This file details the changelog of Capstone.

---------------------------------
Version 3.0.5-rc3: July 31st, 2017


[ Core ]

- Fix compilation for MacOS kernel extension
- cstool to support armbe and arm64be modes
- Add nmake.bat for Windows build
- Fix an integer overflow for Windows kernel driver
- Support to embedded Capstone into MacOS kernel
- cstool: fix mips64 mode
- Fix a compiling error in MS Visual Studio 2015
- Install pkgconfig file with CMake build
- Fix SOVERSION property of CMake build
- Properly handle switching to Endian mode at run-time for Arm, Arm64, Mips & Sparc
- Fix MingW build
- Better handle CMake installation for Linux 64bit


[ X86 ]

- Support BND prefix of Intel MPX extension
- Correct operand size for CALL/JMP in 64bit mode with prefix 0x66
- LOCK NOP is a valid instruction
- Fix ATT syntax for instruction with zero offset segment register
- LES/LDS are invalid in 64bit mode
- Fix number of operands for some MOV instructions


[ ARM ]

- Fix POP reg to update SP register
- Update flags for UADD8 instruction


[ ARM64 ]

- Better performance with new lookup table
- Handle system registers added in ARMv8.1/2


[ Java binding ]

- Better handle input with invalid code


[ Visual Basic binding ]

- New binding

---------------------------------
Version 3.0.5-rc2: March 2nd, 2017


[ Core ]

- Fix build for Visual Studio 2012
- Fix X86_REL_ADDR macro
- Add CS_VERSION_MAJOR, CS_VERSION_MINOR, CS_VERSION_EXTRA
- Better support for embedding Capstone into Windows kernel drivers
- Support to embedded Capstone into MacOS kernel
- Support MacOS 10.11 and up
- Better support for Cygwin
- Support build packages for FreeBSD & DragonflyBSD
- Add a command-line tool "cstool"
- Properly handle switching to Endian mode at run-time for Arm, Arm64, Mips & Sparc


[ X86 ]

- Some random 16-bit code can be handled wrongly.
- Remove abundant operand type X86_OP_FP
- Fix instructions MOVQ, LOOP, LOOPE, LOOPNE, CALL/JMP rel16, REPNE LODSD, MOV *AX, MOFFS, FAR JMP/CALL
- Add X86_REG_EFLAGS for STC and STD
- Fix instruction attributes for SYSEXIT, MOVW, ROL, LGS, SLDT
- Rename registers ST0-ST7 to be consistent with asm output


[ ARM ]

- Properly handle IT instruction
- Fix LDRSB
- Fix writeback for LDR
- Fix Thumb BigEndian setup


[ ARM64 ]

- Fix arith extender
- Fix writeback for LDR
- Rename enum arm64_mrs_reg to arm64_sysreg


[ PowerPC ]

- Print 0 offset for memory operand


[ Sparc ]

- Fix POPC instruction


[ Python binding ]

- Better PyPy support
- Add __version__
- Better support for Python 3
- Fix CS_SKIPDATA_CALLBACK prototype
- Cast skipdata function inside binding to simplify the API


[ Java binding ]

- Better handle input with invalid code


[ PowerShell ]

- New binding

---------------------------------
Version 3.0.4: July 15th, 2015


[ Library ]

- Improve cross-compile for Android using Android NDK.
- Support cross-compile for AArch64 Android (with Linux GCC).
- Removed osxkernel_inttypes.h that is incompatible with BSD license.
- Make it possible to compile with CC having a space inside (like "ccache gcc").


[ X86 ]

- Fix a null pointer dereference bug on handling code with special prefixes.
- Properly handle AL/AX/EAX operand for OUT instruction in AT&T syntax.
- Print immediate operand in positive form in some algorithm instructions.
- Properly decode some SSE instructions.


[ PowerPC ]

- Fixed a memory corruption bug.
- Fixed a memory corruption bug for the engine built in DIET mode.


[ Mips ]

- Fixed instruction ID of SUBU instruction.
- Fixed a memory corruption bug.


[ Arm ]

- Fixed a memory corruption bug on IT instruction.


[ XCore ]

- Fixed a memory corruption bug when instruction has a memory operand.


