[jffi-next] annotated tag upstream/1.0.2 created (now 83ec96f)
Tim Potter
tpot-guest at moszumanska.debian.org
Wed Mar 4 04:51:18 UTC 2015
This is an automated email from the git hooks/post-receive script.
tpot-guest pushed a change to annotated tag upstream/1.0.2
in repository jffi-next.
at 83ec96f (tag)
tagging d99330f5388f1097ef0b5c0fbf6559b2cc7f058e (commit)
tagged by Torsten Werner
on Sat Sep 17 11:20:20 2011 +0200
- Log -----------------------------------------------------------------
Upstream version 1.0.2
Charles Oliver Nutter (8):
Update Darwin native bits.
Updated Darwin.jar for i386 and x86_64.
Update Darwin libs.
Update latest from remote.
Updated build of Darwin native libs.
Updated build of Darwin.jar on x86_64.
Update Darwin.jar.
Flip Darwin native bits.
Philip Jenvey (4):
rebuild i386-FreeBSD stub, built on FreeBSD 6
x86_64-FreeBSD stub, built on FreeBSD 7
sparc-SunOS stub built on Solaris 10
sparcv9-SunOS stub built on Solaris 10
Torsten Werner (2):
remove embedded code
Imported Upstream version 1.0.2
Vladimir Sizikov (5):
Tweaks to make sure JFFI is buildable on Windows
Eclipse settings
JRUBY-4388: [FFI] GetLastError always returns 0 on Windows
Provide i386-Windows binaries
Provide i386-Windows and i86_64-Windows binaries for jffi 1.0.
Wayne Meissner (480):
Initial checkin
Add .hgignore
Sync nb metadata
More .hgignore
Added libffi-3.0.7
Add some useful classes from old jffi
Hook in native build
More native build tweaks
Add a thin wrapper around raw dlopen/dlclose/dlsym
Add i386-linux build to archive
Implement native hooks to dlopen/dlsym/dlclose
Add unit test for dlopen
Add memory alocation/freeing
Remove default comment headers
Add Exceptions
Add CallContext allocation/freeing
Add primitive invokers
Store the function address along with the call context, for one less argument passed on the stack.
Move the fast-int invoker implementations into IntParameterInvoke.c
Avoid calling through ffi on i386 for int parameters - speeds up invocation a good chunk
Add 2 and 3 int argument methods
Rename all call* functions to invoke*
Add calling convention enum
Add Foreign#invokeInt() and InvocationBuffer to support general arguments
Add Int64, Float, Double invoke methods
Add native memory I/O operations
Fix MemoryIO#memchr name
Add MemoryIO#indexOf()
Add Foreign#invokeAddress()
Define all the put methods in InvocationBuffer
Don't copy the arg array if no args supplied
Add InvocationBuffer#putAddress
Fix float, double and address parameters
Implement findSymbol
Free the native function handle when being garbage collected
Move higher level invoke() methods into Invoker
Start on Closure support
clean out unused methods
Store the 64bit address directly as a long
Cleanups
undo some of the cleanups to the fast-int invokers as they slowed things down
Use Invoker.getInstance() instead of Foreign.getForeign()
Rename NativeType to Type
Implement allocateMemory and freeMemory in MemoryIO
Make Foreign#allocateMemory and Foreign#freeMemory package-private
Use MemoryIO#allocateMemory and MemoryIO#freeMemory
clean up fast int invokers
Use the long union field for big-endian for int return values
Use a constant for the param size
Store the size of the raw parameter buffer required
Use raw parameter packing on i386
Calculate the size of raw parameter buffer required, and use it
Fix for Function::rawParameterSize
Add parameter encoders for big-endian 32bit and 64bit
Move all the MemoryIO native functions into Foreign
Add MemoryIO#getStringLength()
Rename HeapInvocationBuffer::LE32rawEncoder to I386RawEncoder
Add a native function to test of the native side is using raw parameter packing, to ensure both java and native side are in agreement
Rename MemoryIO#getMemoryIO to MemoryIO#getInstance()
Rename Foreign#getForeign() to Foreign#getInstance()
Make Foreign package-private now that everything is exposed via better interfaces
Add a sun.misc.Unsafe based MemoryIO implementation
