The directorin typemaps in the director methods now use std::move on the
input parameter when copying the object from the stack to the heap prior
to the callback into the target language, thereby taking advantage of
move semantics if available.
Enhance SWIGTYPE "out" typemaps to use std::move when copying
objects, thereby making use of move semantics when wrapping a function returning
by value if the returned type supports move semantics.
Wrapping functions that return move only types 'by value' now work out the box
without having to provide custom typemaps.
The implementation removed all casts in the "out" typemaps to allow the compiler to
appropriately choose calling a move constructor, where possible, otherwise a copy
constructor. The implementation alsoand required modifying SwigValueWrapper to
change a cast operator from:
SwigValueWrapper::operator T&() const;
to
#if __cplusplus >=201103L
SwigValueWrapper::operator T&&() const;
#else
SwigValueWrapper::operator T&() const;
#endif
This is not backwards compatible for C++11 and later when using the valuewrapper feature
if a cast is explicitly being made in user supplied "out" typemaps. Suggested change
in custom "out" typemaps for C++11 and later code:
1. Try remove the cast altogether to let the compiler use an appropriate implicit cast.
2. Change the cast, for example, from static_cast<X &> to static_cast<X &&>, using the
__cplusplus macro if all versions of C++ need to be supported.
Issue #999Closes#1044
More about the commit:
Added some missing "varout" typemaps for Ocaml which was falling back to
use "out" typemaps as they were missing.
Ruby std::set fix for SwigValueWrapper C++11 changes.
This is necessary for regex-like renames (where you can't use the #define trick
as is done in many of the %keywordwarn directives). It's now unnecessary to print
the "renaming to '`x`'" code explicitly by the kw.swg files.
SWIG_SetPointerZval() requires the zval structure passed to be
initialised (so it can handle the constructor case where the zval is
already initialised as a PHP object.
I couldn't think of a suitable regression test for this, but it fixes 2
issues detected by running director_basic under valgrind.
The typemaps for long long and unsigned long long didn't handle a
string input correctly, and long_long_runme.php had a flawed test
in this case.
Fixes#2155
It seems instanceof_function_ex() isn't a viable option for PHP 7.3 and
earlier - the implementation is strangely wrong.
So let's just provide a compatibility implementation which does what
zend_class_implements_interface() does in 8.x and use that for all 7.x
so there are fewer variants to worry about testing.
The sense of parameter 3 of instanceof_function_ex() appears to have
changed in PHP 7.4 and we need to pass 0 for older versions and 1 for
PHP 7.4. Words fail me.
Since the switch to wrapping classes using PHP's C API, we now
internally need to be able to tell if a PHP object is of or derived
from a class that is wrapped by SWIG so we know if we can offset the
zend_object pointer to get to the swig_object_wrapper. If we try to
do this to an object which isn't wrapped by SWIG then we invoke C/C++
undefined behaviour (and typically get a segmentation fault).
This check is implemented by having a SWIG\wrapped empty interface which
we make all SWIG-wrapped classes implement simply so we can test for it
to detect such classes.
Fixes#2125
We have the swig_type_info available and SWIG_TypeCheckStruct
is more efficient because it uses a pointer comparison instead of the
string comparison SWIG_TypeCheck uses (this change speeds up `make
check-php-test-suite` by about 10%).
Add PHP keyword 'readonly' (added in 8.1) to the list SWIG knows to
automatically rename. This keyword is special in that PHP allows it to
be used as a function (or method) name.
Previously the zend_class_entry for Foo was named SWIGTYPE_Foo_ce, but
this can collide in some cases - e.g. if there's a class named p_Foo
then its zend_class entry will be SWIGTYPE_p_Foo_ce, but that's the same
as the swig_type_info for a class named p_Foo_ce.
Do more initialisation at module load time.
Use a shared set of handlers for cases when the C/C++ object is
destroyed with free().
Most of the code in the free_obj and create_object handlers is the
same for every wrapped class so factor that out into common functions.
This typemap which would wrap C++ bool as PHP int is later overridden
by another which wraps it as PHP bool. The current result is what
we want so just remove the redundant one.
The correct macro to test is PHP_MAJOR_VERSION so these two PHP 8 cases
weren't ever used, which hid that the PHP8 version of the code was
broken in one of them.
Highlighted in #2113.
`SWIG_ErrorCode()`, `SWIG_ErrorMsg()`, `SWIG_FAIL()` and `goto thrown;`
are no longer supported (these are really all internal implementation
details and none are documented aside from brief mentions in CHANGES
for the first three). I wasn't able to find any uses at least in FOSS
code via code search tools.
If you are using these:
Use `SWIG_PHP_Error(code,msg);` instead of `SWIG_ErrorCode(code);
SWIG_ErrorMsg(msg);` (which will throw a PHP exception in SWIG >= 4.1
and do the same as the individual calls in older SWIG).
`SWIG_FAIL();` and `goto thrown;` can typically be replaced with
`SWIG_fail;`. This will probably also work with older SWIG, but
please test with your wrappers if this is important to you.
Fixes#2014
Parameter type errors and some other cases in SWIG-generated wrappers
now throw a PHP exception, which is how PHP's native parameter handling
deals with similar situations.
See #2014, but not closing yet as there may be more cases to convert.
Most pre-defined interfaces are accessible via zend_class_entry*
variables declared in the PHP C API - we can use these to add
an interface at MINIT time (rather than having to wait until RINIT to
look up by name) by having a mapping from PHP interface name to them.
This will also be a little faster than looking up by name.
Closes#2013
PHPCN(x) does a string compare of x with the lower-cased class name,
so x needs to be in lowercase or else the entry has no effect. The
entries for TRUE, FALSE and NULL weren't working as a result.
Add some missing entries, remove some long obsolete entries (from
the "ming" extension for generating SWF files, which was split
out from PHP core in 2008), and entry for "static" as a reserved class
name (`static::` is used for late static bindings, but attempting to
name a PHP class `static` fails because `static` is a keyword and we
also list it as such).
Eliminate redundant and unused includes.
Only include the minimum headers needed before the PHP_MAJOR_VERSION
check in case future PHP versions remove some of the headers we
include.
With modern PHP it only works with the CLI version of PHP, so it's
better to direct users to load the extension via "extension=" in
php.ini.
Suggested by ferdynator in #1529.
A PHP exception now gets translated to a C++ exception to skips over C++
code to get back to PHP, avoiding the need to gate every directorout
typemap on EG(exception).
$needNewFlow in an output typemap is now only relevant when wrapping
to a PHP __construct method, and there the return type is known so
factory.i isn't useful.
