* `table.sort` was improved further. It now guarentees N*log(N) time
complexity in the worst case.
* Fix https://github.com/Roblox/luau/issues/880
We are also working on fixing final bugs and crashes in the new type
solver.
On the CodeGen front we have a few things going on:
* We have a smarter register allocator for the x86 JIT
* We lower more instructions on arm64
* The vector constructor builtin is now translated to IR
---------
Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com>
Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
Once again, all of our changes this week are for new type solver and the
JIT.
In the new type solver, we fixed cyclic type alias handling and multiple
stability issues.
In the JIT, our main progress was for arm64, where, after lowering 36%
of instructions, we start seeing first Luau functions executing
natively.
For x64, we performed code cleanup and refactoring to allow for future
optimizations.
All of our changes this week have been focused on the new type solver
and the JIT.
As we march toward feature parity with the old solver, we've tightened
up a bunch of lingering issues with overload resolution, unsealed
tables, and type normalization. We've also fixed a bunch of crashes and
assertion failures in the new solver.
On the JIT front, we've started work on an A64 backend, improved the IR
analysis in a bunch of cases, and implemented assembly generation for
the builtin functions `type()` and `typeof()`.
---------
Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com>
Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
* A small subset of control-flow refinements have been added to
recognize type options that are unreachable after a
conditional/unconditional code block. (Fixes
https://github.com/Roblox/luau/issues/356).
Some examples:
```lua
local function f(x: string?)
if not x then return end
-- x is 'string' here
end
```
Throwing calls like `error` or `assert(false)` instead of 'return' are
also recognized.
Existing complex refinements like type/typeof and tagged union checks
are expected to work, among others.
To enable this feature, `LuauTinyControlFlowAnalysis` exclusion has to
be removed from `ExperimentalFlags.h`.
If will become enabled unconditionally in the near future.
* Linter has been integrated into the typechecker analysis so that
type-aware lint warnings can work in any mode
`Frontend::lint` methods were deprecated, `Frontend::check` has to be
used instead with `runLintChecks` option set.
Resulting lint warning are located inside `CheckResult`.
* Fixed large performance drop and increased memory consumption when
array is filled at an offset (Fixes
https://github.com/Roblox/luau/issues/590)
* Part of [Type error suppression
RFC](https://github.com/Roblox/luau/blob/master/rfcs/type-error-suppression.md)
was implemented making subtyping checks with `any` type transitive.
---
In our work on the new type-solver:
* `--!nocheck` mode no longer reports type errors
* New solver will not be used for `--!nonstrict` modules until all
issues with strict mode typechecking are fixed
* Added control-flow aware type refinements mentioned earlier
In native code generation:
* `LOP_NAMECALL` has been translated to IR
* `type` and `typeof` builtin fastcalls have been translated to
IR/assembly
* Additional steps were taken towards arm64 support
* Fix#817
* Fix#850
* Optimize math.floor/ceil/round with SSE4.1
* Results in a ~7-9% speedup on the math-cordic benchmark.
* Optimized table.sort.
* table.sort is now ~4.1x faster (when not using a predicate) and ~2.1x
faster when using a simple predicate. Performance may improve further in
the future.
* Reorganize the memory ownership of builtin type definitions.
* This is a small initial step toward affording parallel typechecking.
The new type solver is coming along nicely. We are working on fixing
crashes and bugs.
A few major changes to native codegen landed this week:
* Fixed lowering of Luau IR mod instruction when first argument is a
constant
* Added VM register data-flow/capture analysis
* Fixed issues with optimizations in unreachable blocks
---------
Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com>
Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
* Fixed incorrect lexeme generated for string parts in the middle of an
interpolated string (Fixes https://github.com/Roblox/luau/issues/744)
* DeprecatedApi lint can report some issues without type inference
information
* Fixed performance of autocomplete requests when suggestions have large
intersection types (Solves
https://github.com/Roblox/luau/discussions/847)
* Marked `table.getn`/`foreach`/`foreachi` as deprecated ([RFC:
Deprecate
table.getn/foreach/foreachi](https://github.com/Roblox/luau/blob/master/rfcs/deprecate-table-getn-foreach.md))
* With -O2 optimization level, we now optimize builtin calls based on
known argument/return count.
Note that this change can be observable if `getfenv/setfenv` is used to
substitute a builtin, especially if arity is different.
Fastcall heavy tests show a 1-2% improvement.
* Luau can now be built with clang-cl (Fixes
https://github.com/Roblox/luau/issues/736)
We also made many improvements to our experimental components.