Support passing primitive arrays down to native code
Add putArray() to the InvocationBuffer interface
Start on the native side of callback support
Basic closures working
Fixup Invoker#invokeAddress() to pass the object buffer correctly
Mask off addresses to avoid sign extension from int -> long conversion
Call the Array release function so data gets copied back to java memory
Remove obsolete CallContext
Add errno support
Add a Function constructor that takes the calling convention
Move strlen, allocate and free into MemoryIO.c
Remove unused test
Move all the custom build targets into custom-build.xml
Use an ugly hack via ant to build and assemble every thing
Fixup typo
Rename to jffi
Update i386-linux build
Fix non-linux builds
Add darwin build
Add dlerror()
Replace Address.MASK with Platform.getPlatform().addressMask()
Remove unused code
Optimise functions that take 1 or 2 array arguments by passing the object parameters down as JNI parameters, not enncoded in an array
Update darwin build
Remove debugging println
Bump maven version to 0.1
Added tag 0.1 for changeset 210fcdcc465b
Update i386-linux build
Bump version to 0.2
Add \n at the end of files
Revamp darwin build a bit, so x86_64 and ppc can be dropped in and out by modifying ARCHES
Add new builds for i386-linux and i386-darwin
Bump version to 0.1.1
Added tag 0.1.1 for changeset cf6f1311152e
For darwin, only try to compile/link the required arches
Fix darwin build
Try to call a method on Foreign in Platform#isSupported(), to make sure the stub library actually loads
Figure out which argument encoder to use based on 32/64bit and ByteOrder instead of checking for specific arches
Better detection of amd64/x86_64 arch.
Move USE_RAW and BYPASS_FFI setting into one place
Rework Invoker and MemoryIO to only mask addresses on 32bit machines.
Fixup libc.so/dylib
Port to SunOS (solaris)
Re-arrange SOFLAGS a bit
Add support for all the primitive array types.
Add ParameterFlags#isIn and ParameterFlags#isOut
Add support for direct Buffer arguments
Cleanup big-endian parameter unpacking
Implement fast-long invoker paths
Add fast-long invoke methods to Invoker
Call zero-arg fast-long functions using libffi - i.e. don't bypass
Patch in exec-protection support for all operating systems
Fix ARGPR() macro for amd64
Rename IntParameterInvoke.c to FastIntInvoke.c
Fix typo in ARGPTR
Silence make commands
Re-arrange fast-int invoker code
Add i386-solaris build
Add caching to Library, and rewrite the API a bit to get rid of Address class usage
Enable x86_64 builds for MacOS
Update solaris x86 and amd64 binaries
Update MacOS build for i386 and x86_64
deprecate Address and anywhere it is used in Function
ClosureBuffer isn't used anywhere
Rename putInt{8,16,32,64} to putByte, putShort, putInt, putLong
Fairly large cleanup of Platform
Deprecate more Int{8,16,32,64} named methods and replace them with Byte, Short, Int, long
Latest build for i386-linux
Bump version to 0.2
Added tag 0.2 for changeset f3cd5b6076f1
Add some simple number parameter tests
Cleanup macro argument handling a bit.
Sync netbeans metadata
Rename ParameterFlags to ArrayFlags, and add support for pinned arrays
Sync netbeans meta-data
Fix invokeIIrI() parameter order error
Add windows OS detection
Only include pthread.h for non-win32
Tweak makefile for win32
Sync netbeans metadata
merge
Update builds for amd64 and i386 Solaris
Need stdlib for malloc
Update i386-linux build
Bump version to 0.3
Added tag 0.3 for changeset 10a5b3528278
Set MODEL=64 for x86_64
Remove deprecated methods
Add LGPLv3
Rename custom build to avoid name conflicts
Implement access to ffi_type structures, and ffi struct definitions
Convert Type from an enum to a class, so we can extend it
Pass down the ffi_type handles to Foreign#newFunction instead of the ffi enum type
Keep strong references to the return type and parameter types so they do not get garbage collected until the function does.
