There are far more type names available to use than are suggested by the wiki page. This makes it very confusing what type to use for, say, size_t, when :size_t would actually just work. Some note about this would be nice.

When calling Struct.new, it calls struct_malloc (Struct.c line 176) which creates a MemoryPointer and passes in the parameter to zero/initialize the memory. There are many situations where this is completely unnecessary work.

e.g.
struct foo {
int bar;
long baz;
void *qux;
};

struct foo foo_var;

api_call_initializes_struct(&foo_var);

The API is just going to overlay the memory with its own data or zero it out itself.

So, please provide a way to pass through an argument to Struct.new that disables memory initialization.

I've been working on a FFI library for libudis86, which uses a certain enum (ud_vendor) in a Struct definition. In my version I did the same, although when accessing the enum field in the specs I receive this exception:

ArgumentError in 'FFI::UDis86::UD create should accept a :vendor option'
get not supported for FFI::StructLayout::Field
./spec/ud_spec.rb:23:
/usr/lib64/ruby/1.8/timeout.rb:53:in `timeout'

The offending spec:

it "should accept a :vendor option" do
  ud = UD.create(:vendor => :amd)
  ud.vendor.should == :amd
end

And the Struct layout:

:pc, :uint64,
:vendor, :ud_vendor,
:mapen, :pointer,

I am using FFI 0.6.0 from github.

Pointer#get_byte
ref:http://groups.google.com/group/ruby-ffi/browse_thread/thread/6fc761ef5af7f1f4

http://groups.google.com/group/ruby-ffi/browse_thread/thread/1712f5507b15b467

appears to be binary gem build related

While developing FFI bindings to libmsgpack (github.com/postmodern/ffi-msgpack), I noticed that the msgpack_pack_object function would always return -1 for the Structs I passed to it. It appeared msgpack_pack_object was unable to recognize the type field of the msgpack_object struct and returned -1.

int msgpack_pack_object(msgpack_packer* pk, msgpack_object d)
{
switch(d.type) {
    // ....
        default:
    return -1;
}
}

I later verified this by passing the same msgpack_object structs to msgpack_object_print:

require 'ffi/msgpack'

obj = FFI::MsgPack::MsgObject.new_object(1)
# => #<FFI::MsgPack::MsgObject:0x000000020bbe00>
obj[:type]
# => :positive_integer
obj.pointer.get_bytes(0,8)
# => "\x03\x00\x00\x00\x00\x00\x00\x00"
obj[:values][:u64]
# => 1
stdout = FFI::LibC.fdopen(0,'w')
# => #<FFI::Pointer address=0x2252d20>
FFI::LibC.fflush stdout
# => 0
FFI::MsgPack.msgpack_object_print(stdout,obj)
# => nil
FFI::LibC.fflush stdout
#<UNKNOWN 32488 138993971994384>
# => 0

When msgpack_object_print cannot recognize the type field of the msgpack_object struct it will print this information:

void msgpack_object_print(FILE* out, msgpack_object o)
{
    switch(o.type) {
    // ...
    default:
    // FIXME
    fprintf(out, "#<UNKNOWN %hu %"PRIu64">", o.type, o.via.u64);
}

There needs to be a clean API to change FFI back to using RTLD_GLOBAL instead of RTLD_LOCAL. I have a library that needs the global option.

because of this change:
http://github.com/ffi/ffi/commit/e3e29ce7a1601eb14ecf4b1e3cfe5273b8bf482d

Multi-dimentional aren't supported, as discussed at http://groups.google.com/group/ruby-ffi/browse_thread/thread/378467705717660e.

Therefore structures like:

struct matrix {
    float m[4][4];
}

Can't be binded.