For our new type solver:
* Overhauled data flow analysis system, fixed issues with 'repeat'
loops, global variables and type annotations
* Type refinements now work on generic table indexing with a string
literal
* Type refinements will properly track potentially 'nil' values (like
t[x] for a missing key) and their further refinements
* Internal top table type is now isomorphic to `{}` which fixes issues
when `typeof(v) == 'table'` type refinement is handled
* References to non-existent types in type annotations no longer resolve
to 'error' type like in old solver
* Improved handling of class unions in property access expressions
* Fixed default type packs
* Unsealed tables can now have metatables
* Restored expected types for function arguments
And for native code generation:
* Added min and max IR instructions mapping to vminsd/vmaxsd on x64
* We now speculatively extract direct execution fast-paths based on
expected types of expressions which provides better optimization
opportunities inside a single basic block
* Translated existing math fastcalls to IR form to improve tag guard
removal and constant propagation
We've made a few small changes to reduce the amount of stack we use when
typechecking nested method calls (eg `foo:bar():baz():quux()`).
We've also fixed a small bytecode compiler issue that caused us to emit
redundant jump instructions in code that conditionally uses `break` or
`continue`.
On the new solver, we've switched to a new, better way to handle
augmentations to unsealed tables. We've also made some substantial
improvements to type inference and error reporting on function calls.
These things should both be on par with the old solver now.
The main improvements to the native code generator have been elimination
of some redundant type tag checks. Also, we are starting to inline
particular fastcalls directly to IR.
---------
Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com>
Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
This week we only have updates to new type solver and JIT. Both projects
are still in the process of being built out. Neither are ready for
general use yet.
In the new solver, we fixed issues with recursive type aliases.
Duplicated type parameters are once again reported, exported types are
being recorder and function argument names are placed inside function
types.
We also made improvements to restore parts of bidirectional type
tracking.
On native code generation side, namecall instruction lowering was fixed,
we fixed inconsistencies in IR command definitions and added utility
function to help with constant folding.
* Fix a bug where reading a property from an unsealed table caused
inference to improperly infer the existence of that property.
* Fix#827
We have also made a lot of progress on the new solver and the JIT. Both
projects are still in the process of being built out. Neither are ready
for general use yet.
We are mostly working to tighten up how the new solver handles
refinements and updates to unsealed tables to bring it up to the same
level as the old solver.
---------
Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com>
Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
* Fixed rare use-after-free in analysis during table unification
A lot of work these past months went into two new Luau components:
* A near full rewrite of the typechecker using a new deferred constraint
resolution system
* Native code generation for AoT/JiT compilation of VM bytecode into x64
(avx)/arm64 instructions
Both of these components are far from finished and we don't provide
documentation on building and using them at this point.
However, curious community members expressed interest in learning about
changes that go into these components each week, so we are now listing
them here in the 'sync' pull request descriptions.
---
New typechecker can be enabled by setting
DebugLuauDeferredConstraintResolution flag to 'true'.
It is considered unstable right now, so try it at your own risk.
Even though it already provides better type inference than the current
one in some cases, our main goal right now is to reach feature parity
with current typechecker.
Features which improve over the capabilities of the current typechecker
are marked as '(NEW)'.
Changes to new typechecker:
* Regular for loop index and parameters are now typechecked
* Invalid type annotations on local variables are ignored to improve
autocomplete
* Fixed missing autocomplete type suggestions for function arguments
* Type reduction is now performed to produce simpler types to be
presented to the user (error messages, custom LSPs)
* Internally, complex types like '((number | string) & ~(false?)) |
string' can be produced, which is just 'string | number' when simplified
* Fixed spots where support for unknown and never types was missing
* (NEW) Length operator '#' is now valid to use on top table type, this
type comes up when doing typeof(x) == "table" guards and isn't available
in current typechecker
---
Changes to native code generation:
* Additional math library fast calls are now lowered to x64: math.ldexp,
math.round, math.frexp, math.modf, math.sign and math.clamp
* For autocomplete, additional information is included in Scope for type
alias name locations and names of imported modules
* Improved autocomplete suggestions in 'for' and 'while' loop headers
* String match functions return types are now optional strings and
numbers because match is not guaranteed at runtime
* Fixed build issue on gcc 11 and up (Fixes
https://github.com/Roblox/luau/issues/806)
* Fix autocompletion of if-then-else expressions
* Fix a potential crash surrounding improper use of `%*` in a string
format specifier
* All Python scripts now invoke Python via `python3` rather than
`python`.
* Improved error handling for string interpolation with too many
arguments.
Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com>
Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
* Fixed garbage data in module scopes when type graph is not retained
* LOP_MOVE with the same source and target registers is no longer
generated (Fixes https://github.com/Roblox/luau/issues/793)
* Fixed unions of `nil` types displaying as `?`
* Internal normalization now handles class types which can make
previously failing (incorrectly) sub-typing checks to succeed
* The AST JSON encoder will now stringify infinity and NaN according to the JSON5 spec using the tokens `Infinity`, `-Infinity`, and `NaN`.