raw struct return values are working
Add a Struct class to wrap struct type allocation/cleanup
Passing struct by value works for raw argument packing
First attempt at struct arguments
Add Struct parameter support to HeapInvocationBuffer, and fix up struct parameters for non-raw invocation
Remove deprecated Closure methods, and javadoc most of it
Remove a printf and trim the auto-generated comments
Remove debug printf
Convert Foreign#newClosure() to pass the return and parameter types as pointers to the ffi_type structures
Implement Closure.Buffer#getStruct to get the address of struct parameters passed in to closures
Keep strong references to the return and parameter types so they don't get GCed until the closure handle is
Add tests for callbacks with struct-by-value params
Implement by-value struct return values from closures
Add Invoker#invokeBuffer()
Add an offset parameter to Foreign#invokeArrayWithReturnBuffer so a subsection of an existing buffer can be used as the return value
Split up the output jars. jffi.jar now just contains java classes, jffi-complete.jar contains the native code as well
Use UnitHelper.findSymbol() instead of manually loading the library and looking for a symbol
Mask addresses to 32bit if on a 32bit machine
Need <alloca.h> for solaris
Warn about undefined macro usage
Move endian-ness compatibility into endian.h
Merge commit
Add newline at end of file
Merge solaris fixes
Add some more javadoc
Remove deprecated methods
Add Type.nativeHandles() convenience method to convert from a Type[] array to an array of native ffi_type pointers
Address class is no longer used anywhere, so remove it
Move some code around internally in ClosureManager
Remove deprecated methods from invocationBuffer
More javadoc
Sync netbeans 6.7-m3 metadata
Simplify by using Type.nativeHandles(), and add javadoc
Add C union support as a type of struct with offset=0 for all fields
Rework Type a bit to better match up native ffi_type_foo, and also add aliases for char/short/int/long types
InvocationBuffer#putStruct can deduce the size to copy from the parameter type, so no need to pass it in
Javadoc Function
Simplify fast-int invokers slightly
Use j2p() to cast from the jlong to a pointer
Get rid of the unused 32bit version of the context addrss
Add javadoc to LastError
Add javadoc to ArrayFlags
Add javadoc to ObjectBuffer
Add javadoc to Invoker
Add some leading javadoc
Do not set the -m64 flag on darwin
Use Struct#size() to get the size of buffer required
Use j2p() to convert jlong to C pointer
Add some javadoc to Foreign
Move RTLD_* into Foreign and use them in native code
javadoc fix
Rename some of the invokeArray*() functions
alloca + inline functions == bad karma, so un-inline invokeArray
Use varargs for the Type field/parameter array for both Struct and Function
Clean up and use higher level interfaces
Rename Invoker#invokeBuffer to Invoker#invokeStruct
Add Foreign#invokeArrayWithObjectsReturnStruct
Add builds for Solaris i386 and x86_64
Build for i386-Linux
Fix Foreign#newStruct memory allocation
Update Solaris binaries to fix newStruct allocation bug
Ensure calloc calls are non-zero for functions/callbacks with no arguments
Add RTLD_MEMBER for aix
Make Type#size() and Type#alignment() public
On 32 bit machines, when the return type from a function indicates a size <= sizeof(ffi_sarg), return the return value from the first 32 bits of the retval buffer, not 64 bits.
