File luajit.changes of Package luajit
-------------------------------------------------------------------
Tue Oct 7 09:13:59 UTC 2025 - John Paul Adrian Glaubitz <adrian.glaubitz@suse.com>
- Add loongarch64-support.patch
-------------------------------------------------------------------
Tue Oct 7 09:10:24 UTC 2025 - John Paul Adrian Glaubitz <adrian.glaubitz@suse.com>
- Add riscv-support.patch
-------------------------------------------------------------------
Tue Oct 7 08:46:01 UTC 2025 - John Paul Adrian Glaubitz <adrian.glaubitz@suse.com>
- Add patch for LoongArch support
-------------------------------------------------------------------
Tue Oct 7 08:44:52 UTC 2025 - John Paul Adrian Glaubitz <adrian.glaubitz@suse.com>
- Add patch for RISC-V support
-------------------------------------------------------------------
Tue Oct 7 08:38:47 UTC 2025 - John Paul Adrian Glaubitz <adrian.glaubitz@suse.com>
- Allow build everywhere
-------------------------------------------------------------------
Tue Oct 7 08:37:48 UTC 2025 - John Paul Adrian Glaubitz <adrian.glaubitz@suse.com>
- Test package
-------------------------------------------------------------------
Tue Oct 07 08:12:36 UTC 2025 - mcepl@cepl.eu
- Update to version v2.1-20250826+git.1756211046.dcc9c9ee:
* Windows: Add lua52compat option to msvcbuild.bat.
* x86/x64: Don't use undefined MUL/IMUL zero flag.
* Avoid out-of-range PC for stack overflow error from snapshot restore.
* FFI: Fix dangling CType references (again).
* Fix JIT slot overflow during up-recursion.
* bugfix: table.clone can't work after commit 538a82133ad.
* Prevent Clang UB 'optimization' which breaks integerness checks.
* ARM: Fix soft-float math.min()/math.max().
* ARM64: Fix pass-by-value struct calling conventions.
* Change handling of nil value markers in template tables.
* REVERT: Change handling of nil value markers in template tables.
* macOS: Fix Apple hardened runtime support and put behind build option.
* macOS: Fix support for Apple hardened runtime.
* Fix handling of nil value markers in template tables.
* Fix state restore when recording __concat metamethod.
* Windows: Allow mixed builds with msvcbuild.bat.
* macOS: Add suport for Apple hardened runtime.
* Windows: Clarify installation directory layout.
* Remove Cygwin from docs, since it's not a supported target.
* Improve CLI signal handling on POSIX.
* FFI: Add pre-declared int128_t, uint128_t, __int128 types.
* Use dylib extension for iOS installs, too.
* Change handling of nil value markers in template tables.
* Remove dependency on <limits.h>.
* Restore state when recording __concat metamethod throws OOM.
* MIPS64: Fix pcall() error case.
* Windows: Allow amalgamated static builds with msvcbuild.bat.
* Always close profiler output file.
* Fix override of INSTALL_LJLIBD in the presence of DESTDIR.
* Fix bit op coercion for shifts in DUALNUM builds.
* macOS: Remove obsolete -single_module flag.
* macOS: Workaround for buggy XCode 15.0 - 15.2 linker.
* macOS: Fix macOS 15 / Clang 16 build.
* Fix bit op coercion in DUALNUM builds.
* Drop unused function wrapper.
* ARM64: Use ldr literal to load FP constants.
* FFI: Add missing coercion when recording 64-bit bit.*().
* ARM64: Make tobit conversions match JIT backend behavior.
* FFI: Drop finalizer table rehash after GC cycle.
* Reflect override of INSTALL_LJLIBD in package.path.
* ARM64: Use movi to materialize FP constants.
* Add more FOLD rules for integer conversions.
* Call math.randomseed() without arguments to seed from system entropy.
* Restore state when recording __concat metamethod throws an error.
* Add build flag LUAJIT_DISABLE_TAILCALL to disable tailcall generation.
* Clarify that lj_buf_shrink() does not keep any buffer data.
* OSX: Fix installed luajit.pc.
* change: disable hash computation optimization because of the possibility of severe degradation.
* bugfix： Enabled ppc64le arch on travis and fixed one failing test case.
* Update Nintendo Switch build script.
* Typo.
* Limit number of string format elements to compile.
* FFI: Clarify scalar boxing behavior.
* Fix internal link in docs.
* OSX/iOS: Fix SDK incompatibility.
* Windows/MSVC: Cleanup msvcbuild.bat and always generate PDB.
* Fix compiler warning.
* FFI: Turn FFI finalizer table into a proper GC root.
* OSX/iOS: Always generate 64 bit non-FAT Mach-O object files.
* Show name of NYI bytecode in -jv and -jdump.
* Use generic trace error for OOM during trace stitching.
* Fix serialization format docs.
* feature: add s390x disassembler.
* Handle all types of errors during trace stitching.
* Fix recording of __concat metamethod.
* optimize: [ppc64le] Aligned code as per other archs for next_1 function and relevant code changes.
* Fix zero stripping in %g number formatting.
* DynASM/x86: Add endbr instruction.
* MIPS64 R2/R6: Fix FP to integer conversions.
* Add cross-32/64 bit and deterministic bytecode generation.
* DynASM/x86: Allow [&expr] operand.
* Fix anchoring for string buffer set() method (again).
* Fix anchoring for string buffer set() method.
* Optimize table.new() with constant args to (sinkable) IR_TNEW.
* Fix last commit.
* Improve last commit.
* x86/x64: Don't fuse loads across table.clear.
* ARM64: Fix disassembly of ldp/stp offsets.
* Check for upvalue state transition in IR_UREFO.
* Windows/x86: _BitScan*64 are only available on 64 bit archs.
* Add 'cc' file type for saving bytecode.
* FFI/Windows: Fix type declaration for int64_t and uint64_t.
* FFI: Fix dangling reference to CType in carith_checkarg().
* DynASM/ARM64: Support ldp/stp of q registers.
* ARM64: Use ADR and ADRP to form constants.
* ARM64: Fix disassembly of U12 loads.
* ARM64: Unify constant register handling in interpreter.
* ARM: Fix register hint for FFI calls with FP results.
* ARM64: Fix register hint for FFI calls with FP results.
* ARM64: Restore fp before sp in C stack unwinders.
* Add 'cc' file type for saving bytecode.
* FFI/Windows: Fix type declaration for int64_t and uint64_t.
* Update lj_parse.c
* Update lj_ccall.c
* ppc64le support
* DynASM/ARM64: Support ldp/stp of q registers.
* ARM64: Use ADR and ADRP to form constants.
* ARM64: Fix disassembly of U12 loads.
* ARM64: Unify constant register handling in interpreter.
* ARM: Fix register hint for FFI calls with FP results.
* ARM64: Fix register hint for FFI calls with FP results.
* ARM64: Restore fp before sp in C stack unwinders.
* tests: update test cases to reflect latest change from upstream.
* FFI: Fix ffi.abi("pauth").
* LJ_FR2: Fix stack checks in vararg calls.
* Handle OOM error on stack resize in coroutine.resume and lua_checkstack.
* Restore cur_L for specific Lua/C API use case.
* Consistently use 64 bit constants for 64 bit IR instructions.
* ARM64: Fix IR_HREF code generation for constant FP keys.
* ARM64: Fuse negative 32 bit constants into arithmetic ops again.
* Windows/ARM64: Add MSVC cross-build support for x64 to ARM64.
* IR_MIN/IR_MAX is non-commutative due to underlying FPU ops.
* ARM64: Set fixed interpreter registers before rethrow.
* Windows/ARM64: Update install docs.
* Windows: Call C++ destructors without compiling with /EHa.
* Windows: Pass scratch CONTEXT record to RtlUnwindEx.
* ARM64: External unwinder already restores non-volatile registers.
* Windows/ARM64: Fix exception unwinding (again).
* Windows/ARM64: Fix typo in exception unwinding.
* FFI: Fix 64 bit shift fold rules.
* Windows/ARM64: Support Windows calling conventions.
* Windows/ARM64: Fix exception unwinding.
* ARM64: Remove unneeded IRCALL_* defs for math intrinsics.
* Fix Cygwin build.
* bugfix: Update s390x support.
* Windows/ARM64: Add initial support.
* ARM64: Fuse rotates into logical operands.
* ARM64: Don't fuse sign extensions into logical operands.
* ARM64: Disassemble rotates on logical operands.
* ARM64: Fix LDP/STP fusion (again).
* ARM64: Ensure branch is in range before emitting TBZ/TBNZ.
* ARM64: Improve BC_JLOOP.
* ARM64: Improve integer IR_MUL code generation.
* ARM64: Simplify code generation for IR_STRTO.
* ARM64: Use RID_TMP instead of scratch register in more places.
* ARM64: Improve IR_OBAR code generation.
* ARM64: Improve IR_UREF code generation.
* ARM64: Improve IR_HREF code generation.
* ARM64: Reload BASE via GL instead of spilling it.
* ARM64: Consolidate 32/64-bit constant handling in assembler.
* ARM64: Tune emit_lsptr. Avoid wrong load for asm_prof.
* ARM64: Inline only use of emit_loada.
* ARM64: Improve K13 constant rematerialization.
* tests: only run with 1 core.
* tests: fix one test cases.
* Add randomized register allocation for fuzz testing.
* ARM64: Improve register allocation for integer IR_MUL/IR_MULOV.
* ARM64: Fix register allocation for IR_*LOAD.
* FFI/ARM64/OSX: Handle non-standard OSX C calling conventions.