Hi there,

I've found that the :varargs type doesn't work correctly if you have defined enums before you attach the function. Here's a sample code using GTK functions:

#!/usr/bin/env ruby
#Encoding: UTF-8
require "ffi"
module GTK
  extend FFI::Library

  ffi_lib "gtk-x11-2.0.so"

  enum :type, [ :gtk_message_info, 
                        :gtk_message_warning, 
                        :gtk_message_question, 
                        :gtk_message_error, 
                        :gtk_message_other
                      ]

  enum :buttons, [:gtk_buttons_none, 
                            :gtk_buttons_ok, 
                            :gtk_buttons_close, 
                            :gtk_buttons_cancel, 
                            :gtk_buttons_yes_no, 
                            :gtk_buttons_ok_cancel
                            ]

  attach_function(:gtk_init, [:pointer, :pointer], :void)
  attach_function(:gtk_set_locale, [], :void)
  attach_function(:gtk_message_dialog_new, [:pointer, :int, :type, :buttons, :string], :pointer)
  attach_function(:g_object_set, [:pointer, :varargs], :void) #Here's :varargs used
  attach_function(:gtk_dialog_run, [:pointer], :int)
  attach_function(:gtk_widget_destroy, [:pointer], :void)
end

GTK.gtk_init(nil, nil)
GTK.gtk_set_locale()
dialog = GTK.gtk_message_dialog_new(nil, 0, :gtk_message_info, :gtk_buttons_ok, "Title")
GTK.g_object_set(dialog, :string, "secondary-text", :string, "text", :pointer, nil)
GTK.gtk_dialog_run(dialog)
GTK.gtk_widget_destroy(dialog)

If I put the attaching of g_object_set before the enum definitions, this code works. Otherwise, this error results:

/opt/rubies/ruby-1.9.1-p378/lib/ruby/gems/1.9.1/gems/ffi-0.6.3/lib/ffi/types.rb:46:in `find_type': Unable to resolve type 'string' (TypeError)
  from /opt/rubies/ruby-1.9.1-p378/lib/ruby/gems/1.9.1/gems/ffi-0.6.3/lib/ffi/variadic.rb:47:in `call'
  from (eval):3:in `g_object_set'
  from gtk.rb:36:in `<main>'

Marvin

Pointer#get_uint
ref: http://groups.google.com/group/ruby-ffi/browse_thread/thread/6fc761ef5af7f1f4

or perhaps Pointer#get_int(byte_count_you_want) in case people want to read words or what not.

The code below works fine in MRI but not JRuby. The problem appears to be something strange going on when attempting to inherit from base types like FFI::Pointer, MemoryPointer, Buffer, etc. and overriding initialize() with a call to super().

http://gist.github.com/275646

Am I doing something wrong here? Guessing this has something to do with multiiple creators in the java implementation but that's just a shot in the dark. I dont really understand how JRuby's FFI is implemented well enough to investigate.

Something with overridden pretty_print or 'inspect' appears to be broken. When using any InlineArray struct field, IRB raises an exception as it fails to prettyprint the Array with an error like the following:

NoMethodError: to_s not defined for this array type

Please refer to http://gist.github.com/279607 for a full example, including backtrace.

Hello,

I ran into a "Cannot set :string fields" error from memory_op_put_strptr() in ext/ffi_c/AbstractMemory.c:389.

Is this functionality just not implemented yet (should have raised a NotImplementedError), or is there a reason why users are prohibited from assigning the result of FFI::MemoryPointer.from_string() to a :string field inside a FFI::Struct?

Thanks for your consideration.

On MRI 1.9.1p243 (mingw32) and JRuby 1.4.0 running on a Windows system with the 0.5.1 FFI gem, you can specify the library to use with a Symbol like ffi_lib :kernel32

Attempting the same in a Ruby 1.8 environment gives a can't convert Symbol into String(TypeError) in map_library_name of ffi/ffi.rb due to File.basename in 1.8 not accepting Symbols.

While providing a String library name to ffi_lib works on both 1.8 and 1.9, it is inconsistent with FFI::Library#attach_function which accepts a Symbol as the C function name.

Hi, trying to install under Ruby 1 8 7, debian
Thanks!

brianwolf@lightning:~$ gem install ffi WARNING: Installing to ~/.gem since /usr/lib/ruby/gems/1.8 and
/usr/bin aren't both writable.
Building native extensions. This could take a while...
ERROR: Error installing ffi:
ERROR: Failed to build gem native extension.