[ Python ]

- Support Virtualenv.
- setup.py supports option --user if not in a virtualenv to allow for local usage.
- Properly handle the destruction of Cs object in the case the shared library
  was already unloaded.

---------------------------------
Version 3.0.3: May 08th, 2015


[ Library ]

- Support to embed into Mac OS X kernel extensions.
- Now it is possible to compile Capstone with older C compilers, such as
  GCC 4.8 on Ubuntu 12.04.
- Add "test_iter" to MSVC project.


[ X86 ]

- All shifted instructions SHL, SHR, SAL, SAR, RCL, RCR, ROL & ROR now support
  $1 as first operand in *AT&T* syntax (so we have "rcll $1, %edx" instead of
  "rcll %edx").
- CMPXCHG16B is a valid instruction with LOCK prefix.
- Fixed a segfault on the input of 0xF3.


[ Arm ]

- BLX instruction modifies PC & LR registers.


[ Sparc ]

- Improved displacement decoding for sparc banching instructions.


[ Python binding ]

- Fix for Cython so it can properly initialize.
- X86Op.avx_zero_mask now has c_bool type, but not c_uint8 type.
- Properly support compile with Cygwin & install binding (setup.py).

---------------------------------
Version 3.0.2: March 11th, 2015


[ Library ]

- On *nix, only export symbols that are part of the API (instead of all
  the internal symbols).


[ X86 ]

- Do not consider 0xF2 as REPNE prefix if it is a part of instruction encoding.
- Fix implicit registers read/written & instruction groups of some instructions.
- More flexible on the order of prefixes, so better handle some tricky
  instructions.
- REPNE prefix can go with STOS & MOVS instructions.
- Fix a compilation bug for X86_REDUCE mode.
- Fix operand size of instructions with operand PTR []


[ Arm ]

- Fix a bug where arm_op_mem.disp is wrongly calculated (in DETAIL mode).
- Fix a bug on handling the If-Then block.


[ Mips ]

- Sanity check for the input size for MIPS64 mode.


[ MSVC ]

- Compile capstone.dll with static runtime MSVCR built in.


[ Python binding ]

- Fix a compiling issue of Cython binding with gcc 4.9.

---------------------------------
Version 3.0.1: February 03rd, 2015

[ X86 ]

- Properly handle LOCK, REP, REPE & REPNE prefixes.
- Handle undocumented immediates for SSE's (V)CMPPS/PD/SS/SD instructions.
- Print LJUMP/LCALL without * as prefix for Intel syntax.
- Handle REX prefix properly for segment/MMX related instructions (x86_64).
- Instruction with length > 15 is consider invalid.
- Handle some tricky encodings for instructions MOVSXD, FXCH, FCOM, FCOMP,
  FSTP, FSTPNCE, NOP.
- Handle some tricky code for some X86_64 instructions with REX prefix.
- Add missing operands in detail mode for PUSH , POP , IN/OUT reg, reg
- MOV32ms & MOV32sm should reference word rather than dword.


[ Arm64 ]

- BL & BLR instructions do not read SP register.
- Print absolute (rather than relative) address for instructions B, BL,
  CBNZ, ADR.


[ Arm ]

- Instructions ADC & SBC do not update flags.
- BL & BLX do not read SP, but PC register.
- Alias LDR instruction with operands [sp], 4 to POP.
- Print immediate operand of MVN instruction in positive hexadecimal form.


[ PowerPC ]

- Fix some compilation bugs when DIET mode is enable.
- Populate SLWI/SRWI instruction details with SH operand.


[ Python binding ]

- Fix a Cython bug when CsInsn.bytes returns a shorten array of bytes.
- Fixed a memory leak for Cython disasm functions when we immaturely quit
  the enumeration of disassembled instructions.
- Fix a NULL memory access issue when SKIPDATA & Detail modes are enable
  at the same time.
- Fix a memory leaking bug when when we stop enumeration over the disassembled
  instructions prematurely.
- Export generic operand types & groups (CS_OP_xxx & CS_GRP_xxx).