* FFI: Unify stack setup for C calls in interpreter.
* ARM64: Prevent STP fusion for conditional code emitted by TBAR.
* ARM64: Fix LDP/STP fusing for unaligned accesses.
* Fix typo.
* Add .gitattributes to dynamically resolve .relver.
* Fix for last commit: also remove symlink on uninstall.
* Switch to rolling releases: mark v2.1 as production.
* Remove work-in-progress notice in string buffer docs.
* MIPS: Fix "bad FP FLOAD" assertion.
* Ensure forward progress on trace exit to BC_ITERN.
* ARM64: Add support for ARM64e pointer authentication codes (PAC).
* DynASM/ARM64: Add instructions for ARM64e PAC.
* MIPS32: Declare that the assembler part uses the FR=0 model.
* ARM64: Fix assembly of HREFK (again).
* Fix frame for more types of on-trace error messages.
* Add workaround for bytecode dump of builtins.
* DynASM: Fix regression due to warning fix.
* ARM64: Fix assembly of HREFK.
* Fix maxslots when recording BC_TSETM.
* ARM64: Fix LDP code generation.
* PPC/e500 with SPE enabled: use soft float instead of failing.
* MIPSr6: Add missing files to Makefile install target.
* DynASM: Fix warnings.
* Fix frame for on-trace out-of-memory error.
* Fix last commit.
* Print errors from __gc finalizers instead of rethrowing them.
* tests: fixed travis tests.
* Improve error reporting on stack overflow.
* Allow building sources with mixed LF/CRLF line-endings.
* Fix compiler warning.
* Don't fail for Clang builds, which pretend to be an ancient GCC.
* feature: allow emitting obj files on s390x.
* Fix compiler warning.
* Don't fail for Clang builds, which pretend to be an ancient GCC.
* Revert "change: the order of pairs was indeterminate because it depends on str->sid. (#119)" (#191)
* Disable FMA by default. Use -Ofma or jit.opt.start("+fma") to enable.
* ARM64: Fix code generation for IR_SLOAD with typecheck + conversion.
* PS4/PS5: Fix build scripts.
* Avoid assertion in case of stack overflow from stitched trace.
* NX: Always compile with -fPIC.
* ARM64: Fix IR_SLOAD assembly.
* Merge branch 'v2.1' into v2.1-agentzh (#184)
* Fix trace join to BC_JLOOP originating from BC_ITERN.
* Fix compiler warnings.
* Fix QNX build.
* bugfix： fix math.floor() and math.ceil() (#177)
* Add -F option to override filename in jit.bcsave (luajit -b).
* LJ_GC64: Fix lua_concat().
* Improve GC estimation for userdata with attached managed memory.
* Add missing GC steps to string buffer methods.
* x86/x64: Limit VLOAD fusion to simple cases.
* OSX/iOS/ARM64: Fix generation of Mach-O object files.
* Prevent trace start at BC_ITERL after compiled BC_ITERN.
* ARM64: Allow building with unwinding disabled.
* OSX/iOS/ARM64: Fix bytecode embedding in Mach-O object file.
* LJ_GC64: Fix IR_VARG offset for fixed number of results.
* Make embedded bytecode readable and forward-compatible.
* DynASM: Fix warnings.
* Add Nintendo Switch port.
* Avoid zero-sized arrays in jit_State.
* Don't use jit_State during build with JIT disabled.
* DynASM/ARM64: Fix LSL/BFI* encoding with variable registers.
* Fix ITERN loop detection when hook checks are enabled.
* Fix alias analysis for table length forwarding.
* Fix loop initialization in table.foreach().
* LJ_GC64: Fix HREFK optimization.
* Fix recording of __concat metamethod.
* Add PS5 port.
* x64: Fix RETHI/RETLO swap after call.
* Fix assertion for LJ_KEYINDEX.
* Add missing check for LJ_KEYINDEX in ITERN recording.
* DynASM/ARM64: Fix NOP instruction for aligment
* Fix soft-float IR_POW splitting.
* Fix compiler warning.
* Revert to trival pow() optimizations to prevent inaccuracies.
* Fix string buffer COW handling, part 2.
* *BSD: Fix getentropy() declaration.
* Fix string buffer COW handling.
* bugfix: do not eliminate HSTORE over nkeys and clone (#148)
* Fix pow() optimization inconsistencies.
* Fix FOLD rule for BUFHDR append with intervening buffer use.
* Fix compiled error handling for buffer methods.
* Prevent unroll across BC_ITERN.
* Prevent replay of buffer operation for PHI operands.
* FFI: Ensure library is loaded before de-serializing FFI types.
* Fix HREFK forwarding vs. table.clear().
* Fix FOLD rule for BUFHDR append.
* OSX: Disable unreliable assertion for external frame unwinding.
* bugfix: src/lj_err.c: rroneous assertion failure on x64 macOS.
* Windows: Fix binary output of jit.bcsave to stdout.
* bugfix: This PR fixes issue https://github.com/openresty/luajit2/issues/136 (#144)
* Update Android build docs.
* ARM64: Fix pcall() error case.
* tests: enables TravisCI for arm64.
* refactor: removed duplicated table entries. (#138)
* This PR fixes build issue on s390x introduced with commit 394fb6267acba72ee984edcd331ad1bbde72056a: (#135)
* OSX/ARM64: Fix external unwinding.
* travis-ci: upgrade dist of travis-ci to ubuntu bionic. (#133)
* bugfix: fixed merge error which was introduced by commit 63dee93f4e. (#132)
* Fix string buffer method recording.
* Fix interaction of profiler and ITERN recording.
* Fix compilation of multi-result call to next().
* ARM64: Fix IR_HREF code generation.
* MIPS64: Fix soft-float IR_TOSTR.
* Fix compiler warning.
* FFI: Ensure returned string is alive in ffi.typeinfo().
* OSX/ARM64: Disable unwind info.
* Fix ITERN blacklisting.
* Ensure ITERN forward progress on interpreter bailout.
* ARM64: Reorder interpreter stack frame and fix unwinding.
* Don't bail out to interpreter to JLOOP originating from ITERN.
* FFI: Don't load PC from non-function object in FFI continuation.
* tests: checked in iter.t for testing hash table iterations.
* OSX/ARM64: Disable external unwinding for now.
* Compile table traversals: next(), pairs(), BC_ISNEXT/BC_ITERN.
* Use IR_HIOP for generalized two-register returns.
* Refactor table traversal.
* ARM: Fix symbol display in trace disassembly.
* Refactor IR_TMPREF generation.
* Refactor IR_VLOAD to take an offset.
* MIPS: Fix trace linking.
* String buffers, part 4b: Improve de-serialization checks.
* String buffers, part 4a: Add metatable serialization dictionary.
* PS4: Fix compile.
* String buffers, part 3d: Compile string buffer methods and functions.
* Minor improvements of optimizations.
* Fix error message in lj_lib_checkintrange().
* String buffers, part 2f: Prevent self-put of buffer.
* String buffers, part 3c: Add IRBUFHDR_WRITE mode.
* Fix IRXLOAD_* mode bits description.
* String buffers, part 3b: Change IR_BUFHDR op2 mode bits to mode.
* String buffers, part 3a: Add IR_TMPREF for passing TValues to helpers.
* Add IRCONV_NONE for pass-through INT to I64/U64 type change.
* travis-ci: removed clang compiler mode to save travis credits. (#128)
* Change: Resolve luaL_newstate() return NULL in ppc64le issue (#123)
* Disable unreliable assertion for external frame unwinding.
* Flush and close output file after profiling run.
* Avoid conflict between 64 bit lightuserdata and ITERN key.
* Change: Resolve compilation error in ppc
* doc: fix spelling mistake in README (#127)
* bugfix: disabled the assertion since it might be a false alarm on fedora aarch64.
* feature: added the trace entry and normal exit events in the GC64 interpreter.
* Throw any errors before stack changes in trace stitching.
* String buffers, part 2e: add serialization string dictionary.
* DynASM/ARM64: Fix LSL/BFI* encoding with variable shifts.
* Fix MinGW static build.
* Fix dependencies.
* Fix IR_BUFHDR assembly.
* FFI: Support FFI numbers in string.format() and buf:putf().
* ARM64: More improvements to the generation of immediates.
* String buffers, part 2d: basic string buffer methods.
* String buffers, part 2c: abstract out string.format.
* Abstract out on-demand loading of FFI library.
* String buffers, part 2b: Add extended string buffers. Use in serializer.
* String buffers, part 2a: internal SBuf reorg. Use full pointers.
* DynASM/ARM64: Fix ADRP encoding with absolute address.
* change: Introduce a new macro LUAJIT_TEST_FIXED_ORDER for testing so that lua table would have predictable key iteration order.
* DynASM: Bump version to 1.5.0.
* DynASM/ARM64: Add .long expr. Add .quad/.addr expr + refs.
* DynASM/x86: Fix x64 .aword refs. Add .qword, .quad, .addr and .long.
* FFI/ARM64/OSX: Fix vararg call handling.
* Prevent compile of __concat with tailcall to fast function.
* Fix IR_RENAME snapshot number. Follow-up fix for a32aeadc.
* Fix compiler warning.
* change: the order of pairs was indeterminate because it depends on str->sid. (#119)
* DynASM: Fix global label references
* DynASM/ARM64: Add VREG support.
* Fix build with busybox grep.
* NetBSD: Use PROT_MPROTECT() and disable getentropy().
* Allow disabling the serializer.
* BSD: Fix build with BSD grep.
* Fix .bat file builds.
* OSX: Fix build by hardcoding external frame unwinding.
