts" (apicall:module:) right)

Any recommendations how to fix this - fix RBParser or - change this pragma, for example, like this <apicall: #(bool SetCursorPos (long long)) module: 'user32.dll'>

nicolai

_,,,^..^,,,_ best, Eliot