/usr/bin/ruby1.8 -rubygems /usr/lib/ruby/gems/1.8/gems/rake-0.8.7/bin/rake RUBYARCHDIR=/home/brianwolf/.gem/ruby/1.8/gems/ffi-0.6.2/lib RUBYLIBDIR=/home/brianwolf/.gem/ruby/1.8/gems/ffi-0.6.2/lib
/usr/bin/ruby1.8: no such file to load -- ubygems (LoadError)

Gem files will remain installed in /home/brianwolf/.gem/ruby/1.8/gems/ffi-0.6.2 for inspection.
Results logged to /home/brianwolf/.gem/ruby/1.8/gems/ffi-0.6.2/gen/gem_make.out
brianwolf@lightning:~$ ruby --version
ruby 1.8.7 (2008-08-11 patchlevel 72) [x86_64-linux]

http://github.com/ffi/ffi/tree/master/spec/ffi/

On Dec 4, 2009, at 2:51 PM, Brian Ford wrote:
The specs need quite a bit of cleanup to conform to the rubyspec style guide.

http://rubyspec.org/wiki/rubyspec/Style_Guide

mike dalessio wrote:
We've had a couple of threads about this in the last 6 months, and although
I think FFI developers are in favor of a unifying / standard test suite,
nobody has had the time or the will to be point-man for the project.

I'm willing to help unify and standardize, if someone is willing to
coordinate and pester all the people who need to be pestered.

Ernie wrote:
Deal.

http://groups.google.com/group/ruby-ffi/browse_thread/thread/e9a24d52f776ab04

rdp@li49-39:~/dev/downloads/ffi$ rake
(in /home/rdp/dev/downloads/ffi)
rake-compiler must be configured first to enable cross-compilation
rake-compiler must be configured first to enable cross-compilation
rake-compiler must be configured first to enable cross-compilation
rake-compiler must be configured first to enable cross-compilation
cd build/i686-linux/ffi_c/1.8.6
GNU Make 3.81
Copyright (C) 2006 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.
There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE.

This program built for i486-pc-linux-gnu
make
Configuring libffi
/bin/sh: Can't open ../../../../ext/ffi_c/libffi/configure
make: *** [/home/rdp/dev/downloads/ffi/build/i686-linux/ffi_c/1.8.6/libffi/.libs/libffi_convenience.a] Error 2
rake aborted!
Command failed with status (2): [make...]

(See full trace by running task with --trace)

however running ruby extconf.rb && make within the ext/ffi_c directory works, so I think it's just a path problem.
-r

I've added a large enum (560 elements) named :ud_mnemonic_code to ffi-udis86. I've confirmed that I can manually map Integers to Symbols with the enum. But when I use the enum type in function or struct definitions, all Integers are mapped to nil.

ud = FFI::UDis86::UD.create { |ud| 0x90 }
ud.next_insn
# => 1
ud.to_asm
# => "nop "
ud[:mnemonic]
# => nil
ud.offset_of(:mnemonic)
# => 200
ud.to_ptr.get_uint(200)
# => 322
FFI::UDis86.enum_type(:ud_mnemonic_code)[322]
# => :ud_inop

http://github.com/sophsec/ffi-udis86/blob/master/lib/udis86/types.rb#L19
http://github.com/sophsec/ffi-udis86/blob/master/lib/udis86/ud.rb#L30

Using multiple modules to avoid namespace pollution, the automagic detection of the library name fails:

module Windows
  module Kernel32
    extend FFI::Library
    # ffi_lib 'kernel32' # need to comment out to make it work

    attach_function :AllocConsole, [], :int32
    attach_function :GetLastError, [], :uint32
  end
end

Producing the following output while the ffi_lib invoke is still commented:

C:/Users/Luis/.gem/ruby/1.8/gems/ffi-0.5.1-x86-mingw32/lib/ffi/library.rb:77:in `attach_function': Function 'AllocConsole' not found in [[current process]] (FFI::NotFoundError)

I think this is because is using @self.name@ and not excluding the nested modules of it.

http://gist.github.com/367604

running

$ ruby extconf.rb
$ make
$ ruby issue.rb

results in
/Users/daniel/.rvm/gems/ruby-1.9.1-p378/gems/ffi-0.6.3/lib/ffi/library.rb:61:in block in ffi_lib': Could not open library 'issue': dlopen(issue, 5): image not found. Could not open library 'libissue.dylib': dlopen(libissue.dylib, 5): image not found (LoadError) but it should probably just load theissue.bundle`.
ffi_lib ["issue.bundle", "issue"]
can be used as a workaround.