* String buffers, part 1: object serialization.
* Reorganize lightuserdata interning code.
* Upgrade docs to HTML5. It's about time.
* FFI: Handle zero-fill of struct-of-NYI.
* ARM64: Improve generation of immediates.
* Handle on-trace OOM errors from helper functions.
* Use weak guards for on-trace allocations.
* PPC: Fix GG_State loads.
* MIPS: Fix handling of long-range spare jumps.
* Cleanup and enable external unwinding for more platforms.
* iOS: Don't use getentropy() since it's disallowed in the App Store.
* Linux/ARM64: Make mremap() non-moving due to VA space woes.
* Fix trace exit register dump for some archs.
* feature: enable debuginfo in our luajit build by default.
* Fix binary number literal parsing.
* Fix warning.
* bugfix: fixed gcc warnings when -DLUAJIT_DISABLE_JIT is specified.
* Detect SSE4.2 support dynamically
* bugfix: jit.prngstate() did not allow integer arg which broke backward compatibility.
* Fix BC_NOT on s390x
* doc: fixed use case for jit.prngstate(). (#113)
* bugfix: fixed compatibility regression with MinGW gcc
* bugfix: lua_resetthread: we forgot to reset the "exdata2" field.
* feature: added new Lua API functions thread.exdata2 as well as C API funcs lua_getexdata2 and lua_setexdata2.
* feature: added new LuaJIT C API lua_resetthread.
* doc: added new section "New C API".
* Actually move saved_jit_base to the end of global_State
* Move all luajit2-specific elements to the end of state structs
* bugfix: compilation error "for’ loop initial declarations are only allowed in C99 mode" would happen with older gcc versions like 4.8.
* Add support for full-range 64 bit lightuserdata.
* Adjust SSE4.1 str_hash to replace hash_sparse and hash_dense
* Make jit.prngstate use the new PRNG code
* ARM64: Followup fix for exit branch patching.
* x64: Fix 64 bit shift code generation.
* Revert "no clever memory over-reads when LUAJIT_USE_VALGRIND is defined."
* Revert "optimize: lj_str_new: tests the full hash value before doing the full string comparison on hash collisions. thanks Shuxin Yang for the patch."
* Revert "optimize: lj_str_new: uses randomized hash functions based on crc32 when -msse4.2 is specified."
* Revert "bugfix: fixed compatibility regression with MinGW gcc. this bug had appeared in commit 7923c63."
* Revert "bugfix: FFI C parsers could not parse some C constructs like `__attribute((aligned(N)))` and `#pragma`."
* Revert "style: minor coding style fixes."
* Revert "bugfix: fixed a clang warning in lj_str.c regarding unused str_fastcmp() when macro LUAJIT_USE_VALGRIND is defined."
* Revert "Fix arm64 register allocation issue for XLOAD."
* Another fix for lua_yield() from C hook.
* Fix lua_yield() from C hook.
* LJ_GC64: Always snapshot functions for non-base frames.
* Handle old OSX/iOS without getentropy().
* Fix Clang build.
* Android/ARM: Fix build with recent NDK.
* Fix compiler warning.
* Fix OSX build.
* Follow-up fix for iOS build.
* OSX/iOS: Handle iOS simulator and ARM64 Macs.
* Fix pointer check for non-GC64 mode.
* Windows: Fix NtAllocateVirtualMemory prototype.
* Add jit.security().
* Redesign and harden string interning.
* Use a securely seeded global PRNG for the VM.
* Cleanup some arch defines and fix builds.
* ARM: Implement FLOAD from GG_State.
* Improve assertions.
* Optimize table length computation with hinting.
* Remove pow() splitting and cleanup backends.
* Cleanup math function compilation and fix inconsistencies.
* Don't compile math.modf() anymore.
* Fix math.min()/math.max() inconsistencies.
* Fix narrowing of unary minus.
* Cleanup CPU detection and tuning for old CPUs.
* ARM64: Fix {AHUV}LOAD specialized to nil/false/true.
* ARM, ARM64, PPC: Fix TSETR fallback.
* Remove unused file.
* Windows: Make actual use of internal allocator optimization.
* Patch for PPC64 support
* FFI/ARM64: Fix pass-by-value struct calling conventions.
* Make string to number conversions fail on NUL char.
* Remove support for de-facto dead archs.
* DynASM/x86: Fix BMI instructions.
* Minor fixes.
* MIPS: Add MIPS64 R6 port.
* Fix string.char() recording with no arguments.
* Followup fix for embedded bytecode loader.
* travis-ci: add -DLUAJIT_DISABLE_GC64 mode to the tests matrix.
* Revert "Move all register allocations out of the asm_href loop"
* Revert "ARM: Fix up condition codes for conditional arithmetic insn"
* Revert "Fix arm64 register allocation issue for XLOAD." from Patrick Galizia (commit 49c4d0f4e).
* Revert "bugfix: fixed a segfault when unsinking 64-bit pointers."
* Revert "bugfix: committed a much less aggressive patch to fix an infinite loop in lua table lookups."
* Properly fix pointer diff in string.find().
* x64: Enable LJ_GC64 mode by default.
* FFI: Eliminate hardcoded string hashes.
* Fix interaction between profiler hooks and finalizers.
* Don't use STRREF for pointer diff in string.find().
* doc: readme.md: improve completeness and readability.
* Move all register allocations out of the asm_href loop
* bugfix: committed a much less aggressive patch to fix an infinite loop in lua table lookups.
* Revert "bugfix: thanks Julien Desgats for the report and Peter Cawley for the patch."
* Fix unsinking of 64 bit constants.
* tests: now --- err section accepts perl regex refs.
* [aarch64] Allocate LJ_TISNUM early
* [aarch64] Fix crash with side traces under register pressure
* thread.exdata: Build recff_thread_exdata only for LJ_HASFFI
* doc: reword description of table.isempty()
* jit.prngstate: Return a sane value (0) for LUAJIT_DISABLE_JIT
* Fix os.date() for timezone change awareness
* ARM64: Add support for FNMADD and FNMSUB.
* aarch64: Use the xzr register whenever possible
* Remove redundant emit_check_ofs
* Fix arm64 register allocation issue for XLOAD.
* Fix arm64 register allocation issue for XLOAD.
* Added initial s390x support.
* doc: mentioned the increase in the maximum number of upvalues in the miscellaneous section.
* feature: increased the maximum number of upvalues allowed from 60 to 120.
* ARM: Fix up condition codes for conditional arithmetic insn
* bugfix: thanks Julien Desgats for the report and Peter Cawley for the patch.
* bugfix: guarded the jit_prngstate builtin with the LJ_HAS_JIT macro.
* style: minor coding style fixes.
* feature: luajit -bl: dump the constant tables (KGC and KN) for each lua proto object as well.
* bugfix: fixed a segfault when unsinking 64-bit pointers.
* travis: introduced a matrix of XCFLAGS and added the GC64 test mode to it.
* travis-ci: updated image to xenial.
* doc: section layout changes.
* doc: added the markdown version of the README.
* tests: updated our unit tests to reflect recent changes in commit 864c72e31.
* bugfix: ffi.C.FUNC(): it lacked a write barrier which might lead to use-after-free issues and memory corruptions.
* bugfix: LuaJIT's jit.v module might lead to segfaults due to buggy Lua stack traversal code. thanks Thibault Charbonnier for the report in #40.
* ARM64: Avoid side-effects of constant rematerialization.
* feature: implemented the new table.nkeys Lua API.
* feature: implemented the new table.isarray() builtin function which returns true for purely array-like Lua tables and false otherwise.
* docs: added documentation for table.isempty().
* feature: implemented the table.isempty() Lua API.
* doc: minor style fix.
* bugfix: we broke the arm build in the commit c844a613. thanks Alec Muffett for the report in #37.
* bugfix: fixed assertion failure "lj_record.c:92: rec_check_slots: Assertion `nslots <= 250' failed" found by stressing our edgelang compiler.
* Revert "bugfix: 16-bit MCLink.mapofs and GCtrace.nsnapmap fields would overflow for large Lua programs, leading to segfaults; enlarged them to 32-bit."
* Improve luaL_addlstring().
* Fix os.date() for wider libc strftime() compatibility.
* bugfix: 16-bit MCLink.mapofs and GCtrace.nsnapmap fields would overflow for large Lua programs, leading to segfaults; enlarged them to 32-bit.
* change: run the LJ_64 branch in asm_hrefk when LUAJIT_USE_VALGRIND and LUAJIT_ENABLE_GC64 are both supplied.
* testing: added 2 more (passing) tests for the thread.exdata API.
* feature: implemented the new Lua and C API functions for thread exdata.
* Actually implement maxirconst trace limit.
* MIPS/MIPS64: Fix TSETR barrier (again).
* Fix memory probing allocator to check for valid end address, too.
* DynASM/x86: Fix vroundps/vroundpd encoding.
* Revert "debugging: added an (expensive) assertion to check GC objects in current stack upon trace exiting. thanks Mike Pall."
* DynASM: Fix warning.
* ARM64: Fix exit stub patching.
* ARM64: Fix write barrier in BC_USETS.
* feature: luajit.h: defined the macro OPENRESTY_LUAJIT for our branch of LuaJIT.
* doc: documented what we change in our OpenResty branch of LuaJIT.
* Windows: Add UWP support, part 1.
* From Lua 5.3: assert() accepts any type of error object.
* x86: Disassemble FMA3 instructions.
* DynASM/x86: Add FMA3 instructions.
* added copyright notices for our own patches and additions.
* Give expected results for negative non-base-10 numbers in tonumber().