Add sarg and arg members to union
Raise a compilation error if the endian-ness of the current platform cannot be determined
Add more closure tests
For byte, short and int return values, always write a full native 32/64 bit word
Patches for AIX support from Jean-Dominique Morani <jdmorani at gmail.com>
Add aix to the jffi-complete build
Remove dummy libs from empty ppc-AIX.jar
AIX gcc does not support -mimpure-text, so remove
Add sparc/sparcv9 jars to jffi-complete.jar build
Change AIX jni library extension to be '.a' instead of '.so'
Throw ppc-Darwin support into the ring as well, now that ppc-AIX works
Add -lpthread for aix
Fix up SOFLAGS and CFLAGS, and remove the old libtest building rules
No -mimpure-text on AIX
Sync netbeans metadata for 6.7beta
Merge commit
Turn off C warnings for MacOS
Update SunOS binaries for x86 and amd64
Update build for i386-linux
Added tag 0.4 for changeset 76cb761c198b
Bump maven version to 0.4
Sync up with ruby-ffi libtest/GNUmakefile
Sync with ruby-ffi libtest
Expose the address, return type, parameter types of the function
More syncing with ruby-ffi libtest
Add EnumTest.c
Add sparc and sparcv9 cpu detection
Just use 'gcc' on solaris instead of hardcoding the path, and move the -std=c99 flag to CFLAGS so it doesn't get propagated to the libffi build
Better use of findstring for -m$(MODEL) flag
Better use of findstring for -m$(MODEL) flag setting
Remove hardcoded path to gcc, and move -std=c99 to CFLAGS
Remove SOFLAGS from LDFLAGS since it is already listed on the link line
Remove unused LIBEXT var
Add ppc-AIX and sparc-SunOS binaries.
Update to correctly built amd64-SunOS stub lib
Added tag 0.4.1 for changeset f5213e028343
Implement experiemental alternative interface to FFI; this one takes an address of a native result buffer, and an array of addresses of parameters.
Use $(realpath) to ensure the value returned from /usr/bin/which ccache is a valid path
Add newDirectByteBuffer() and getDirectBufferAddress()
Implement MemoryIO#getZeroTerminatedByteArray to read in an ascii or utf8 stringin one hit, instead of searching for the terminator, then copying in the appropriate number of bytes.
newDirectBuffer takes an int as the length param, not a long
Add experimental interface to call JNI functions
Expose some JNI methods up to java land
Fix check for darwin when not adding -Wundef and friends to warning flags
Use the varargs version of the Function constructor
Convert some inline functions into macros
Add branch prediction hints in critical areas
I hate merge commits
Remove unused functions
Update i386-Linux build
Fix cut'n'paste typo where OUT array params were being treated as pinned arrays, and disable the pinned array path.
Add Foreign#getVersion to retrieve the version compiled in to the native stub
Normalize the cpu name when building the native jars, and use that normalized name to load the stub libary.
Update solaris builds for i386 and x86_64
Update build for i386-linux
sync netbeans metadata
Fix for i386-SunOS and x86_64-SunOS jar file names
Implement MemoryIO#putZeroTerminatedByteArray() to write a ascii/utf-8 string and NUL terminator in one operation
For the length limited version of MemoryIO#getZeroTerminatedByteArray(), use an int as the length param, since a long just does not make sense
Add backwards compatible native function for getZeroTerminatedByteArray(long, long)
Add one and two object optimizations for int64 returns
Re-arrange the invoke() macros a bit
Sync netbeans metadata
Add version of fast-int invokers that do not save errno
Make errno saving optional
Add a couple more branch prediction hints
Move the errno saving into a non-inline function, as gcc was un-inlining it anyway in certain functions, and this way there is only one copy of it
Add Foreign#getFunctionAddress() to return the C function contained in a function context instance
Move deprecated native functions to deprecated.c
Add Function constructor that takes a boolean flag that indicates whether calls should save errno or not
Bump version to 0.5, remove native stub lib munging
Add newline at EOF
Added tag 0.5 for changeset b468818704f7
Update i386 and x86_64 builds for solaris for 0.5
Update i386-linux build for 0.5
Zero out the old binaries
Update x86_64-linux build for 0.5
Add back the binary lib munging
sync netbeans metadata
Update libffi to 3.0.8
Use malloc instead of alloca for the struct return buffer
Small cleanup to alignment
Add ArrayFlags.CLEAR to clear the temporary native memory bounce buffer for OUT parameters, so junk is not copied back to java if the function does not fill out the entire buffer
First raft of win32 fixes
Merge commit
Add private ArrayFlags constructor
Re-enable ccache for solaris
Apply +5 hammer of javadoc to MemoryIO
Javadoc the Foreign equivalents of MemoryIO methods
Javadoc the Foreign equivalents of MemoryIO methods
Add javadoc for allocateMemory and freeMemory
Add LastError#setError()
Move the stub-loading configuration from Platform to Init
Small tweaks to where offset calculations are done
Setup for cross compiling from unix to mingw32
Add -static-libgcc to win32 LDFLAGS
Handle NULL paths in dlopen
Rework struct returns a bit. No longer attempt to handle them using ffi_raw_call, instead just convert from raw parameters to ptrarray and use ffi_call
More tweaks to mingw32 compilation
Strip debug symbols from cross-mingw32
Merge commit
Remove use of ffi_raw_call for struct returns in the invokeArrayWithObjects path as well
Use a loop to set ffiArgs[i] when calling with a struct return value, instead of ffi_raw_to_ptrarray(), since the latter does not handle i386 structs correctly.