---------------------------------
Version 3.0: November 19th, 2014

[ API ]

- New API: cs_disasm_iter & cs_malloc. See docs/README for tutorials.
- Renamed cs_disasm_ex to cs_disasm (cs_disasm_ex is still supported, but
  marked obsolete to be removed in future)
- Support SKIPDATA mode, so Capstone can jump over unknown data and keep going
  from the next legitimate instruction. See docs/README for tutorials.
- More details provided in cs_detail struct for all architectures.
- API version was bumped to 3.0.


[ Bindings ]

- Python binding supports Python3 (besides Python2).
- Support Ocaml binding.
- Java: add close() method to be used to deinitialize a Capstone object when
  no longer use it.


[ Architectures ]

- New architectures: Sparc, SystemZ & XCore.
- Important bugfixes for Arm, Arm64, Mips, PowerPC & X86.
- Support more instructions for Arm, Arm64, Mips, PowerPC & X86.
- Always expose absolute addresses rather than relative addresses (Arm, Arm64,
  Mips, PPC, Sparc, X86).
- Use common instruction operand types REG, IMM, MEM & FP across all
  architectures (to enable cross-architecture analysis).
- Use common instruction group types across all architectures (to enable
  cross-architecture analysis).


[ X86 ]

- X86 engine is mature & handles all the malware tricks (that we are aware of).
- Added a lot of new instructions (such as AVX512, 3DNow, etc).
- Add prefix symbols X86_PREFIX_REP/REPNE/LOCK/CS/DS/SS/FS/GS/ES/OPSIZE/ADDRSIZE.
- Print immediate in positive form & hexadecimal for AND/OR/XOR instructions.
- More friendly disassembly for JMP16i (in the form segment:offset)


[ Mips ]

- Engine added supports for new hardware modes: Mips32R6 (CS_MODE_MIPS32R6) &
  MipsGP64 (CS_MODE_MIPSGP64).
- Removed the ABI-only mode CS_MODE_N64.
- New modes CS_MODE_MIPS32 & CS_MODE_MIPS64 (to use instead of CS_MODE_32 &
  CS_MODE_64).


[ ARM ]

- Support new mode CS_MODE_V8 for Armv8 A32 encodings.
- Print immediate in positive form & hexadecimal for AND/ORR/EOR/BIC instructions


[ ARM64 ]

- Print immediate in hexadecimal for AND/ORR/EOR/TST instructions.


[ PowerPC ]

- Do not print a dot in front of absolute address.


[ Other features ]

- Support for Microsoft Visual Studio (so enable Windows native compilation).
- Support CMake compilation.
- Cross-compile for Android.
- Build libraries/tests using XCode project
- Much faster, while consuming less memory for all architectures.

---------------------------------
Version 2.1.2: April 3rd, 2014

This is a stable release to fix some bugs deep in the core. There is no update
to any architectures or bindings, so bindings version 2.1 can be used with this
version 2.1.2 just fine.

[ Core changes]

- Support cross-compilation for all iDevices (iPhone/iPad/iPod).
- X86: do not print memory offset in negative form.
- Fix a bug in X86 when Capstone cannot handle short instruction.
- Print negative number above -9 without prefix 0x (arm64, mips, arm).
- Correct the SONAME setup for library versioning (Linux, *BSD, Solaris).
- Set library versioning for dylib of OSX.

---------------------------------
Version 2.1.1: March 13th, 2014

This is a stable release to fix some bugs deep in the core. There is no update
to any architectures or bindings, so bindings version 2.1 can be used with this
version 2.1.1 just fine.

[ Core changes]

- Fix a buffer overflow bug in Thumb mode (ARM). Some special input can
  trigger this flaw.
- Fix a crash issue when embedding Capstone into OSX kernel. This should
  also enable Capstone to be embedded into other systems with limited stack
  memory size such as Linux kernel or some firmwares.
- Use a proper SONAME for library versioning (Linux).

---------------------------------
Version 2.1: March 5th, 2014

[ API changes ]

- API version has been bumped to 2.1.
- Change prototype of cs_close() to be able to invalidate closed handle.
  See http://capstone-engine.org/version_2.1_API.html for more information.
- Extend cs_support() to handle more query types, not only about supported
  architectures. This change is backward compatible, however, so existent code
  do not need to be modified to support this.
- New query type CS_SUPPORT_DIET for cs_support() to ask about diet status of
  the engine.
- New error code CS_ERR_DIET to report errors about newly added diet mode.
- New error code CS_ERR_VERSION to report issue of incompatible versions between
  bindings & core engine.


