Stas Boukarev authored Jan 12, 2013 and Luís Oliveira committed Feb 21, 2013

Fixes issue with ABCL not finding things like "strcmp" when no libraries
are loaded.

Alejandro R Sedeño authored Jan 03, 2013 and Luís Oliveira committed Jan 21, 2013

Use C-WRITE instead of C-PRINT-SYMBOL when emitting types to deal with
the fact that some types are more than just one symbol now.

e.g. 'foo vs '(:struct foo) and '(:pointer (:struct foo))

Fixes: Bug 1065432

More library names and fix :sizet typo for windows.  Expand affirmative return from structure-by-value-p to include typedefs, so that foreign function calls by value may use the typedefed names to declare variable.  Includes tests.

Joshua Elsasser authored Dec 15, 2012 and Luís Oliveira committed Jan 06, 2013

This is especially useful when the type and count have been groveled.

Joshua Elsasser authored Nov 24, 2012 and Luís Oliveira committed Jan 06, 2013

This means that forms such as the following are allowed:

(cstruct foo "struct foo"
  (mybar "mybar" :type (:struct bar)))

Joshua Elsasser authored Nov 22, 2012 and Luís Oliveira committed Jan 06, 2013

This allows % to be safely used in lisp symbol names which pass
through the groveler, such as the following:

(cvar ("errno" %errno) :int)

Joshua Elsasser authored Nov 22, 2012 and Luís Oliveira committed Jan 06, 2013

This means that integer size and signedness is determined by the
groveler in cases like this:

(cstruct autotype "struct mystruct"
 (myint "myint" :type :auto)
 (myarray "myarray" :type :auto :count :auto))

Joshua Elsasser authored Nov 22, 2012 and Luís Oliveira committed Jan 06, 2013

The :count :auto logic assumed the struct member was an array which
filled the remainder of the struct, and would generate incorrect cffi
cstruct forms when any other struct members followed one declared
:count :auto.

The following C struct definition and groveler cstruct form
demonstrate this:

struct charbuf {
	char three_chars[3];
	int clobbered_int
};

(cstruct charbuf "struct charbuf"
  (three-chars "three_chars" :type :char :count :auto)
  (clobbered-int "clobbered_int" :type :int))

Juan Jose Garcia Ripoll authored Nov 22, 2012 and Luís Oliveira committed Jan 06, 2013

The previous fix did not work with functions that return no values.
Surround the code with braces, so that the C compiler does not
complain about the extern declaration.

Juanjo Garcia-Ripoll authored Nov 22, 2012 and Luís Oliveira committed Jan 06, 2013

Allow the ECL backend to support different combinations of FFI in
the same code by customizing a single special variable.

Stas Boukarev authored Nov 23, 2012 and Stelian Ionescu committed Nov 24, 2012

consing aliens.

In defcstruct, there was a line, removed in be626770, that set a
default type class name by appending "-TCLASS".  Without this line,
the class will be directly FOREIGN-STRUCT-TYPE instead of a subclass.
This is acceptable if the translate methods are not going to be
redefined, but if they are, any change will apply to all instances
(including instances of subclasses of other structures, which may call
these with call-next-method).  In fact, there is a check in
define-translation-method to prevent redefinition with that macro, but
of course a defmethod would be just as bad.

While in principal one can specify the :class argument, this isn't
always possible, for example, if the structure was made by groveling.
Even so, it is a bad idea to make a default which has the potential
for this kind of harm.

This fix now permits GSLL to compile and load correctly.

easye authored Oct 29, 2012 and Luís Oliveira committed Nov 04, 2012

The interactive restart when reloading callbacks is no longer needed.

A callable function pointer is now returned by CALLBACK and
GET-CALLBACK, which wasn't the case previously.

Now down to 25 failing tests!

Callbacks "automacro-ly" now get a translation layer to convert back
from native types to ones which ABCL expects.  This translation is
currently a work in progress, as not all cases are covered correctly.

(Stas Boukarev) MAKE-FUNCTION-POINTER typo.

Refactored to remove compile warnings about MAKE-IMMEDIATE-OBJECT.

CFFI-SYS::%LOAD-FOREIGN-LIBRARY tries harder to figure out which
library to load.

Docstrings added.

easye authored Oct 29, 2012 and Luís Oliveira committed Nov 04, 2012

The "-fPIC" flag will always be necessary when using GCC-derived compilers as I understand it.

Additionally, Solaris compilation options now unconditionally tries a
64bit build.  Contemporary Solaris (since Solaris 10), are hybrid
32/64 bit environments where "uname -m" reports "i86pc".  Whether one
wants the 32 or 64 bit version of the libraries depends on the type of
the Lisp implementation doing the loading, which is currently not
available in the environment in which make(1) is executing.

This header defines the PRI{fmt}{type} printf format specifiers used by
foreign-type-to-printf-specification.

Nicolas Martyanoff authored Nov 04, 2012 and Luís Oliveira committed Nov 04, 2012

The BITFIELD grovel syntax always used %i to print the foreign values. But
when the base type is larger than an int, the resulting values will be
incorrect.

We introduce the FOREIGN-TYPE-TO-PRINTF-SPECIFICATION function which returns a
printf string containing a single format specification for a foreign type, and
use it in the BITFIELD grovel syntax.

Nicolas Martyanoff authored May 19, 2012 and Luís Oliveira committed Nov 04, 2012

On FreeBSD, GCC only looks in directories of the base
system (/usr/include). We need to manually add /usr/local/include if we want
to be able to bind libraries installed by packages or ports.

Add method expand-to-foreign-dyn-indirect for type foreign-type-alias so that foreign-string-type will be expanded correctly (as a pointer).

Fix an error in the recursive call in libffi-type-pointer for
foreign-struct-type, now parse the slot-type, which is unparsed.  This
makes FSBV.3 and FSBV.4 succeed.  Removed quote from test FSBV.2; this
makes FSBV.2 succeed.

* Instead of a special variable, the parsed struct type instance is now
  annotated to let the type system know whether we're dealing with
  deprecated semantics (struct type canonicalizes to :pointer) or not.

  This approach should hopefully be much less error-prone.

* Disabled translations for bare structs.

* Fixed and commented out some bogus tests.