Add some fairly thin wrappers around mmap/munmap and VirtualAlloc/VirtualFree
Make cross-mingw32 depend on init so vars get setup correctly
Instead of trying to abstract mmap(2) on unix vs VirtualAlloc() on win32, just export them both in Foreign, so a better API can be done later in, all in java
Javadoc
Make all fields private
The native interface version is now defined in the ant build file, and propagated to code
When loading the native stub library, look for a specific version
Cleanup some warnings
tweak run.jvmargs to pass -d64 or -d32 to junit
Dummy out all jars
OpenBSD fixes
FreeBSD fixes
Strip out debugging symbols from the shared lib
Lock around calls to newClosure and freeClosure
Correct name of OutOfMemory exception
Print the stack trace if freeClosure() ever fails in the finalizer
Fix memory leaks in freeClosure() and clean up a bit of code
Add test for Foreign#freeClosure()
Allow explicit control over closure freeing
Use '-S' instead of '--strip-debug'
Rebuild binaries for Darwin
Rebuild i386-Linux binary
Sync netbeans metadata
Rebuild binaries for i386 and x86_64 solaris
Rebuild for i386-FreeBSD
Merge
Fix cross build when build/classes is missing
Rebuild for i386-Windows
Added tag 0.6 for changeset a39aa492a425
Build Version.java as 1.5
Fix CCACHE
Build FastIntInvoke.o with frame pointer saving
Add float returning variants for fast-int invoke
Use explicit -march and -mtune flags for i386 arches
Tweak the bypass code paths to produce slightly better i386 asm
Type that describes an array
Do not specify -march on darwin
bump version to 0.7
Make Array, Struct, Union all inherit from Aggregate, to centralize the cleanup
Dump any exceptions during finalization to stderr
Build fixes for AIX
Merge commit
Dump any exceptions during finalization to stderr
Add explicit free() method
Start splitting out the basic call info from Function into a separate class
Add memmove and memcpy to Foreign, and adapt MemoryIO#copyMemory() to use memcpy if the regions overlap
Default to LAZY | LOCAL flags
Rename free to dispose to be more consistent
Fold replicated code into a macro
Enable ffi bypass for int args on x86_64
Remove define BYPASS_FFI from jffi.h
Add constructor which takes a Function for binary backwards compat
use getInt() for byte and short returns
read a native long for all 8, 16, 32bit return values
Move ffi_callX into a common place
add ffi_call{4, 5, 6}
Add ClosurePool.[ch]
Put preliminary bits in place for closure pooling
Use a ClosurePool for all closure allocations
Turn the CallContext cache into a SoftReference cache, and tweak the Signature hashCode and equals methods for performance
Build script now produces jffi-${CPU}-${OS}.jar instead of just ${CPU}-${OS}.jar
Split out the CallContextCache into a separate utility class
Keep a strong reference to the ClosurePool in the Handle to avoid it being garbage collected until all closures allocated from the pool have also been collected.