* bugfix: fixed compatibility regression with MinGW gcc. this bug had appeared in commit 7923c63.
* change: travis: used `cd -` instead of `cd ../luajit2`.
* Revert "bugfix: io module: read("*a") fails to automatically resume the read() system call upon EINTR. thanks Michal Cichra for the report in openresty/resty-cli#35."
* feature: implemented the table.clone() builtin Lua API.
* DynASM/x86: Add BMI1 and BMI2 instructions.
* MIPS64: Fix soft-float +-0.0 vs. +-0.0 comparison.
* Fix ARMv8 (32 bit subset) detection.
* Fix IR_BUFPUT assembly.
* bugfix: io module: read("*a") fails to automatically resume the read() system call upon EINTR. thanks Michal Cichra for the report in openresty/resty-cli#35.
* feature: implemented new API function jit.prngstate() for reading or setting the current PRNG state number used in the JIT compiler.
* MIPS64: Fix xpcall() error case.
* ARM64: Fix xpcall() error case (really).
* ARM64: Fix xpcall() error case.
* MIPS64: Fix register allocation in assembly of HREF.
* ARM64: Fix assembly of HREFK.
* optimize: use more appressive JIT compiler parameters as the default to help large OpenResty Lua apps.
* LJ_GC64: Make ASMREF_L references 64 bit.
* LJ_GC64: Fix ir_khash for non-string GCobj.
* MIPS64: Hide internal function.
* x64/LJ_GC64: Fix type-check-only variant of SLOAD.
* PPC: Add soft-float support to JIT compiler backend.
* x64/LJ_GC64: Fix fallback case of asm_fuseloadk64().
* travis: added luajit2-test-suite.
* PPC: Add soft-float support to interpreter.
* x64/LJ_GC64: Fix emit_rma().
* bugfix: fixed a clang warning in lj_str.c regarding unused str_fastcmp() when macro LUAJIT_USE_VALGRIND is defined.
* bugfix: FFI C parsers could not parse some C constructs like `__attribute((aligned(N)))` and `#pragma`.
* Remove unused define.
* FreeBSD/x64: Avoid changing resource limits, if not needed.
* MIPS64: Add soft-float support to JIT compiler backend.
* RELEASE LuaJIT-2.1.0-beta3
* Update changelog.
* FFI: Update docs on compiler limitations.
* Add FOLD rules for mixed BAND/BOR with constants.
* FFI: Compile bitfield loads/stores.
* MIPS64: Fix stores of MULTRES.
* MIPS64: Fix write barrier in BC_USETV.
* ARM64: Fix stores to vmstate.
* Document added C API extensions from Lua 5.2/5.3.
* From Lua 5.2: Add lua_tonumberx() and lua_tointegerx().
* From Lua 5.2: Add luaL_setmetatable().
* From Lua 5.2: Add luaL_testudata().
* From Lua 5.3: Add lua_isyieldable().
* From Lua 5.2: Add lua_copy().
* From Lua 5.2: Add lua_version().
* Refactor with LUA_OK.
* Allow building on Haiku OS.
* Add some more changes and extensions from Lua 5.2.
* Remove old Lua 5.0 compatibility defines.
* ARM64: Add big-endian support.
* optimize: lj_str_new: uses randomized hash functions based on crc32 when -msse4.2 is specified.
* x64/LJ_GC64: Fix emit_loadk64().
* LJ_GC64: Fix BC_CALLM snapshot handling.
* x64/LJ_GC64: Fix assembly of CNEWI with 64 bit constant pointer.
* ARM64: Fix Nintendo Switch build.
* ARM64: Fix XLOAD/XSTORE with FP operand.
* Remove Lua 5.0 compatibility defines.
* optimize: lj_str_new: tests the full hash value before doing the full string comparison on hash collisions. thanks Shuxin Yang for the patch.
* feature: added the bytecode option `L` to display lua source line numbers.
* MIPS64, part 2: Add MIPS64 hard-float JIT compiler backend.
* Fix FOLD rules for math.abs() and FP negation.
* Fix soft-float math.abs() and negation.
* x64/LJ_GC64: Fix warning for DUALNUM build.
* x64/LJ_GC64: Fix (currently unused) integer stores in asm_tvptr().
* ARM64: Cleanup and de-cargo-cult TValue store generation.
* Make checkptrGC() actually work.
* ARM64: Fix AREF/HREF/UREF fusion.
* Fix extension docs about package.searchers.
* LJ_GC64: Add build options and install instructions.
* Add some more extensions from Lua 5.2/5.3.
* ARM64: Remove unused variables in disassembler.
* ARM64: Fuse BOR/BXOR and BNOT into ORN/EON.
* Add "proto" field to jit.util.funcinfo().
* ARM64: Use the correct FUSE check.
* ARM64: Fuse BOR(BSHL, BSHR) into EXTR/ROR.
* Add missing FOLD rule for 64 bit shift+BAND simplification.
* ARM64: Fix code generation for S19 offsets.
* ARM64: Fuse various BAND/BSHL/BSHR/BSAR combinations.
* ARM64: Fuse FP multiply-add/sub.
* ARM64: Fuse XLOAD/XSTORE with STRREF/ADD/BSHL/CONV.
* ARM64: Reorganize operand extension definitions.
* ARM64: Add missing ldrb/strb instructions to disassembler.
* ARM64: Fix pc-relative loads of consts. Cleanup branch codegen.
* ARM64: Make use of tbz/tbnz and cbz/cbnz.
* Document 47 bit limit for lightuserdata.
* Eliminate use of lightuserdata derived from static data pointers.
* ARM64: Emit more efficient trace exits.
* Generalize deferred constant handling in backend to 64 bit.
* ARM64: Reject special case in emit_isk13().
* ARM64: Allow full VA range for mcode allocation.
* ARM64: Add JIT compiler backend.
* Whitespace.
* Fix amalgamated build.
* Increase range of GG_State loads via IR_FLOAD with REF_NIL.
* MIPS: Fix TSETR barrier.
* Report parent of stitched trace.
* LJ_GC64: Various followup fixes.
* LJ_GC64: Fix HREF for pointers.
* LJ_FR2: Fix slot 1 handling.
* LJ_GC64: Various fixes.
* LJ_GC64: Fix jit.on/off.
* Fix -jp=a mode for builtins.
* Windows/x86: Add MSVC flags for debug build with exception interop.
* Fix exit status for 'luajit -b'.
* Must preserve J->fold.ins (fins) around call to lj_ir_ksimd().
* Emit bytecode in .c/.h files with unsigned char type.
* Set arg table before evaluating LUA_INIT and -e chunks.
* Properly clean up state before restart of trace assembly.
* MIPS: Support MIPS16 interlinking.
* Revert "Revert "Windows/x86: Add full exception interoperability.""
* x64/LJ_GC64: Fix code generation for IR_KNULL call argument.
* LJ_GC64: Set correct nil value when clearing a cdata finalizer.
* LJ_GC64: Ensure all IR slot fields are initialized.
* LJ_GC64: Allow optional use of the system memory allocator.
* MIPS: Fix build failures and warnings.
* Proper fix for LJ_GC64 changes to asm_href().
* MIPS64, part 1: Add MIPS64 support to interpreter.
* DynASM/MIPS: Add missing MIPS64 instructions.
* Fix compiler warning.
* x64/LJ_GC64: Fix __call metamethod for tailcall.
* Fix collateral damage from LJ_GC64 changes to asm_href().
* Cleanup install docs.
* Use MAP_TRYFIXED for the probing memory allocator, if available.
* x64/LJ_GC64: Add missing backend support and enable JIT compilation.
* LJ_FR2: Add support for trace recording and snapshots.
* LJ_GC64: Update IR type sizes.
* LJ_GC64: Add support for 64 bit GCobj constants in the IR.
* Strip out old infrastructure for 64 bit constants.
* Embed 64 bit constants directly in the IR, using two slots.
* Always walk IR constants in ascending order.
* Simplify GCtrace * reference embedding for trace stitching.
* Make the IR immovable after assembly.
* Add ra_addrename().
* Load SIMD constants with IR_FLOAD from GG_State.
* Add IR_FLOAD with REF_NIL for field loads from GG_State.
* Move common 32/64 bit in-memory FP constants to jit_State.
* LJ_GC64: Introduce IRT_PGC.
* Workaround for MinGW headers lacking some exception definitions.
* Fix dependencies.
* removed the GCC 4 requirement for x86 for older systems like Solaris i386.
* Revert "Windows/x86: Add full exception interoperability."
* Windows/x86: Add full exception interoperability.
* Use the GDB JIT API in a thread-safe manner.
* Constrain value range of lj_ir_kptr() to unsigned 32 bit pointers.
* Rewrite memory block allocator.
* x64/LJ_GC64: Fix BC_UCLO check for fast-path.
* x86: Improve disassembly of BMI2 instructions.
* Windows/x64/LJ_GC64: Fix math.frexp() and math.modf() (again).
* x86: Generate BMI2 shifts and rotates, if available.
* x86: Detect BMI2 instruction support.
* x64/LJ_GC64: Fix JIT glue code in interpreter.
* Fix compiler warnings.
* Fix formatting of some small denormals at low precision.
* ARM: Fix build problem with external frame unwinding.
* MIPS soft-float: Fix code generation for HREF.
* DynASM/x86: Add ADX instructions.
* RELEASE LuaJIT-2.1.0-beta2
* Update changelog.
* Fix Android/x86 build.
* Fix build with JIT disabled.
* Use internal implementation for converting FP numbers to strings.
