Sync to upstream/release/666 (#1747)

Another week, another release. Happy spring! 🌷 

## New Type Solver

- Add typechecking and autocomplete support for user-defined type
functions!
- Improve the display of type paths, making type mismatch errors far
more human-readable.
- Enhance various aspects of the `index` type function: support function
type metamethods, fix crashes involving cyclic metatables, and forward
`any` types through the type function.
- Fix incorrect subtyping results involving the `buffer` type.
- Fix crashes related to typechecking anonymous functions in nonstrict
mode.

## AST

- Retain source information for type packs, functions, and type
functions.
- Introduce `AstTypeOptional` to differentiate `T?` from `T | nil` in
the AST.
- Prevent the transpiler from advancing before tokens when the AST has
errors.

## Autocomplete

- Introduce demand-based cloning and better module isolation for
fragment autocomplete, leading to a substantial speedup in performance.
- Guard against recursive unions in `autocompleteProps`.

## Miscellaneous

- #1720 (thank you!)

## Internal Contributors

Co-authored-by: Andy Friesen <[email protected]>
Co-authored-by: Ariel Weiss <[email protected]>
Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Talha Pathan <[email protected]>
Co-authored-by: Vighnesh Vijay <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>

Sync to upstream/release/665 (#1732)

Hello all! Another week, another Luau release!

# Change to `lua_setuserdatametatable`

This release fixes #1710: `lua_setuserdatametatable` is being changed so
that it _only_ operates on the top of the stack: the `idx` parameter is
being removed. Prior to this, `lua_setuserdatametable` would set the
metatable of the value in the stack at `idx`, but _always_ pop the top
of the stack. The old behavior is available in this release as
`lua_setuserdatametatable_DEPRECATED`.

# General

This release exposes a generalized implementation of require-by-string's
autocomplete logic. `FileResolver` can now be optionally constructed
with a `RequireSuggester`, which provides an interface for converting a
given module to a `RequireNode`. Consumers of this new API implement a
`RequireNode` to define how modules are represented in their embedded
context, and the new API manages the logic specific to
require-by-string, including providing suggestions for require aliases.
This enhancement moves toward integrating require-by-string's semantics
into the language itself, rather than merely providing a specification
for community members to implement themselves.

# New Type Solver
* Fixed a source of potential `Luau::follow detected a Type cycle`
internal compiler exceptions when assigning a global to itself.
* Fixed an issue whereby `*no-refine*` (a type which should not be
visible at the end of type checking) was not being properly elided,
causing inference of class-like tables to become unreadable / induce
crashes in autocomplete.
* Fixed a case of incomplete constraint solving when performing basic
math in a loop

# Fragment Autocomplete
* Fixed several crashes related to not properly filling in scope
information for the fragments
* Fixed a source of memory corruption by isolating the return type of a
fragment when it is type checked.
* Improved performance by opting not to clone persistent types for the
fragment (e.g.: built in types)
 
# Internal Contributors
Co-authored-by: Andy Friesen <[email protected]>
Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Talha Pathan <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Vighnesh Vijay <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>

---------

Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Alexander Youngblood <[email protected]>
Co-authored-by: Menarul Alam <[email protected]>
Co-authored-by: Aviral Goel <[email protected]>
Co-authored-by: Vighnesh <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>
Co-authored-by: Ariel Weiss <[email protected]>

Sync to upstream/release/664 (#1715)

As always, a weekly Luau update!
This week we have further improvements to new type solver, fixing a few
of the popular issues reported. The fragment autocomplete is even more
stable and we believe it's ready for broader use.

Aside from that we have a few general fixes/improvements:
* Fixed data race when multi-threaded typechecking is used, appearing as
a random crash at the end of typechecking
* AST data is now available from `Luau::Module`

## New Type Solver

* Fixed type refinements made by function calls which could attach `nil`
as an option of a type before (Fixes #1528)
* Improved bidirectional typechecking in tables (Fixes #1596)
* Fixed normalization of negated types
* `getmetatable()` on `any` type should no longer report an error

## Fragment Autocomplete

* Fixed auto-complete suggestions being provided inside multiline
comments
* Fixed an assertion failure that could happen when old type solver was
used
* Fixed issues with missing suggestions when multiple statements are on
the same line
* Fixed memory safety issues

## Internal Contributors

Co-authored-by: Andy Friesen <[email protected]>
Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Vighnesh Vijay <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>

Sync to upstream/release/663 (#1699)

Hey folks, another week means another Luau release! This one features a
number of bug fixes in the New Type Solver including improvements to
user-defined type functions and a bunch of work to untangle some of the
outstanding issues we've been seeing with constraint solving not
completing in real world use. We're also continuing to make progress on
crashes and other problems that affect the stability of fragment
autocomplete, as we work towards delivering consistent, low-latency
autocomplete for any editor environment.