[ Core changes ]

- On memory usage, Capstone uses about 40% less memory, while still faster
  than version 2.0.
- All architectures are much smaller: binaries size reduce at least 30%.
  Especially, X86-only binary reduces from 1.9MB to just 720KB.
- Support "diet" mode, in which engine size is further reduced (by around 40%)
  for embedding purpose. The price to pay is that we have to sacrifice some
  non-critical data fields. See http://capstone-engine.org/diet.html for more
  details.


[ Architectures ]

- Update all 5 architectures to fix bugs.
- PowerPC:
	- New instructions: FMR & MSYNC.
- Mips:
	- New instruction: DLSA
- X86:
	- Properly handle AVX-512 instructions.
	- New instructions: PSETPM, SALC, INT1, GETSEC.
	- Fix some memory leaking issues in case of prefixed instructions such
	  as LOCK, REP, REPNE.


[ Python binding ]

- Verify the core version at initialization time. Refuse to run if its version
  is different from the core's version.
- New API disasm_lite() added to Cs class. This light API only returns tuples of
  (address, size, mnemonic, op_str), rather than list of CsInsn objects. This
  improves performance by around 30% in some benchmarks.
- New API version_bind() returns binding's version, which might differ from
  the core's API version if the binding is out-of-date.
- New API debug() returns information on Cython support, diet status & archs
  compiled in.
- Fixed some memory leaking bugs for Cython binding.
- Fix a bug crashing Cython code when accessing @regs_read/regs_write/groups.
- Support diet mode.


[ Java binding ]

- Fix some memory leaking bugs.
- New API version() returns combined version.
- Support diet mode.
- Better support for detail option.


[ Miscellaneous ]

- make.sh now can uninstall the core engine. This is done with:

	$ sudo ./make.sh uninstall

----------------------------------
Version 2.0: January 22nd, 2014

Release 2.0 deprecates verison 1.0 and brings a lot of crucial changes.

[ API changes ]

- API version has been bumped to 2.0 (see cs_version() API)
- New API cs_strerror(errno) returns a string describing error code given
  in its only argument.
- cs_version() now returns combined version encoding both major & minor versions.
- New option CS_OPT_MODE allows to change engine’s mode at run-time with
  cs_option().
- New option CS_OPT_MEM allows to specify user-defined functions for dynamically
  memory management used internally by Capstone. This is useful to embed Capstone
  into special environments such as kernel or firware.
- New API cs_support() can be used to check if this lib supports a particular
  architecture (this is necessary since we now allow to choose which architectures
  to compile in).
- The detail option is OFF by default now. To get detail information, it should be
  explicitly turned ON. The details then can be accessed using cs_insn.detail
  pointer (to newly added structure cs_detail)


[ Core changes ]

- On memory usage, Capstone uses much less memory, but a lot faster now.
- User now can choose which architectures to be supported by modifying config.mk
  before compiling/installing.


[ Architectures ]

- Arm
     - Support Big-Endian mode (besides Little-Endian mode).
     - Support friendly register, so instead of output sub "r12,r11,0x14",
	 we have "sub ip,fp,0x14".
- Arm64: support Big-Endian mode (besides Little-Endian mode).
- PowerPC: newly added.
- Mips: support friendly register, so instead of output "srl $2,$1,0x1f",
     we have "srl $v0,$at,0x1f".
- X86: bug fixes.


[ Python binding ]

- Python binding is vastly improved in performance: around 3 ~ 4 times faster
  than in 1.0.
- Cython support has been added, which can further speed up over the default
  pure Python binding (up to 30% in some cases)
- Function cs_disasm_quick() & Cs.disasm() now use generator (rather than a list)
  to return succesfully disassembled instructions. This improves the performance
  and reduces memory usage.


[ Java binding ]

- Better performance & bug fixes.


[ Miscellaneous ]

- Fixed some installation issues with Gentoo Linux.
- Capstone now can easily compile/install on all *nix, including Linux, OSX,
  {Net, Free, Open}BSD & Solaris.

----------------------------------
[Version 1.0]: December 18th, 2013

- Initial public release.