Produced on OSX 10.6 Ruby 1.9.1

I just tried to load s dylib on OSx 10.6 with ffi-0.6.3. Loading works fine but i get this error while attaching the function in my lib:

`attach_function': Function 'calc_a_lot' not found in libcalcalot.dylib

g++ compile line:
g++ -dynamiclib -o libcalcalot.dylib calc_a_lot.cpp

nm says the function is there:
.....
0000000000001016 (__TEXT,__text) external _Z9calcalotPKcS0
.....

I just upgraded to FFI 0.6.0 and noticed FFI functions that return uint types, attempt to return a String on Ruby 1.9.1. Example from the ffi-udis86 library (http://github.com/sophsec/ffi-udis86).

attach_function :ud_disassemble, [:pointer], :uint

require 'udis86'
FFI::UDis86::UD.open('file') do |ud|
  puts ud.next_insn.inspect
end
TypeError: can't convert String into Integer
from /underground/code/sophsec/ffi-udis86/lib/udis86/ud.rb:376:in `ud_disassemble'
from /underground/code/sophsec/ffi-udis86/lib/udis86/ud.rb:376:in `next_insn'
from (irb):3:in `block in irb_binding'
from /underground/code/sophsec/ffi-udis86/lib/udis86/ud.rb:152:in `call'
from /underground/code/sophsec/ffi-udis86/lib/udis86/ud.rb:152:in `block in open'
from /underground/code/sophsec/ffi-udis86/lib/udis86/ud.rb:149:in `open'
from /underground/code/sophsec/ffi-udis86/lib/udis86/ud.rb:149:in `open'
from (irb):3
from /usr/bin/irb19:12:in `<main>'

I'm getting unknown CPU when trying to install the gem on a Sheeva Plug.

/proc/cpuinfo:

root@ubuntu:/var/lib/gems/1.8/gems/ffi-0.6.3/ext/ffi_c# cat /proc/cpuinfo
Processor : Feroceon 88FR131 rev 1 (v5l)
BogoMIPS : 1192.75
Features : swp half thumb fastmult edsp
CPU implementer : 0x56
CPU architecture: 5TE
CPU variant : 0x2
CPU part : 0x131
CPU revision : 1

Hardware : Marvell SheevaPlug Reference Board
Revision : 0000
Serial : 0000000000000000

$ sudo gem install ffi
Building native extensions. This could take a while...
ERROR: Error installing ffi:
ERROR: Failed to build gem native extension.

/opt/local/bin/ruby extconf.rb
checking for ffi.h in /usr/local/include,/opt/local/include... no
checking for ffi_call() in -lffi... yes
checking for ffi_prep_closure()... yes
checking for ffi_raw_call()... no
checking for rb_thread_blocking_region()... no
creating extconf.h
creating Makefile

make
/usr/bin/gcc-4.2 -I. -I. -I/opt/local/lib/ruby/1.8/i686-darwin10 -I. -DRUBY_EXTCONF_H="extconf.h" -I/opt/local/include -D_XOPEN_SOURCE -D_DARWIN_C_SOURCE -I/opt/local/include -fno-common -O2 -arch x86_64 -fno-common -pipe -fno-common -Werror -Wunused -Wformat -Wimplicit -Wreturn-type -arch x86_64 -c AbstractMemory.c
In file included from AbstractMemory.h:37,
from AbstractMemory.c:36:
Types.h:74:17: error: ffi.h: No such file or directory
In file included from Types.h:75,
from AbstractMemory.h:37,
from AbstractMemory.c:36:
Type.h:44: error: expected specifier-qualifier-list before ‘ffi_type’
In file included from AbstractMemory.c:38:
Function.h:52: error: expected specifier-qualifier-list before ‘ffi_type’
make: *** [AbstractMemory.o] Error 1

Gem files will remain installed in /opt/local/lib/ruby/gems/1.8/gems/ffi-0.5.4 for inspection.
Results logged to /opt/local/lib/ruby/gems/1.8/gems/ffi-0.5.4/ext/ffi_c/gem_make.out

The first Windows example seg faults (using ffi 0.6.2). In what way should the example be written such that it would work?
http://wiki.github.com/ffi/ffi/windows-examples

C:\Source\windoze>ruby test3.rb
test3.rb:14: [BUG] Segmentation fault
ruby 1.9.1p243 (2009-07-16 revision 24175) [i386-mingw32]