## New Type Solver

- Fix a bug in user-defined type functions where `print` would
incorrectly insert `\1` a number of times.
- Fix a bug where attempting to refine an optional generic with a type
test will cause a false positive type error (fixes #1666)
- Fix a bug where the `refine` type family would not skip over
`*no-refine*` discriminants (partial resolution for #1424)
- Fix a constraint solving bug where recursive function calls would
consistently produce cyclic constraints leading to incomplete or
inaccurate type inference.
- Implement `readparent` and `writeparent` for class types in
user-defined type functions, replacing the incorrectly included `parent`
method.
- Add initial groundwork (under a debug flag) for eager free type
generalization, moving us towards further improvements to constraint
solving incomplete errors.

## Fragment Autocomplete

- Ease up some assertions to improve stability of mixed-mode use of the
two type solvers (i.e. using Fragment Autocomplete on a type graph
originally produced by the old type solver)
- Resolve a bug with type compatibility checks causing internal compiler
errors in autocomplete.

## Lexer and Parser

- Improve the accuracy of the roundtrippable AST parsing mode by
correctly placing closing parentheses on type groupings.
- Add a getter for `offset` in the Lexer by @aduermael in #1688
- Add a second entry point to the parser to parse an expression,
`parseExpr`

## Internal Contributors

Co-authored-by: Andy Friesen <[email protected]>
Co-authored-by: Ariel Weiss <[email protected]>
Co-authored-by: Aviral Goel <[email protected]>
Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: James McNellis <[email protected]>
Co-authored-by: Talha Pathan <[email protected]>
Co-authored-by: Vighnesh Vijay <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>

---------

Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Alexander Youngblood <[email protected]>
Co-authored-by: Menarul Alam <[email protected]>
Co-authored-by: Aviral Goel <[email protected]>
Co-authored-by: Vighnesh <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>

Sync to upstream/release/662 (#1681)

## What's new

This update brings improvements to the new type solver, roundtrippable
AST parsing mode and closes multiple issues reported in this repository.

* `require` dependency tracing for non-string requires now supports `()`
groups in expressions and types as well as an ability to type annotate a
value with a `typeof` of a different module path
* Fixed rare misaligned memory access in Compiler/Typechecker on 32 bit
platforms (Closes #1572)

## New Solver

* Fixed crash/UB in subtyping of type packs (Closes #1449)
* Fixed incorrect type errors when calling `debug.info` (Closes #1534
and Resolves #966)
* Fixed incorrect boolean and string equality comparison result in
user-defined type functions (Closes #1623)
* Fixed incorrect class types being produced in user-defined type
functions when multiple classes share the same name (Closes #1639)
* Improved bidirectional typechecking for table literals containing
elements that have not been solved yet (Closes #1641)

## Roundtrippable AST

* Added source information for `AstStatTypeAlias`
* Fixed an issue with `AstTypeGroup` node (added in #1643) producing
invalid AST json. Contained type is now named 'inner' instead of 'type'
* Fixed end location of the `do ... end` statement

---

Internal Contributors:

Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Talha Pathan <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Vighnesh Vijay <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>

Sync to upstream/release/661 (#1664)

# General
- Additional logging enabled for fragment autocomplete.

## Roundtrippable AST
- Add a new `AstNode`, `AstGenericType`
- Retain source information for `AstExprTypeAssertion`
## New Type Solver
- New non-strict mode will report unknown symbol errors, e.g
```
foo = 5
local wrong1 = foob <- issue warning
```
- Fixed a bug where new non-strict mode failed to visit large parts of
the program.
- We now infer the types of unnanotated local variables in statements
with multiple assignments, e.g. `local x: "a", y, z = "a", f()`
- Fixed bugs in constraint dispatch ordering.
- Fixed a bug that caused an infinite loop between `Subtyping`,
`OverloadResolution`, and `Type Function Reduction`, by preventing calls
to `Type Function Reduction` being re-entrant.
- Fixed a crash in bidirectional type inference caused by asserting read
and write properties on a type that was readonly.

## Runtime
- Fix a stack overflow caused by `luaL_checkstack` consuming stack space
even if the function fails to reserve memory.
- Using '%c' with a 0 value in Luau string.format will append a '\0'.
Resolves #1650

## Miscellaneous
- Miscellaneous small bugfixes for the new solver.

**Full Changelog**:
0.660...0.661
----
Co-authored-by: Ariel Weiss <[email protected]>
Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Talha Pathan <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Vighnesh Vijay <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>

---------

Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Alexander Youngblood <[email protected]>
Co-authored-by: Menarul Alam <[email protected]>
Co-authored-by: Aviral Goel <[email protected]>

Sync to upstream/release/660 (#1643)

# General

This release introduces initial work on a Roundtrippable AST for Luau,
and numerous fixes to the new type solver, runtime, and fragment
autocomplete.