* Windows/x64/LJ_GC64: Fix math.frexp() and math.modf().
* MIPS: Add soft-float support to JIT compiler backend.
* Generalize LJ_SOFTFP dependencies in lj_asm.c.
* PS4: Switch default build to amalgamated and LJ_GC64 mode.
* MIPS: Switch to dual-number mode. Fix soft-float interpreter.
* Add proper DynASM dependency.
* FFI: Parse #line NN and #NN.
* Always merge snapshots without instructions inbetween.
* Rollback due to HREFK + load fwd must restore guardemit state.
* DynASM/x64: Fix for full VREG support.
* x86: Disassemble AES instructions.
* x86: Improve disassembly of AVX shift instructions.
* DynASM/x86: Add AVX AES instructions.
* ARM: Add external frame unwinding.
* DynASM/x64: Add full VREG support.
* FFI: Properly unsink non-standard cdata allocations.
* Makefile: ensure we always install the symlink for "luajit".
* MIPS soft-float, part 2: Add soft-float FFI support.
* MIPS soft-float, part 1: Add soft-float support to interpreter.
* x64/LJ_GC64: Fix stack growth in vararg function setup.
* Document minimum compiler requirements for 2.1 branch.
* Update changelog.
* iOS/ARM64: Add build instructions.
* x86/x64: Disassemble AVX/AVX2 instructions.
* Whitespace.
* DynASM: Bump version to 1.4.
* DynASM/x86: Add AVX and AVX2 opcodes.
* DynASM/x86: Add AES-NI opcodes.
* Fix pairs() recording.
* Add collectgarbage("isrunning").
* FFI: Add ssize_t declaration.
* Parse Unicode string escape \u{XX...}.
* Re-enable trace stitching.
* ARM64: Fix ELF bytecode saving.
* RELEASE LuaJIT-2.1.0-beta1
* ARM64: Add support for saving bytecode as object files.
* Update changelog.
* Add note about bytecode compatibility wrt. LJ_GC64/LJ_FR2.
* Remove LuaJIT 1.x change history.
* Parse binary number literals (0bxxx).
* ARM64: Fix __call metamethod handling for tail calls.
* Fix Xbox One port.
* Disable table allocation bump optimization (for now).
* Add Xbox One port.
* Fix table allocation bump optimization (sigh).
* x64/LJ_GC64: Fix ipairs().
* Fix table allocation bump optimization (again).
* x64/LJ_GC64: Fix upvalue write barrier.
* x64/LJ_GC64: Fix setmetatable().
* Fix table allocation bump optimization.
* LJ_GC64: Fix frame unwinding.
* Fix for last commit.
* Bump table allocations retroactively if they grow later on.
* Cleanup.
* x64: Add LJ_GC64 mode interpreter.
* Fix broken Windows build.
* Disable trace stitching (for now) due to a design mistake.
* ARM64: Fix math.floor/math.ceil for string args.
* DynASM/PPC: Add sub/shift/rotate/clear instruction aliases.
* DynASM/PPC: Add support for parameterized shifts/masks.
* PPC: Fix cross-endian builds.
* PPC: Fix write barrier in BC_TSETR.
* PPC64: Add build infrastructure.
* LJ_FR2: Fix bytecode generation for method lookups.
* DynASM/PPC: Add missing PPC64 instructions and various extensions.
* iOS/ARM64: Fix target OS detection.
* LJ_GC64: Return true for ffi.abi("gc64").
* ARM64: Add FFI support.
* iOS/ARM64: The frame pointer is required.
* LJ_FR2: Fix lua_settable() and lua_setfield().
* DynASM/ARM64: Fix checks for scaled immediates.
* Fix MSVC build.
* ARM64: Update docs.
* ARM64: Compatibility fixes for Clang.
* PPC/e500: Drop support for this architecture.
* ARM64: Add build infrastructure and initial port of interpreter.
* Add LJ_GC64 mode: 64 bit GC object references.
* Add LJ_FR2 mode: Two-slot frame info.
* DynASM/ARM64: Various fixes.
* Cleanup of TValue setters. No functional changes.
* Cleanup of memory vs. GC sizes. No functional changes.
* x86: Fix stack slot reservation for FP math functions.
* Disable debug info for generated *.S to avoid confusing gdb.
* Avoid dependency on frame layout for LJ_POST_FIXCOMP.
* Cleanup of frame handling. No functional changes.
* x86/x64: Drop internal x87 math functions. Use libm functions.
* x86/x64: Call external symbols directly from interpreter code.
* DynASM/ARM64: Initial commit of ARM64 module.
* debugging: added an (expensive) assertion to check GC objects in current stack upon trace exiting. thanks Mike Pall.
* Revert "applied a new versiono of the patch for DragonFlyBSD compatibility from lhmwzy in #2."
* FFI: Add ffi.typeinfo().
* Eliminate hmask guard for forwarded HREFK.
* Compile debug.getmetatable().
* fixed compilation errors on Solaris when -DLUA_USE_TRACE_LOGS is enabled.
* disabled trace stitching by default for now since it may trigger random lua stack corruptions when using with ngx_lua.
* feature: jit.dump: output Lua source location after every BC.
* bugfix: fixed build regression on i386 introduced by the LUA_USE_TRACE_LOGS feature.
* feature: added internal memory-buffer-based trace entry/exit/start-recording event logging, mainly for debugging bugs in the JIT compiler. it requires -DLUA_USE_TRACE_LOGS when building.
* feature: save g->jit_base to g->saved_jit_base before lj_err_throw clears g->jit_base which makes it impossible to get Lua backtrace in such states.
* checked in Mike Pall's interim fix for the assertion failure "ir->o == IR_CONV && ir->op2 == ((IRT_NUM<<5)|IRT_INT)" in lj_snap_replay.
* applied a new versiono of the patch for DragonFlyBSD compatibility from lhmwzy in #2.
* Fix FOLD of string concatenations.
* reverted the patch for DragonFlyBSD because it is wrong.
* Fix alias analysis for table.len vs. table.clear.
* relaxed the heuristic limit further for loopunroll.
* bugfix: fixed a typo in commit 395fc489 for DragonFly BSD. thanks lhmwzy for the report.
* Specialize to ffid for non-monomorphic builtins.
* feature: applied John Marino's patch for compiling LuaJIT on DragonFlyBSD. thanks lhmwzy for #1.
* Fix Xbox 360 build.
* Use own buffer for LOOP substitution table.
* Don't start a trace on side exit to non-Lua function.
* turned off an optimization for 64-bit architectures when valgrind is enabled.
* Don't start stitched trace with CALLMT, too.
* Fix KBASE for Lua functions below stitched fast functions.
* Fix final snapshot for trace stitching.
* x86: Fix stack slot counting for IR_CALLA (affects table.new).
* Show trailing NYI item for stitched trace with -jv.
* Revert "added Mike Pall's temporary workaround for fixing a segmentation fault when a lua error is thrown."
* Add trace stitching.
* Revert "Revert "temporary work-around for the bogus "loop unroll limit reached" failures.""
* added Mike Pall's temporary workaround for fixing a segmentation fault when a lua error is thrown.
* Revert "temporary work-around for the bogus "loop unroll limit reached" failures."
* Add table.clear().
* Abstract out post-registration handling of pre-registered modules.
* no clever memory over-reads when LUAJIT_USE_VALGRIND is defined.
* fixed a regression in the last commit.
* temporary work-around for the bogus "loop unroll limit reached" failures.
* allow use of system malloc on x86_64.
* updated .gitignore a bit.
* FFI: Compile conversions from functions to function pointers.
* Compile BC_TSETM, e.g. {1,2,3,f()}.
* Compile getfenv(0).
* FFI: Compile lightuserdata to void * conversion.
* Add table.new().
* Properly fix loading of embedded bytecode.
* Fix loading of embedded bytecode.
* Fix profiler/hook interaction.
* Fix build with profiler but without JIT compiler.
* Protect g->hookmask with lock when using profiler thread.
* Fix profiler stack dump for string chunks.
* Fix -jp=a mode.
* Low-overhead profiler, part 7: console ports.
* Fix non-FFI build.
* Fix PS3 build.
* Low-overhead profiler, part 6: documentation.
* Add -jp=fl mode and minor fixes.
* Low-overhead profiler, part 5: add annotation mode for -jp.
* Switch from -jp=n<top_n> to -jp=m<min_pct> as a threshold.
* Load jit.util module on demand (saves 1 KB).
* Low-overhead profiler, part 4: JIT compiler support.
* Add notice when no samples were collected by -jp.
* Fix string.format("%#X", 1).
* Install new jit.* modules, too.
* Add low-overhead profiler. Part 3: command line interface.
* Add low-overhead profiler. Part 2: low-level Lua API.
* Add internal API to add sub-modules to package.preload.
* Add low-overhead profiler. Part 1: interpreter, low-level C API.
* Add common helper for pointer to string formatting.
* Save currently executing lua_State in g->cur_L.
* Use g->jit_base for on/off-trace detection.
* Fix compiler warning.
* Refactor string buffer growth helpers.
* Fix -jdump broken by 2ef2c34d.
* Minor change to lj_lib_pushcc().
* FFI: Refactor ffi.gc()/__gc. Compile ffi.gc(cdata, nil), too.
* FFI: Compile VLA/VLS and large cdata allocs with default initialization.
* Refactor lj_cdata_newv().
* Don't try to reinvent memcpy().
* Remove use of printf() in luajit.c.
* Get rid of module() in all internal modules.
* Compile bit.tohex().
* Another fix for bit.tohex().
* Compile string.format().
* FFI: Load FFI library on-demand for -jdump.