-- control frame ----------
c:0004 p:---- s:0013 b:0013 l:000012 d:000012 CFUNC  :GetLocalTime
c:0003 p:0049 s:0009 b:0008 l:001b14 d:001f24 EVAL   test3.rb:14
c:0002 p:---- s:0004 b:0004 l:000003 d:000003 FINISH
c:0001 p:0000 s:0002 b:0002 l:001b14 d:001b14 TOP
---------------------------
-- Ruby level backtrace information-----------------------------------------
test3.rb:14:in `GetLocalTime'
test3.rb:14:in `<main>'

Thanks,
-Charles

Would it be possible to add a condition to FFI::ConstantGenerator#dump_constants so that the join does not happen if there is no prefix?

--- a/lib/ffi/tools/const_generator.rb
+++ b/lib/ffi/tools/const_generator.rb
@@ -117,7 +117,7 @@ module FFI

def dump_constants(io)
  @constants.each do |name, constant|
-     name = [@prefix, name].join '.'
+     name = [@prefix, name].join '.' if @prefix
      io.puts "#{name} = #{constant.converted_value}"
  end
end

class BITMAPFILEHEADER < FFI::Struct
  layout(
    :Type, :int16, 0,
    :Size, :int32, 2,
    :Reserved1, :int16, 6,
    :Reserved2, :int16, 8,
    :OffBits, :int32, 10
  )
end

(implementing http://msdn.microsoft.com/en-us/library/dd183374%28VS.85%29.aspx )

BITMAPFILEHEADER.layout returns #<FFI::Type:02E67BD0 size=16 alignment=4>.

The size should be 14, not 16.

The size should maybe be implemented to just look at the highest offset and add the size of that field? Highest offset 10 + size 4 gives the correct size for this struct.

Is there a way to override the alignment? Since that seems to be the issue here, telling it to align by 2 (or 1) instead of 4 would also give the correct size, I think (and then I wouldn't have to give explicit offsets for each member).

Hi,

Trying to get ffi working on a stock CentOS5 system [gcc (GCC) 4.1.2 20080704 (Red Hat 4.1.2-46)] but the native bits of the gem build fail to compile b/c of the warnings as errors flags, which yield dozens of warnings, ending with:

In v0.5.4:
gcc -I. -I. -I/usr/lib/ruby/1.8/i386-linux -I. -I/usr/lib/ruby/gems/1.8/gems/ffi-0.5.4/ext/ffi_c/libffi/include -DRUBY_EXTCONF_H="extconf.h" -fPIC -O2 -g -march=i386 -mcpu=i686 -Wall -fPIC -Werror -Wunused -Wformat -Wimplicit -Wreturn-type -c Call.c
-mcpu=' is deprecated. Use-mtune=' or '-march=' instead.
cc1: warnings being treated as errors
Call.c: In function ‘getString’:
Call.c:415: warning: control reaches end of non-void function
make: *** [Call.o] Error 1

In v0.5.0:
gcc -I. -I. -I/usr/lib/ruby/1.8/i386-linux -I. -I/usr/lib/ruby/gems/1.8/gems/ffi-0.5.0/ext/ffi_c/libffi/include -DHAVE_EXTCONF_H -Werror -Wunused -Wformat -Wimplicit -Wreturn-type -fPIC -O2 -g -march=i386 -mcpu=i686 -Wall -fPIC -c AbstractMemory.c
-mcpu=' is deprecated. Use-mtune=' or '-march=' instead.
cc1: warnings being treated as errors
AbstractMemory.c: In function ‘get_pointer_value’:
AbstractMemory.c:160: warning: control reaches end of non-void function
make: *** [AbstractMemory.o] Error 1

I don't know details...

ffi/lib/ffi/library.rb

def callback(*args)
  raise ArgError, "wrong number of arguments" if args.length < 2 || args.length > 3
  name, params, ret = if args.length == 3
    args
  else
    [ nil, args[0], args[1] ]
  end

  options = Hash.new
  options[:convention] = defined?(@ffi_convention) ? @ffi_convention : :default
  options[:enums] = @ffi_enums if defined?(@ffi_enums)
  options[:blocking] = @ffi_blocking if defined?(@ffi_blocking) # ?
  cb = FFI::CallbackInfo.new(find_type(ret), params.map { |e| find_type(e) }, options)

  # Add to the symbol -> type map (unless there was no name)
  unless name.nil?
    @ffi_callbacks = Hash.new unless defined?(@ffi_callbacks)
    @ffi_callbacks[name] = cb
  end

  cb
end

~/Repositories/sys-uptime-> sudo gem install ffi
Password:
Building native extensions. This could take a while...
ERROR: Error installing ffi:
ERROR: Failed to build gem native extension.