## Roundtrippable AST

To support tooling around source code transformations, we are extending
the parser to retain source information so that we can re-emit the
initial source code exactly as the author wrote it. We have made
numerous changes to the Transpiler, added new AST types such as
`AstTypeGroup`, and added source information to AST nodes such as
`AstExprInterpString`, `AstExprIfElse`, `AstTypeTable`,
`AstTypeReference`, `AstTypeSingletonString`, and `AstTypeTypeof`.

## New Type Solver

* Implement `setmetatable` and `getmetatable` type functions.
* Fix handling of nested and recursive union type functions to prevent
the solver from getting stuck.
* Free types in both old and new solver now have an upper and lower
bound to resolve mixed mode usage of the solvers in fragment
autocomplete.
* Fix infinite recursion during normalization of cyclic tables.
* Add normalization support for intersections of subclasses with negated
superclasses.

## Runtime
* Fix compilation error in Luau buffer bit operations for big-endian
machines.

## Miscellaneous
* Add test and bugfixes to fragment autocomplete.
* Fixed `clang-tidy` warnings in `Simplify.cpp`.

**Full Changelog**:
0.659...0.660

---

Co-authored-by: Ariel Weiss <[email protected]>
Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Talha Pathan <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Vighnesh Vijay <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>

---------

Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Alexander Youngblood <[email protected]>
Co-authored-by: Menarul Alam <[email protected]>

Sync to upstream/release/659 (#1637)

## What's Changed

General performance improvements and bug fixes. `lua_clonetable` was
added too.

### General

## Runtime
- Improvements were made to Luau's performance, including a
`lua_clonetable` function and optimizations to string caching. Buffer
read/write operations were optimized for big-endian machines.
## New Solver
- Crashes related to duplicate keys in table literals, fragment AC
crashes, and potential hash collisions in the StringCache.
- We now handle user-defined type functions as opaque and track interior
free table types.
## Require By String
- Require-by-string path resolution was simplified.

**Full Changelog**:
0.658...0.659

---

Co-authored-by: Ariel Weiss <[email protected]>
Co-authored-by: Andy Friesen <[email protected]>
Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Vighnesh Vijay <[email protected]>
Co-authored-by: Yohoo Lin <[email protected]>

---------

Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Alexander Youngblood <[email protected]>

Sync to upstream/release/658 (#1625)

## What's Changed

### General
- Allow types of tables to diverge after using `table.clone` (fixes
#1617).
- Allow 2-argument vector.create in Luau.
- Fix a crash when suggesting autocomplete after encountering parsing
errors.
- Add lua_tolstringatom C API which returns the string length (whether
or not the atom exists) and which extends the existing lua_tostringatom
function the same way lua_tolstring/lua_tostring do.
- Luau now retains the DFGs of typechecked modules.

### Magic Functions Migration Note
We've made a change to the API used to define magic functions.

Previously, we had a set of function pointers on each `FunctionType`
that would be invoked by the type inference engine at the correct point.

The problem we'd run into is that they were all `std::function`s, we'd
grown quite a few of them, and Luau allocates tens of thousands of types
as it performs type inference. This adds up to a large amount of memory
for data that isn't used by 99% of types.

To slim things down a bit, we've replaced all of those `std::function`s
with a single `shared_ptr` to a new interface called `MagicFunction`.
This slims down the memory footprint of each type by about 50 bytes.

The virtual methods of `MagicFunction` have roughly 1:1 correspondence
with the old interface, so updating things should not be too difficult:

* `FunctionType::magicFunction` is now `MagicFunction::handleOldSolver`
* `FunctionType::dcrMagicFunction` is now `MagicFunction::infer`
* `FunctionType::dcrMagicRefinement` is now `MagicFunction::refine`
* `FunctionType::dcrMagicTypeCheck` is now `MagicFunction::typeCheck`

**Full Changelog**:
0.657...0.658

---

Co-authored-by: Andy Friesen <[email protected]>
Co-authored-by: Ariel Weiss <[email protected]>
Co-authored-by: Aviral Goel <[email protected]>
Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Talha Pathan <[email protected]>
Co-authored-by: Varun Saini <[email protected]>
Co-authored-by: Vighnesh Vijay <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>

Sync to upstream/release/657 (#1619)

## General
- Fix a parsing bug related to the starting position of function names.
- Rename Luau's `Table` struct to `LuaTable`.

## New Solver
- Add support for generics in user-defined type functions
([RFC](https://rfcs.luau.org/support-for-generic-function-types-in-user-defined-type-functions.html)).
- Provide a definition of `math.lerp` to the typechecker.
- Implement error suppression in `string.format`.
  - Fixes #1587.
- Ensure function call discriminant types are always filled when
resolving `FunctionCallConstraint`.

---

Co-authored-by: Ariel Weiss <[email protected]>
Co-authored-by: Hunter Goldstein <[email protected]>
Co-authored-by: Talha Pathan <[email protected]>
Co-authored-by: Vyacheslav Egorov <[email protected]>