[RFC] FFI Improvements: Consistency and soliving some problems

Posted 28 days ago by [email protected] — view source

unread

Hi all!

I would like to start discussion about the «FFI Improvements» RFC. At the moment, I do not have the right to create wiki page, so I post it on the github: https://github.com/SerafimArts/php-rfcs/blob/ffi-improvements/rfcs/0000-ffi-improvements.md

The proposal:

Contains improvements regarding the harmonization of supported FFI directives («define FFI_XXX») in different operating modes (cdef and load).
- And he suggests solving the problem of the inability to work with some types of libraries that contain loading of other libraries.

In addition, I want to add that I am not a C developer, and I understand that this reduces the chances of adopting this RFC. However, I think that the proposal is useful and if approved, someone will help to implement it.

Posted 26 days ago by Dan Ackroyd — view source

unread

I would like to start discussion about the «FFI Improvements» RFC. At the moment, I do not have the right to create wiki page, so I post it on the github: https://github.com/SerafimArts/php-rfcs/blob/ffi-improvements/rfcs/0000-ffi-improvements.md

Thanks for starting this discussion. I think FFI definitely needs to
be improved before it will see widespread adoption.

Please note that for backward compatibility, allow option with
passing a string as the second argument to the FFI::cdef() method.

Rather than changing the function signature of FFI::cdef, I think it
would be better to introduce a new function, and then at some point
deprecate and remove the old one (if anyone cares to).

So rather than FFI::cdef() it could be FFI::ghij() or maybe even a
name that is somewhat meaningful.

For FFI_LIB, FFI_SCOPE, and FFI_LIB_DIR your proposal sounds sensible.
It would be really good if someone involved in getting FFI into core
could comment if there was any specific reason for not doing that
already.

One other thing that probably also should be addressed is the
behaviour around closures that was noted in the RFC
https://wiki.php.net/rfc/ffi#php_callbacks:

It's possible to assign PHP closure to native variable of
function pointer type (or pass it as a function argument).
This works, but this functionality is not supported on all libffi
platforms, it is not efficient and leaks resources by the end of
request. It's recommended to minimize the usage of PHP callbacks.

I really don't know what can or should be done for that, but having a
feature that can't be used safely seems like a bad feature.

cheers
Dan
Ack

Posted 26 days ago by [email protected] — view source

unread

One other thing that probably also should be addressed is the
behaviour around closures that was noted in the RFC
https://wiki.php.net/rfc/ffi#php_callbacks :

During work with FFI, I did not encounter problems associated with this. There are more significant ones, for example, the lack of a cast of types in structures:

$struct = FFI::cdef(‘struct Test { char* string_field; }’)->new(‘Test’);
$struct->string_field = «string»; // Error

However, these internal problems including a memory leak when working with anonymous functions, I found it unnecessary in this RFC. Maybe a little later.

So rather than FFI::cdef() it could be FFI::ghij() or maybe even a
name that is somewhat meaningful.

I'm afraid in this case, developers will use an easier to use feature. And this means that just so we can’t deprecate it later.

However, if this argument is not very convincing, then as the name of the new function, I would suggest «FFI::fromSource(...)».

Четверг, 9 июля 2020, 11:01 +03:00 от Dan Ackroyd [email protected]:

I would like to start discussion about the «FFI Improvements» RFC. At the moment, I do not have the right to create wiki page, so I post it on the github: https://github.com/SerafimArts/php-rfcs/blob/ffi-improvements/rfcs/0000-ffi-improvements.md

Thanks for starting this discussion. I think FFI definitely needs to
be improved before it will see widespread adoption.

Please note that for backward compatibility, allow option with
passing a string as the second argument to the FFI::cdef() method.

Rather than changing the function signature of FFI::cdef, I think it
would be better to introduce a new function, and then at some point
deprecate and remove the old one (if anyone cares to).

So rather than FFI::cdef() it could be FFI::ghij() or maybe even a
name that is somewhat meaningful.

For FFI_LIB, FFI_SCOPE, and FFI_LIB_DIR your proposal sounds sensible.
It would be really good if someone involved in getting FFI into core
could comment if there was any specific reason for not doing that
already.

One other thing that probably also should be addressed is the
behaviour around closures that was noted in the RFC
https://wiki.php.net/rfc/ffi#php_callbacks :

It's possible to assign PHP closure to native variable of
function pointer type (or pass it as a function argument).
This works, but this functionality is not supported on all libffi
platforms, it is not efficient and leaks resources by the end of
request. It's recommended to minimize the usage of PHP callbacks.

I really don't know what can or should be done for that, but having a
feature that can't be used safely seems like a bad feature.

cheers
Dan
Ack

--

To unsubscribe, visit: https://www.php.net/unsub.php

--
Kirill Nesmeyanov