/usr/local/bin/ruby extconf.rb
checking for #include <ffi.h>
... no
checking for ffi_call() in -lffi... no
checking for ffi_prep_closure()... no
checking for ffi_raw_call()... no
checking for rb_thread_blocking_region()... no
creating extconf.h
creating Makefile

make
make: Fatal error in reader: ./libffi.gnu.mk, line 15: Unexpected end of line seen

Gem files will remain installed in /usr/local/lib/ruby/gems/1.8/gems/ffi-0.5.4 for inspection.
Results logged to /usr/local/lib/ruby/gems/1.8/gems/ffi-0.5.4/ext/ffi_c/gem_make.out

~/Repositories/sys-uptime-> gem env
RubyGems Environment:

• RUBYGEMS VERSION: 1.3.5
• RUBY VERSION: 1.8.6 (2009-08-04 patchlevel 383) [i386-solaris2.10]
• INSTALLATION DIRECTORY: /usr/local/lib/ruby/gems/1.8
• RUBY EXECUTABLE: /usr/local/bin/ruby
• EXECUTABLE DIRECTORY: /usr/local/bin
• RUBYGEMS PLATFORMS:
  • ruby
  • x86-solaris-2.10
• GEM PATHS:
  • /usr/local/lib/ruby/gems/1.8
  • /export/home/djberge/.gem/ruby/1.8
• GEM CONFIGURATION:
  • :update_sources => true
  • :verbose => true
  • :benchmark => false
  • :backtrace => false
  • :bulk_threshold => 1000
• REMOTE SOURCES:
  • http://gems.rubyforge.org/

Version 0.5.3, ext/ffi_c/Makefile has the CCFLAG "-fno-common-Werror", which I think should be "-fno-common -Werror" (with a space between the two flags). This causes an unrecognized command line option error.

This code

http://pastie.org/839395

seems to seg fault for me, though in reality I think it should actually return two sets of matching values.

Certainly top_window.read_int should at least work (I think HWND is a pointer to an int), and it appears to seg fault.

ref: http://groups.google.com/group/ruby-ffi/browse_thread/thread/ff20fdb3ec612585

Thanks.

These files are missing from the gemspec at the moment, causing gem install to fail.

ext/ffi_c/ClosurePool.c
ext/ffi_c/ClosurePool.h

if ffi gets vendored in an rails project, as an dependency of capybara, then it just breaks rake in peaces.

• PROJ - Constant not defined in all ffi Rakefiles.
  I had to remove all Rakefiles, and bundle my own version of ffi to get Rake working again.

Don't see this on my local machine (Mac OS X 10.6), but I am seeing it on RHEL 5.3:

...
gcc -I. -I. -I/usr/lib/ruby/1.8/i386-linux -I. -I/home/rails.wincent.com/deploy/releases/20091126200856/vendor/gems/ffi-0.5.3/ext/ffi_c/libffi/include -DRUBY_EXTCONF_H=\"extconf.h\"  -fPIC -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector --param=ssp-buffer-size=4 -m32 -march=i386 -mtune=generic -fasynchronous-unwind-tables -Wall -fno-strict-aliasing  -fPIC-Werror -Wunused -Wformat -Wimplicit -Wreturn-type   -c Pointer.c
cc1: error: unrecognized command line option "-fPIC-Werror"
make: *** [Pointer.o] Error 1

Notice how "-fPIC" and "-Werror" get mashed together, causing the error.