* Improve autocompletion of table keys if the type of that key is a union of string singletons.
* Type mismatch errors now mention if unification failed in covariant or
invariant context, to explain why sometimes derived class can't be
converted to base class or why `T` can't be converted into `T?` and so
on
* Class type indexing is no longer an error in non-strict mode (still an
error in strict mode)
* Fixed cyclic type packs not being displayed in the type
* Added an error when unrelated types are compared with `==`/`~=`
* Fixed false positive errors involving sub-type tests an `never` type
* Fixed miscompilation of multiple assignment statements (Fixes
https://github.com/Roblox/luau/issues/754)
* Type inference stability improvements
* Type inference of `a and b` and `a or b` has been improved (Fixes
https://github.com/Roblox/luau/issues/730)
* Improved error message when `for ... in x` loop iterates over a value
that could be 'nil'
* Return type of `next` not includes 'nil' (Fixes
https://github.com/Roblox/luau/issues/706)
* Improved type inference of `string` type
* Luau library table type names are now reported as `typeof(string)`/etc
instead of just `string` which was misleading
* Added parsing error when optional type alias type parameter wasn't
provided after `=` token
* Improved tagged union type refinement in conditional expressions, type
in `else` branch should no longer include previously handled union
options
* Reduce the stack utilization of type checking.
* Improve the error message that's reported when a delimiting comma is
missing from a table literal. eg
```lua
local t = {
first = 1
second = 2
}```
* https://github.com/Roblox/luau/pull/719
* Improved `Failed to unify type packs` error message to be reported as
`Type pack 'X' could not be converted into 'Y'`
* https://github.com/Roblox/luau/pull/722
* 1% reduction in executed instruction count by removing a check in fast
call dispatch
* Additional fixes to reported error location of OOM errors in VM
* Improve `math.sqrt`, `math.floor` and `math.ceil` performance on
additional compilers and platforms (1-2% geomean improvement including
8-9% on math-cordic)
* All thrown exceptions by Luau analysis are derived from
`Luau::InternalCompilerError`
* When a call site has fewer arguments than required, error now reports
the location of the function name instead of the argument to the
function
* https://github.com/Roblox/luau/pull/724
* Fixed https://github.com/Roblox/luau/issues/725
Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com>
Co-authored-by: Andy Friesen <afriesen@roblox.com>
* Support `["prop"]` syntax on class definitions in definition files.
(#704)
* Improve type checking performance for complex overloaded functions
* Fix rare cases of incorrect stack traces for out of memory errors at
runtime
* Reoptimized math.min/max/bit32 builtins assuming at least 2 arguments are used (1-2% lift on some benchmarks)
* Type errors that mention function types no longer have redundant parenthesis around return type
* Luau REPL now supports --compile=remarks which displays the source code with optimization remarks embedded as comments
* Builtin calls are slightly faster when called with 1-2 arguments (~1% improvement in some benchmarks)
- Fix rare type checking bugs with invalid generic types escaping the
module scope
- Fix type checking of variadic type packs in certain cases
- Implement type normalization, which resolves a large set of various
issues with unions/intersections in type checker
- Improve parse errors for trailing commas in function calls and type
lists
- Reduce profiling skew when using --profile with very high frequencies
- Improve performance of `lua_getinfo` (`debug.info`, `debug.traceback`
and profiling overhead are now 20% faster/smaller)
- Improve performance of polymorphic comparisons (1-2% lift on some
benchmarks)
- Improve performance of closure creation (1-2% lift on some benchmarks)
- Improve string comparison performance (4% lift on string sorting)
- Type aliases can no longer override primitive types; attempts to do
that will result in a type error
- Fix misleading type error messages for mismatches in expression list
length during assignment
- Fix incorrect type name display in certain cases
- setmetatable/getmetatable are now ~2x faster
- tools/perfstat.py can be used to display statistics about profiles
captured via --profile switch
- Improve type error messages for argument count mismatch in certain
cases
- Fix type checking variadic returns when the type is incompatible which
type checked in certain cases
- Reduce size of upvalue objects by 8 bytes on 64-bit platforms
- Reduce I$ footprint of interpreter by 1.5KB
- Reduce GC pause during atomic stage for programs with a lot of threads
- Remove support for bytecode v2
- Remove type definitions of
`utf8.nfcnormalize`/`nfdnormalize`/`graphemes` that aren't supported by
standalone Luau library
- Add `lua_costatus` to retrieve extended thread status (similar to
`coroutine.status`)
- Improve GC sweeping performance (2-10% improvement on allocation-heavy
benchmarks)
- Improve ComparisonPrecedence lint suggestions for three-way comparisons (X < Y < Z)