* Refactor CCallInfo representation for split arguments.
* Fix bit.tohex() broken by 7d5acc29.
* Adapt -jdump to unified string.format("%x", -1) behavior.
* Abstract out default BUFHDR emitter.
* Fix compiler warnings.
* Big renaming of string buffer/formatting/conversion functions.
* Refactor raw object to pointer or string conversions.
* Refactor internal string formatting.
* Refactor bit.tohex().
* Refactor string.format().
* Compile string.find() for fixed string searches (no patterns).
* Partially refactor string.find().
* Inhibit CSE for table.concat().
* Compile __concat metamethod.
* Fix BC_ISTYPE for WIN64 calling conventions.
* Compile table.concat().
* Refactor table.concat().
* Optimize BUFHDR code generation.
* Compile string.rep().
* Refactor string.rep().
* Abstract out some common buffer operations.
* Replace string.len with bytecode builtin.
* Change semantics of buffer ops to simplify CSE and DCE.
* Fix MSVC build.
* Compile string.reverse(), string.lower(), string.upper().
* Refactor string.reverse(), string.lower(), string.upper().
* Fix FOLD rules for appending to string buffers.
* Fix os.date() broken by d38d10a3.
* Compile string.char().
* Use explicit conversion type for IR_TOSTR. Add char conversion.
* Fuse string creation into concats. Optimize single-char concats.
* Always emit TOSTR for non-string CAT operands. Fuse in backend.
* Fix PHI barrier in BUFPUT FOLD rule.
* Combine IR instruction dispatch for all assembler backends.
* Use same HREF+EQ/NE optimization in all assembler backends.
* Reorganize generic operations common to all assembler backends.
* Compile string concatenations (BC_CAT).
* Move buffer shrinking to atomic GC phase.
* Add generic load/store with offset to assembler backends.
* Tune string concatenation.
* Optimize and refactor printing of numbers to buffers.
* Use string buffer for string.format().
* Use string buffer for string.dump().
* Use string buffer for os.date().
* Use string buffer for table.concat().
* Clean up TValue to buffer conversions.
* FFI: Add 64 bit bitwise operations.
* ARM: Rearrange global_State to satisfy imm12 encoding constraints.
* String buffer refactoring, part 4.
* String buffer refactoring, part 3.
* Cleanup lexer source code.
* String buffer refactoring, part 2.
* String buffer refactoring, part 1.
* Replace table.remove with bytecode builtin.
* Fix genlibbc.
* Replace table.getn/foreach/foreachi with bytecode builtins.
* Add special bytecodes for builtins.
* Remove obsolete non-truncating number to integer conversions.
* Don't modify generated libbc header if unchanged.
* Replace math.deg/math.rad with builtin Lua function.
* Add support for embedding LuaJIT bytecode for builtins.
* x86/x64: Drop xmm register/memory move tuning for K8.
* x86: Remove x87 support from interpreter.
* BRANCH LuaJIT-2.1
-------------------------------------------------------------------
Tue Mar 25 16:04:27 UTC 2025 - mcepl@cepl.eu
- Update to version 5.1.2.1.0+git.1741730670.538a821:
* Change handling of nil value markers in template tables.
* REVERT: Change handling of nil value markers in template tables.
* macOS: Fix Apple hardened runtime support and put behind build option.
* macOS: Fix support for Apple hardened runtime.
* Fix handling of nil value markers in template tables.
* Avoid unpatching bytecode twice after a trace flush.
* Fix state restore when recording __concat metamethod.
* Windows: Allow mixed builds with msvcbuild.bat.
* macOS: Add suport for Apple hardened runtime.
* Add compatibility string coercion for fp:seek() argument.
* Windows: Clarify installation directory layout.
* Remove Cygwin from docs, since it's not a supported target.
* Improve CLI signal handling on POSIX.
* FFI: Add pre-declared int128_t, uint128_t, __int128 types.
* Use dylib extension for iOS installs, too.
* Change handling of nil value markers in template tables.
* Fix recording of BC_VARG.
* Reject negative getfenv()/setfenv() levels to prevent compiler warning.
* Bump copyright date.
* Force fallback source name for stripped bytecode.
* Remove dependency on <limits.h>.
* Restore state when recording __concat metamethod throws OOM.
* MIPS64: Fix pcall() error case.
* Fix detection of inconsistent renames due to sunk values.
-------------------------------------------------------------------
Fri Mar 7 05:12:33 UTC 2025 - Muhammad Akbar Yanuar Mantari <mantarimay@pm.me>
- Enable lua 5.2 compatibility (Needed for build aegisubs)
-------------------------------------------------------------------
Fri Dec 6 07:23:43 UTC 2024 - Adrian Schröter <adrian@suse.de>
- No loongarch64 support yet
- turn around the logic to known supported architectures
-------------------------------------------------------------------
Wed Nov 20 07:23:12 UTC 2024 - mcepl@cepl.eu
- Update to version 5.1.2.1.0+git.1731601260.fe71d0f:
* Windows: Allow amalgamated static builds with msvcbuild.bat.
* Always close profiler output file.
* Fix override of INSTALL_LJLIBD in the presence of DESTDIR.
* Fix bit op coercion for shifts in DUALNUM builds.
-------------------------------------------------------------------
Wed Nov 06 16:31:22 UTC 2024 - mcepl@cepl.eu
- Update to version 5.1.2.1.0+git.1727870382.97813fb:
* macOS: Remove obsolete -single_module flag.
* macOS: Workaround for buggy XCode 15.0 - 15.2 linker.
* macOS: Fix macOS 15 / Clang 16 build.
* Fix bit op coercion in DUALNUM builds.
* Fix compiliation of getmetatable() for UDTYPE_IO_FILE.
* Remove ancient RtlUnwindEx workaround for MinGW64.
* Drop unused function wrapper.
-------------------------------------------------------------------
Wed Nov 6 16:25:34 UTC 2024 - Matej Cepl <mcepl@cepl.eu>
- Update luajit-s390x.patch from
gh#openresty/luajit2!195 (following the advice on
gh#LuaJIT/LuaJIT!631#issuecomment-2459913131).
-------------------------------------------------------------------
Mon Aug 26 18:41:09 UTC 2024 - mcepl@cepl.eu
- Update to version 5.1.2.1.0+git.1724512491.f725e44:
* Fix limit check in narrow_conv_backprop().
* Always use IRT_NIL for IR_TBAR.
* ARM64: Use ldr literal to load FP constants.
* FFI: Add missing coercion when recording 64-bit bit.*().
* ARM64: Make tobit conversions match JIT backend behavior.
* ARM: Make hard-float tobit conversions match JIT backend behavior.
* FFI: Drop finalizer table rehash after GC cycle.
* Fix another potential file descriptor leak in luaL_loadfile*().
* MIPS32: Fix little-endian IR_RETF.
* Correctly close VM state after early OOM during open.
* Fix potential file descriptor leak in luaL_loadfile*().
* Reflect override of INSTALL_LJLIBD in package.path.
* ARM64: Use movi to materialize FP constants.
* Add more FOLD rules for integer conversions.
* Different fix for partial snapshot restore due to stack overflow.
* Fix IR_ABC hoisting.
* Limit CSE for IR_CARG to fix loop optimizations.
* Call math.randomseed() without arguments to seed from system entropy.
* Restore state when recording __concat metamethod throws an error.
* Add build flag LUAJIT_DISABLE_TAILCALL to disable tailcall generation.
* Clarify that lj_buf_shrink() does not keep any buffer data.
* OSX: Fix installed luajit.pc.
* FFI: Fix various issues in recff_cdata_arith.
* Fix predict_next() in parser (for real now).
* FFI: Fix __tostring metamethod access to enum cdata value.
* Fix typo.
* Handle partial snapshot restore due to stack overflow.
* Update Nintendo Switch build script.
* Prevent sanitizer warning in snap_restoredata().
* Typo.
* Limit number of string format elements to compile.
* FFI: Clarify scalar boxing behavior.
* Fix internal link in docs.
* OSX/iOS: Fix SDK incompatibility.
* Windows/MSVC: Cleanup msvcbuild.bat and always generate PDB.
* Fix segment release check in internal memory allocator.
* Fix compiler warning.
* FFI: Turn FFI finalizer table into a proper GC root.
* OSX/iOS: Always generate 64 bit non-FAT Mach-O object files.
* Show name of NYI bytecode in -jv and -jdump.
* Use generic trace error for OOM during trace stitching.
* Fix serialization format docs.
* Handle all types of errors during trace stitching.
* Fix recording of __concat metamethod.
* Prevent down-recursion for side traces.
* Check frame size limit before returning to a lower frame.
* FFI: Treat cdata finalizer table as a GC root.
* Handle stack reallocation in debug.setmetatable() and lua_setmetatable().
- Update patches:
- luajit-lua-versioned.patch
- luajit-s390x.patch (update to the latest base code)
-------------------------------------------------------------------
Fri Mar 01 20:05:18 UTC 2024 - mcepl@cepl.eu
- Update to version 5.1.2.1.0+git.1707061634.0d313b2:
* Rework stack overflow handling.
* Preserve keys with dynamic values in template tables when saving bytecode.
* Prevent include of luajit_rolling.h.
* Fix documentation bug about '\z' string escape.
* Fix zero stripping in %g number formatting.
* Fix unsinking of IR_FSTORE for NULL metatable.
* DynASM/x86: Add endbr instruction.
* MIPS64 R2/R6: Fix FP to integer conversions.
* Add cross-32/64 bit and deterministic bytecode generation.