Adding an extra space at the start of "CFLAGS" in the extconf.rb makes it go away:

diff --git a/vendor/gems/ffi-0.5.3/ext/ffi_c/extconf.rb b/vendor/gems/ffi-0.5.3/ext/ffi_c/extconf.rb
index 08ef0b4..6e20fcb 100644
--- a/vendor/gems/ffi-0.5.3/ext/ffi_c/extconf.rb
+++ b/vendor/gems/ffi-0.5.3/ext/ffi_c/extconf.rb
@@ -18,7 +18,7 @@ $defs << "-DHAVE_EXTCONF_H" if $defs.empty? # needed so create_header works
 create_header

 $CFLAGS << "-mwin32 " if Config::CONFIG['host_os'] =~ /cygwin/
-$CFLAGS << "-Werror -Wunused -Wformat -Wimplicit -Wreturn-type "
+$CFLAGS << " -Werror -Wunused -Wformat -Wimplicit -Wreturn-type "

 create_makefile("ffi_c")
 unless libffi_ok

When installing gem an error occurs:

===> Installing for rubygem-ffi-0.5.0
===> rubygem-ffi-0.5.0 depends on file: /usr/local/bin/gem18 - found
===> rubygem-ffi-0.5.0 depends on file: /usr/local/bin/ruby18 - found
===> rubygem-ffi-0.5.0 depends on file: /usr/local/bin/ruby18 - found
===> Generating temporary packing list
===> Checking if devel/rubygem-ffi already installed
/usr/bin/env /usr/local/bin/gem18 install -l --no-update-sources --no-ri --install-dir /usr/local/lib/ruby/gems/1.8 /usr/distfiles/rubygem/ffi-0.5.0.gem -- --build-args
Building native extensions. This could take a while...
ERROR: Error installing /usr/distfiles/rubygem/ffi-0.5.0.gem:
ERROR: Failed to build gem native extension.

/usr/local/bin/ruby18 extconf.rb --build-args
checking for ffi_closure_alloc() in -lffi... no
checking for rb_thread_blocking_region()... no
creating Makefile
creating extconf.h

make
"./libffi.bsd.mk", line 12: Malformed conditional ("${srcdir}" == ".")
"./libffi.bsd.mk", line 14: if-less else
"./libffi.bsd.mk", line 16: if-less endif
make: fatal errors encountered -- cannot continue

Gem files will remain installed in /usr/local/lib/ruby/gems/1.8/gems/ffi-0.5.0 for inspection.
Results logged to /usr/local/lib/ruby/gems/1.8/gems/ffi-0.5.0/ext/ffi_c/gem_make.out
*** Error code 1

Here is content of gem_make.out:
/usr/local/bin/ruby18 extconf.rb --build-args
checking for ffi_closure_alloc() in -lffi... no
checking for rb_thread_blocking_region()... no
creating Makefile
creating extconf.h

make
"./libffi.bsd.mk", line 12: Malformed conditional ("${srcdir}" == ".")
"./libffi.bsd.mk", line 14: if-less else
"./libffi.bsd.mk", line 16: if-less endif
make: fatal errors encountered -- cannot continue

Just starting with ffi. Trying to wrap a library that contains the following:
typedef long double SmiFloat128;

There doesn't seem to be a :long_double type. How can I add that?

gem install ffi
Building native extensions. This could take a while...
ERROR: Error installing ffi:
ERROR: Failed to build gem native extension.

C:/data/Ruby/bin/ruby.exe extconf.rb install ffi
checking for rb_thread_blocking_region()... no
creating extconf.h
creating Makefile

nmake

Microsoft (R) Program Maintenance Utility Version 9.00.30729.01
Copyright (C) Microsoft Corporation. All rights reserved.

makefile(183) : fatal error U1002: syntax error : invalid macro invocation '$'
Stop.

ENV:
windows 7
visual studio express 2008
ruby 1.8.6

Because of this the Rakefile from the gem is unusable (and in Gentoo I'm falling back to download a tarball from the git tag).

Thanks,
Diego

Here is the spec that fails.
http://gist.github.com/205271

I'm running under OSX 10.6.
repro steps:
1: apply patch
2: run rake
3: spec fails
4: vim build/i686-darwin10/ffi_c/1.8.7/Makefile
5: change -O2 to -O0 in the CFLAGS variable
6: rm -rf build/i686-darwin10/ffi_c/1.8.7/.o
7: rm -rf build/i686-darwin10/ffi_c/1.8.7/.bundle
8: rake
9: spec passes

currently it appears you can only specify one library:

module Hello
  extend FFI::Library
  ffi_lib 'msvcrt'
  ffi_lib 'user32'

  ffi_convention :default
  attach_function 'FindWindowA', [ :string, :string], :pointer
end

fails if you switch the ffi_lib commands.