- Improve type checking stability
- Improve location information for errors when parsing invalid type annotations
- Compiler now generates bytecode version 3 in all configurations
- Improve performance of comparisons against numeric constants on AArch64
- Fix DeprecatedGlobal warning text in cases when the global is deprecated without a suggested alternative
- Fix an off-by-one error in type error text for incorrect use of string.format
- Reduce stack consumption further during parsing, hopefully eliminating stack overflows during parsing/compilation for good
- Mark interpolated string support as experimental (requires --fflags=LuauInterpolatedStringBaseSupport to enable)
- Simplify garbage collection treatment of upvalues, reducing cache misses during sweeping stage and reducing the cost of upvalue assignment (SETUPVAL); supersedes #643
- Simplify garbage collection treatment of sleeping threads
- Simplify sweeping of alive threads, reducing cache misses during sweeping stage
- Simplify management of string buffers, removing redundant linked list operations
- Fix autocomplete not suggesting globals defined after the cursor (fixes#622)
- Improve type checker stability
- Reduce parser C stack consumption which fixes some stack overflow crashes on deeply nested sources
- Improve performance of bit32.extract/replace when width is implied (~3% faster chess)
- Improve performance of bit32.extract when field/width are constants (~10% faster base64)
- Heap dump now annotates thread stacks with local variable/function names
* Run clang-format
* Contains a preliminary implementation of deferred constraint resolution
* Reduce stack usage by some recursive functions
* Fix a bug when smartCloning a BoundTypeVar
* Remove some GC related flags from VM
* Adds a currently unused x86-64 assembler as a prerequisite for possible future JIT compilation
* Fix a bug in table iteration (closes Possible table iteration bug #504)
* Improved warning method when function is used as a type
* Fix a bug with unsandboxed iteration with pairs()
* Type of coroutine.status() is now a union of value types
* Bytecode output for tests/debugging now has labels
* Improvements to loop unrolling cost estimation
* Report errors when the key obviously doesn't exist in the table
- Rework transaction log used for type checking which should result in more robust type checking internals with fewer bugs
- Reduce the amount of memory consumed by type checker on large module graphs
- Type checker now errors on attempts to change the type of imported module fields
- The return type of newproxy is now any (fixes#296)
- Implement new number printing algorithm (Schubfach) which makes tostring() produce precise (round-trippable) and short decimal output up to 10x faster
- Fix lua_Debug::linedefined to point to the line with the function definition instead of the first statement (fixes#265)
- Fix minor bugs in Tab completion in Repl
- Repl now saves/restores command history in ~/.luau_history
- Fix some cases where type checking would overflow the native stack
- Improve autocomplete behavior when assigning a partially written function call (not currently exposed through command line tools)
- Improve autocomplete type inference feedback for some expressions where previously the type would not be known
- Improve quantification performance during type checking for large types
- Improve type checking for table literals when the expected type of the table is known because of a type annotation
- Fix type checking errors in cases where required module has errors in the resulting type
- Fix debug line information for multi-line chained call sequences (Add function name information for "attempt to call a nil value" #255)
- lua_newuserdata now takes 2 arguments to match Lua/LuaJIT APIs better; lua_newuserdatatagged should be used if the third argument was non-0.
- lua_ref can no longer be used with LUA_REGISTRYINDEX to prevent mistakes when migrating Lua FFI (Inconsistency with lua_ref #247)
- Fix assertions and possible crashes when executing script code indirectly via metatable dispatch from lua_equal/lua_lessthan/lua_getfield/etc. (Hitting a crash in an assert after lua_equal is called. #259)
- Fix flamegraph scripts to run under Python 2
- A series of major optimizations to type checking performance on complex
programs/types (up to two orders of magnitude speedup for programs
involving huge tagged unions)
- Fix a few issues encountered by UBSAN (and maybe fix s390x builds)
- Fix gcc-11 test builds
- Fix a rare corner case where luau_load wouldn't wake inactive threads
which could result in a use-after-free due to GC
- Fix CLI crash when error object that's not a string escapes to top level
- Fix Makefile suffixes on macOS
Co-authored-by: Rodactor <rodactor@roblox.com>
Changes:
- Support for time tracing for analysis/compiler (not currently exposed
through CLI)
- Support for type pack arguments in type aliases (#83)
- Basic support for require(path) in luau-analyze
- Add a lint warning for table.move with 0 index as part of
TableOperation lint
- Remove last STL dependency from Luau.VM
- Minor VS2022 performance tuning
Co-authored-by: Rodactor <rodactor@roblox.com>