* DynASM/x86: Allow [&expr] operand.
* Check for IR_HREF vs. IR_HREFK aliasing in non-nil store check.
* Respect jit.off() on pending trace exit.
* Simplify handling of instable types in TNEW/TDUP load forwarding.
* Only emit proper parent references in snapshot replay.
* Fix anchoring for string buffer set() method (again).
* ARM: Fix stack restore for FP slots.
* Document workaround for multilib vs. cross-compiler conflict.
* Fix anchoring for string buffer set() method.
* Fix runtime library flags for MSVC debug builds.
* Fix .debug_abbrev section in GDB JIT API.
* Optimize table.new() with constant args to (sinkable) IR_TNEW.
* Emit sunk IR_NEWREF only once per key on snapshot replay.
* Fix last commit.
* x86/x64: Don't fuse loads across IR_NEWREF.
* Improve last commit.
* x86/x64: Don't fuse loads across table.clear.
* x86/x64: Add more red zone checks to assembler backend.
* Add stack check to pcall/xpcall.
* Invalidate SCEV entry when returning to lower frame.
* FFI: Fix pragma push stack limit check and throw on overflow.
* ARM64: Fix disassembly of ldp/stp offsets.
* Check for upvalue state transition in IR_UREFO.
* x64: Properly fix __call metamethod return dispatch.
* Windows/x86: _BitScan*64 are only available on 64 bit archs.
* Add 'cc' file type for saving bytecode.
* FFI/Windows: Fix type declaration for int64_t and uint64_t.
* FFI: Fix dangling reference to CType in carith_checkarg().
* DynASM/ARM64: Support ldp/stp of q registers.
* ARM64: Use ADR and ADRP to form constants.
* ARM64: Fix disassembly of U12 loads.
* ARM64: Unify constant register handling in interpreter.
* ARM: Fix register hint for FFI calls with FP results.
* ARM64: Fix register hint for FFI calls with FP results.
* ARM64: Restore fp before sp in C stack unwinders.
* FFI: Fix ffi.abi("pauth").
* Maintain chain invariant in DCE.
* LJ_FR2: Fix stack checks in vararg calls.
* Follow-up fix for stack overflow handling cleanup.
* Handle OOM error on stack resize in coroutine.resume and lua_checkstack.
* Restore cur_L for specific Lua/C API use case.
* Consistently use 64 bit constants for 64 bit IR instructions.
* ARM64: Fix IR_HREF code generation for constant FP keys.
* ARM64: Fuse negative 32 bit constants into arithmetic ops again.
* Handle all stack layouts in (delayed) TRACE vmevent.
* Add missing coercion when recording select(string, ...)
* Cleanup stack overflow handling.
* Windows/ARM64: Add MSVC cross-build support for x64 to ARM64.
* IR_MIN/IR_MAX is non-commutative due to underlying FPU ops.
* ARM64: Set fixed interpreter registers before rethrow.
* Windows/ARM64: Update install docs.
* Windows: Call C++ destructors without compiling with /EHa.
* Windows: Pass scratch CONTEXT record to RtlUnwindEx.
* ARM64: External unwinder already restores non-volatile registers.
* Windows/ARM64: Fix exception unwinding (again).
* Windows/ARM64: Fix typo in exception unwinding.
* FFI: Fix 64 bit shift fold rules.
* Windows/ARM64: Support Windows calling conventions.
* Windows/ARM64: Fix exception unwinding.
* ARM64: Remove unneeded IRCALL_* defs for math intrinsics.
* Fix Cygwin build.
* Allow path overrides in genversion.lua with minilua, too.
* Windows/ARM64: Add initial support.
* Improve architecture detection error messages.
* ARM64: Fuse rotates into logical operands.
* ARM64: Don't fuse sign extensions into logical operands.
* ARM64: Disassemble rotates on logical operands.
* ARM: Fix stack check code generation.
* ARM64: Fix LDP/STP fusion (again).
* ARM64: Ensure branch is in range before emitting TBZ/TBNZ.
* Fix mcode limit check for non-x86 archs.
* ARM64: Improve BC_JLOOP.
* ARM64: Improve integer IR_MUL code generation.
* ARM64: Simplify code generation for IR_STRTO.
* ARM64: Use RID_TMP instead of scratch register in more places.
* ARM64: Improve IR_OBAR code generation.
* ARM64: Improve IR_UREF code generation.
* ARM64: Improve IR_HREF code generation.
* ARM64: Reload BASE via GL instead of spilling it.
* ARM64: Consolidate 32/64-bit constant handling in assembler.
* ARM64: Tune emit_lsptr. Avoid wrong load for asm_prof.
* ARM64: Inline only use of emit_loada.
* ARM64: Improve K13 constant rematerialization.
* Add NaN check to IR_NEWREF.
* Allow override of paths for genversion.lua.
* Fix native MinGW build.
- Also refresh luajit-s390x.patch from the rebased pull request.
-------------------------------------------------------------------
Mon Oct 9 20:09:43 UTC 2023 - Matej Cepl <mcepl@cepl.eu>
- Adjust luajit-lua-versioned.patch to have correct LUA_LJDIR
(bsc#1215592).
-------------------------------------------------------------------
Wed Sep 06 21:51:55 UTC 2023 - mcepl@cepl.eu
- Update to version 5.1.2.1.0+git.1693350652.41fb94d:
* Add randomized register allocation for fuzz testing.
* ARM64: Improve register allocation for integer IR_MUL/IR_MULOV.
* ARM64: Fix register allocation for IR_*LOAD.
* Update external MSDN URL in code.
* FFI/ARM64/OSX: Handle non-standard OSX C calling conventions.
* FFI: Unify stack setup for C calls in interpreter.
* ARM64: Prevent STP fusion for conditional code emitted by TBAR.
* ARM64: Fix LDP/STP fusing for unaligned accesses.
* Handle table unsinking in the presence of IRFL_TAB_NOMM.
* Use fallback name for install files without valid .git or .relver.
* Handle non-.git checkout with .relver in .bat-file builds.
* Fix external C call stack check when using LUAJIT_MODE_WRAPCFUNC.
* Fix predict_next() in parser (again).
- Update luajit-lua-versioned.patch to work with the git checkout
created tarball. The point of the patch is to extend the
version number so that it is always bigger than 2.2.0 version
of moonjit, which is Obsoleted by this package.
-------------------------------------------------------------------
Wed Aug 23 14:18:06 UTC 2023 - mcepl@cepl.eu
- Update to version 2.1.0~beta3+git.1692716794.03c3112:
* Fix typo.
* Handle the case when .git is not a directory.
* Add .gitattributes to dynamically resolve .relver.
* Add .gitattributes to dynamically resolve .relver.
* Fix for last commit: also remove symlink on uninstall.
* Switch to rolling releases: mark v2.1 as production.
* Fix Windows build scripts for rolling releases.
* Switch MSVC and console build scripts to rolling releases.
* Switch build system to rolling releases.
* Update documentation for switch to rolling releases.
* Bump copyright date.
* Remove work-in-progress notice in string buffer docs.
* MIPS: Fix "bad FP FLOAD" assertion.
* Ensure forward progress on trace exit to BC_ITERN.
* ARM64: Add support for ARM64e pointer authentication codes (PAC).
* DynASM/ARM64: Add instructions for ARM64e PAC.
* Fix maxslots when recording BC_VARG, part 3.
* Fix predict_next() in parser.
* MIPS32: Declare that the assembler part uses the FR=0 model.
* ARM64: Fix assembly of HREFK (again).
* Fix frame for more types of on-trace error messages.
* Add workaround for bytecode dump of builtins.
* DynASM: Fix regression due to warning fix.
* Fix base register coalescing in side trace.
* ARM64: Fix assembly of HREFK.
* Fix maxslots when recording BC_VARG, part 2.
* Fix maxslots when recording BC_TSETM.
* Fix maxslots when recording BC_VARG.
* Fix register mask for stack check in head of side trace.
* FFI: Fix ffi.metatype() for non-raw types.
* ARM64: Fix LDP code generation.
* MIPSr6: Add missing files to Makefile install target.
* DynASM: Fix warnings.
* Fix frame for on-trace out-of-memory error.
* Fix handling of instable types in TNEW/TDUP load forwarding.
* Fix compiler warning.
* Fix last commit.
* Print errors from __gc finalizers instead of rethrowing them.
* Fix TDUP load forwarding after table rehash.
* Fix canonicalization of +-0.0 keys for IR_NEWREF.
* Improve error reporting on stack overflow.
* Allow building sources with mixed LF/CRLF line-endings.
* Fix compiler warning.
* Don't fail for Clang builds, which pretend to be an ancient GCC.
* Avoid negation of signed integers in C that may hold INT*_MIN.
* Correct fix for stack check when recording BC_VARG.
* Disable FMA by default. Use -Ofma or jit.opt.start("+fma") to enable.
* FFI: Fix dangling reference to CType. Improve checks.
* ARM64: Fix code generation for IR_SLOAD with typecheck + conversion.
* PS4/PS5: Fix build scripts.
* Avoid assertion in case of stack overflow from stitched trace.
-------------------------------------------------------------------
Mon Aug 21 13:04:40 UTC 2023 - Matej Cepl <mcepl@suse.com>
- Update luajit-s390x.patch from gh#LuaJIT/LuaJIT#631 to the
current form.
-------------------------------------------------------------------
Wed Jan 25 10:52:05 UTC 2023 - Michal Suchanek <msuchanek@suse.de>
- Opencode %make_build to prevent build failure when not defined.