<FFI::NotFoundError: Function 'FindWindowA' not found in [msvcrt]>

It would be convenient if you could specify more than one.
a la

<FFI::NotFoundError: Function 'FindWindowA' not found in [msvcrt, user32]>

etc.

This code in MRI
http://gist.github.com/260343

lib/ffi/library.rb line 99

The documentation in the source for attach_function is incorrect. It states that if you want an alternate name for the method, you have to specify it after the +name+ when, in fact, you have to specify it before the +name+.

Sorry I cannot provide a patch, I do not have time right now to and this is a simple wording fix.

I am defining a callback to ruby that gets registered with a C library. To do so I'm using the callback api. In 0.5.0 the callback works, but in the current 0.6.0 branch it does not. The attachment breaks at 8cbc3f5, during which some of the compile flags in extconf.rb got changed.

I found I could make the callback work again by messing with the compile flags in a local copy. I did the following:

• checkout 610fedd
• rake test (all passes)
• ran the test script (fails)
• change the extconf.rb as below
• rake test (all passes)
• ran the test script (passes)

All of this was done on OS X 10.5.8.

Test Script

module Foo
  extend FFI::Library
  ffi_lib 'path/to/my.bundle'

  callback :my_callback, [], :int
  attach_function :InitializeEnvironment, [], :void 
  attach_function :DefineFunction, [ :string, :char, :my_callback, :string ], :int
end

cb = Proc.new { 0 }
Foo.InitializeEnvironment
res = Foo.DefineFunction("ruby-callback", ?d, cb, "RubyCallback")
puts(res == 1 ? 'pass' : 'fail')

The updated extconf.rb

...
$defs << "-DHAVE_EXTCONF_H" if $defs.empty? # needed so create_header works
$defs << "-DHAVE_RAW_API"        # This is what I added
create_header
...

line 22 defines ARCH but the exception case on line 30 refers to "ARCH_" with an underscore

Hello,

I'm using ffi (0.5.1) with ruby 1.8.6 (2009-08-04 patchlevel 383) [i686-linux].

I have the following example.rb file:

require 'ffi'

module Example
  extend FFI::Library
  ffi_lib './example.so'

  attach_variable :cvar, :char
end

When I try to assign to that attached variable:

 Example.cvar   =  "S"

I get this error:

% ruby -rubygems runme.rb
(eval):6:in `[]=': can't convert String into Integer (TypeError)
  from (eval):6:in `cvar='
  from runme.rb:15

Shouldn't FFI do "S"[0,1].ord to get the integer value?

Also, why the (eval) in the error backtrace? Please see this patch.

Thanks for your consideration.

It's a minor issue..

http://github.com/ffi/ffi/blob/master/spec/ffi/pointer_spec.rb#L186

It says that it raises ArgumentError, but actually it raises TypeError...

Rubyspec catches ArgumentError, and it obviously fails

http://github.com/rubyspec/rubyspec/blob/master/optional/ffi/pointer_spec.rb#L144

Which one is correct?

When defining a layout for a C Struct that contains an inline Array of other Structs, FFI 0.5.3 will raise this exception:

/usr/lib64/ruby/gems/1.8/gems/ffi-0.5.3/lib/ffi/struct.rb:175:in `initialize': wrong argument type Class (expected Data) (TypeError)
    from /usr/lib64/ruby/gems/1.8/gems/ffi-0.5.3/lib/ffi/struct.rb:175:in `add_array'
    from /usr/lib64/ruby/gems/1.8/gems/ffi-0.5.3/lib/ffi/struct.rb:175:in `array_layout'
    from /usr/lib64/ruby/gems/1.8/gems/ffi-0.5.3/lib/ffi/struct.rb:106:in `layout'
    from /underground/code/sophsec/ffi-udis86/lib/udis86/ud.rb:14

I found the line of code which was triggering the exception was:

  ud.rb:32:             :operand, [Operand, 3],

UD: http://github.com/sophsec/ffi-udis86/blob/master/lib/udis86/ud.rb
Operand: http://github.com/sophsec/ffi-udis86/blob/master/lib/udis86/operand.rb

Currently there is a mswin32 binary gem uploaded on rubyforge. Please, provide a mingw32 precompiled gem too.