Add S390-Linux support
More s390x fixes
Removed tag 0.6
Remove duplicate 0.6 tag entries
Ignore libffi.info
Drop ppc and 10.4 support, since neither want to compile on MacOS 10.6
Add invokeNNNrN() style entry points which handle any primitive number, including floats and longs - at least on i386 and x86_64
Add newlines to the end of files that need it
remove obsolete binaries from archive
Implement 6 arg versions of fast-long and fast-number invokers, and use ffi-bypass for fast-long
Fix windows build
Remove superfluous redefinition of caddr_t
Import AsmJit from svn
Hook in AsmJit
Protect some headers against C++
Move all the page allocation/freeing/setting functions into memory.c
disable maven tests
Add windows to jffi-complete.jar
Comment out native library build - just bundle up the java class files
Make all Fast*Invoke.c files be compiled -fomit-frame-pointer
Extend FastIntInvoke to handle up to six integer arguments
Turn ClosurePool into a slab allocator
Add a mini-stress test for closure allocation
Set slab->pool in new_slab
Add some more closure pool stress tests
Rename ClosurePool::active to partial
Start splitting up ClosureManager
Re-implement the closure pooling in java, with simple magazine allocation in native code
Sync netbeans metadata
Move ClosurePool into its own java file
Remove ClosureMagazine and ClosureProxy - they are both subsumed in CLosurePool now
Remove obsolete closure pool related code
Remove closure pool methods
Handle struct-by-value returns with object parameters
Fix alternative version of invokeStruct() for object parameters
Build Version.java from pom.xml
Rename memory.[ch] to MemoryUtil.[ch]
BSD fixes
Add newlines at end of file
Update to libffi-3.0.9rc3
Add jffi-prefix to win32 cross compiled jar
Propagate any loading/version mismatch errors up in Foreign.getInstance()
Sync netbeans metadata
Add Type.nativeHandles(List<Type>)
Add licence header to files which did not have them
Re-enable FFI bypass on x86_64
Add PageManager, a thin wrapper around mmap/munmap/mprotect, and the windows equivalents
Add wrappers around jni RegisterNatives and UnregisterNatives
Remove AsmJit
more AsmJit removal
Revert to C compiler, not c++
Simplify _WIN32/_WIN64 abi code
Re-enable pinned array parameters
Rework array handling to make use of alloca() instead of a fixed-size stack buffer
Add explicit newArray and freeAggregate JNI calls
Make builtin type and alias type construction a bit more robust
Fix jffi_arraySize() and better error reporting
Fix yet another bug I introduced when cleaning up array handling
Export the address of the jffi_save_errno function so it can be used from jaffl
Make all the jni objects depend on the libffi build
Update to libffi-3.0.9
merge commit
Re-arrange call to alloca() to explicitly save the result in a local var
If jffi.boot.library.path is not set, try to get the property from /com/kenai/jffi/boot.properties
Throw up the dlerror() message as an exception from native code, so it does not get overwritten
Use alloca(3) for all stack allocations
Convert .hgignore to .gitignore
Pass -1 as the fd arg to mmap, since some unixen might require it
Add a lazy cache for builtin type info, so loading the Type class does not trigger the jni stub to load (or fail to load if it does not exist).
Add init() method to Foreign to initialize any class/method/field ids
Enable building against the system libffi via USE_SYSTEM_LIBFFI=1 or ant -Duse.system.libffi=1
Can now set LIBFFI_CFLAGS and LIBFFI_LIBS when calling make, and it will use those instead of pkg-config
Rebuild i386-linux binaries on ubuntu 6.06
Rebuild x86_64-Linux binaries
Rebuild darwin binaries for 0.7
Add LICENSE file with LGPL pre-amble
Add *~ to .gitignore
Bump pom version
Bump version to 1.0.0
Rebuild 1.0 for x86_64 on ubuntu 6.06
Rebuild 1.0 stub for i386 on ubuntu 6.06
empty out old binaries
Rebuild 1.0 stub for Darwin
Add possible fix for building on powerpc64 systems
Add binary stub for ppc-Linux. Built by Antoine Toulme <antoine at lunar-ocean.com>
More powerpc64 build fixes
Always use longSize == 32 for windows
Bump pom version to 1.0.1
Make everything build for 10.4 by using gcc-4.0 on darwin. PPC is still broken.
Simplify and cleanup cpu MODEL calculation
Cleanup findbugs warnings
Do not mask the upper 32bits of a long when using it to store an address
Bump pom version to 1.0.2
wayne (1):
Add newline at end of file
-----------------------------------------------------------------------
No new revisions were added by this update.
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/pkg-java/jffi-next.git
More information about the pkg-java-commits
mailing list