-------------------------------------------------------------------
Tue Nov 22 23:18:13 UTC 2022 - mcepl@cepl.eu
- Update to version 2.1.0~beta3+git.1669107176.46aa45d:
* NX: Always compile with -fPIC.
* Clarify contact page.
* Ensure correct stack top for OOM error message.
* ARM64: Fix IR_SLOAD assembly.
* Fix trace join to BC_JLOOP originating from BC_ITERN.
* Fix compiler warnings.
* Fix last edit.
* Remove dead Wiki links.
* Fix QNX build.
* Add -F option to override filename in jit.bcsave (luajit -b).
* Patch luajit.pc with INSTALL_INC, if customized.
* LJ_GC64: Fix lua_concat().
* Prevent use of RTLD_DEFAULT when NO_RTLD_DEFAULT is defined.
* Improve GC estimation for userdata with attached managed memory.
* Add missing GC steps to string buffer methods.
* x86/x64: Limit VLOAD fusion to simple cases.
* OSX/iOS/ARM64: Fix generation of Mach-O object files.
* Prevent trace start at BC_ITERL after compiled BC_ITERN.
* ARM64: Allow building with unwinding disabled.
* FFI: Fix sizeof expression in C parser for reference types.
* FFI: Fix ffi.alignof() for reference types.
* FFI: Allow ffi.metatype() for typedefs with attributes.
* OSX/iOS/ARM64: Fix bytecode embedding in Mach-O object file.
* LJ_GC64: Fix IR_VARG offset for fixed number of results.
* x86/x64: Fix math.ceil(-0.9) result sign.
* Grammar and spell check.
* Make embedded bytecode readable and forward-compatible.
* DynASM: Fix warnings.
* Fix warnings in minilua.c.
* Update console build instructions.
* Add Nintendo Switch port.
* Avoid zero-sized arrays in jit_State.
* Don't use jit_State during build with JIT disabled.
* DynASM/ARM64: Fix LSL/BFI* encoding with variable registers.
* Fix ITERN loop detection when hook checks are enabled.
* Prevent C compiler undefined-behavior optimization.
* Fix alias analysis for table length forwarding.
* Fix loop initialization in table.foreach().
* LJ_GC64: Fix HREFK optimization.
* Fix recording of __concat metamethod.
* Cleanup of system and architecture support docs.
* Add PS5 port.
* Fix assertions.
* x64: Fix RETHI/RETLO swap after call.
* Fix assertion for LJ_KEYINDEX.
* Add missing check for LJ_KEYINDEX in ITERN recording.
* DynASM/ARM64: Fix NOP instruction for aligment
- Was forced to remove perhaps in future important patches:
- luajit-ppc64-replace-asserts.patch
- 0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch
-------------------------------------------------------------------
Tue Nov 22 16:22:18 UTC 2022 - Michael Gorse <mgorse@suse.com>
- Also exclude s390x from the build (boo#1204557).
-------------------------------------------------------------------
Tue Nov 1 16:01:53 UTC 2022 - Michael Gorse <mgorse@suse.com>
- Add ppc64 and ppc64le to ExcludeArch for now. This can be
removed if we rebase the ppc64 patch, but it is no longer being
maintained and seems unlikely to land upstream (bsc#1204557).
-------------------------------------------------------------------
Tue Oct 4 19:00:14 UTC 2022 - Dirk Müller <dmueller@suse.com>
- exclude luajit from building on riscv64 - not yet supported
-------------------------------------------------------------------
Wed Mar 23 21:33:11 UTC 2022 - mcepl@cepl.eu
- Update to version 2.1.0~beta3+git.1647772157.43ebb949:
* Fix soft-float IR_POW splitting.
* Fix BC_UCLO insertion for returns.
* Fix compiler warning.
* Revert to trival pow() optimizations to prevent inaccuracies.
* Fix string buffer COW handling, part 2.
* Fix install docs for Android.
* *BSD: Fix getentropy() declaration.
* Fix string buffer COW handling.
* Fix command-line argv handling.
* Don't export internal symbol.
* Fix pow() optimization inconsistencies.
* Always exit after machine code page protection change fails.
* Fix FOLD rule for BUFHDR append with intervening buffer use.
* Fix compiled error handling for buffer methods.
* Prevent unroll across BC_ITERN.
* Prevent replay of buffer operation for PHI operands.
* Save trace recorder state around VM event call.
* Bump copyright date.
* FFI: Ensure library is loaded before de-serializing FFI types.
* Fix HREFK forwarding vs. table.clear().
* Fix FOLD rule for BUFHDR append.
* OSX: Disable unreliable assertion for external frame unwinding.
* Limit exponent range in number parsing.
* Fix tonumber("-0") in dual-number mode.
* Limit work done in SINK pass.
* Fix ABC FOLD rule with constants.
* Prevent CSE of a REF_BASE operand across IR_RETF.
* Windows: Fix binary output of jit.bcsave to stdout.
* Fix FOLD rule for x-0.
* Update Android build docs.
* Update Android build docs.
* ARM64: Fix pcall() error case.
* OSX/ARM64: Fix external unwinding.
* Fix string buffer method recording.
* Fix interaction of profiler and ITERN recording.
* Fix compilation of multi-result call to next().
* ARM64: Fix IR_HREF code generation.
* MIPS64: Fix soft-float IR_TOSTR.
* MIPS: Fix register allocation in assembly of HREF.
* Fix compiler warning.
* Windows/x64: Document MSVC flags for C++ exception interoperability.
* FFI: Ensure returned string is alive in ffi.typeinfo().
* OSX/ARM64: Disable unwind info.
* Fix stack allocation after on-trace stack check.
* Fix ITERN blacklisting.
* Ensure ITERN forward progress on interpreter bailout.
* ARM64: Reorder interpreter stack frame and fix unwinding.
* Don't bail out to interpreter to JLOOP originating from ITERN.
* FFI: Don't load PC from non-function object in FFI continuation.
* FFI: Don't load PC from non-function object in FFI continuation.
* FFI: Fix missing cts->L initialization in argv2ctype().
* OSX/ARM64: Disable external unwinding for now.
* Compile table traversals: next(), pairs(), BC_ISNEXT/BC_ITERN.
* Use IR_HIOP for generalized two-register returns.
* Refactor table traversal.
* ARM: Fix symbol display in trace disassembly.
* Refactor IR_TMPREF generation.
* Refactor IR_VLOAD to take an offset.
* MIPS: Fix trace linking.
* String buffers, part 4b: Improve de-serialization checks.
* String buffers, part 4a: Add metatable serialization dictionary.
* Consider slots used by upvalues in use-def analysis.
* Prevent loop in snap_usedef().
* Fix io.close() error message.
* Fix io.close().
* Fix minilua vararg stack handling.
* PS4: Fix compile.
* Avoid out-of-range number of results when compiling select(k, ...).
* String buffers, part 3d: Compile string buffer methods and functions.
* Minor improvements of optimizations.
* Fix error message in lj_lib_checkintrange().
* String buffers, part 2f: Prevent self-put of buffer.
* String buffers, part 3c: Add IRBUFHDR_WRITE mode.
* Fix IRXLOAD_* mode bits description.
* String buffers, part 3b: Change IR_BUFHDR op2 mode bits to mode.
* String buffers, part 3a: Add IR_TMPREF for passing TValues to helpers.
* Add IRCONV_NONE for pass-through INT to I64/U64 type change.
* Fix jit.dump() output for IR_CONV.
- Patches broken in the need of rebase (temporarily switched off):
- 0004-Add-ppc64-support-based-on-koriakin-GitHub-patchset.patch
- luajit-ppc64-replace-asserts.patch
-------------------------------------------------------------------
Wed Oct 27 20:20:36 UTC 2021 - Fabian Vogt <fvogt@suse.com>
- Replace luajit-ppc64-remove-asserts.patch with
luajit-ppc64-replace-asserts.patch to use the proper assert
functions instead of removing them.
-------------------------------------------------------------------
Mon Oct 25 21:55:16 UTC 2021 - Michael Gorse <mgorse@suse.com>
- Add luajit-ppc64-remove-asserts.patch: remove calls to
lua_assert. Fixes the build with our revision (bsc#1191615).
- Add luajit-s390x.patch: add s390x support, based on Fedora's
patch and ported to our revision.
- Remove ExcludeArch, since package currently builds on s390x.
-------------------------------------------------------------------
Thu Oct 7 15:09:55 UTC 2021 - Michael Gorse <mgorse@suse.com>
- Exclude s390x for now. There is a not-yet-upstreamed port
available, but we would need to rebase it for our release.
-------------------------------------------------------------------
Mon Jul 12 17:15:55 UTC 2021 - Matej Cepl <mcepl@suse.com>
- Clean up SPEC. We don't need update-alternatives after all.
-------------------------------------------------------------------
Tue Jun 29 17:17:36 UTC 2021 - munix9@googlemail.com
- Update to version 2.1.0~beta3+git294.e9577376:
* Disable unreliable assertion for external frame unwinding.
* Flush and close output file after profiling run.
* Avoid conflict between 64 bit lightuserdata and ITERN key.
-------------------------------------------------------------------
Thu Jun 24 10:15:35 UTC 2021 - munix9@googlemail.com
- Update to version 2.1.0~beta3+git291.3f9389ed:
* Throw any errors before stack changes in trace stitching.
* DynASM/x86: Add missing escape in pattern.
* Fix HTML.
* String buffers, part 2e: add serialization string dictionary.
* DynASM/ARM64: Fix LSL/BFI* encoding with variable shifts.
* Fix MinGW static build.
* Fix dependencies.
* Fix IR_BUFHDR assembly.
