Typechecker Improvements

• Commit by mheiber: correct handling of parent:: and self This applies to the method attribute <<__NeedsConcrete>>. If you mark a method on an abstract type with this attribute, all calls to this method must originate from a non-abstract type. This is useful when you need to call other methods your declared as abstract on yourself. This change makes some false positive errors go away.
• Commit by Michael Thomas : Fix bug which disallows all wildcard hints in poly lambda It is now possible to use $x is vec<_> in a lambda with a generic. You can write generic lambdas like so
  $refl = function<T>(T $t): T { return $t; }; This is new in hhvm@next. HHVM tried to prevent you from writing taking a parameter of type vec<_>, since an underscore in a generic parameter does not make sense. This check accidentally banned the use of underscored generics everywhere in the lambda, including the body, where as and is expressions can be used.
• Commit by mheiber: reified check: be more careful leaking type params through this You are not allowed to reify a type parameter with an erased inner generic. You are sometimes allowed to reify the this type. This commit adds a check to make sure that the this type you are reifying does not include an erased type parameter. This code now raises a type error.
• Commit by vassilmladenov: Kill the old class_pointer_levels flags, simplify errors The following new typechecker options have been introduced: class_pointer_ban_classname_new, class_pointer_ban_classname_type_structure, class_pointer_ban_classname_static_prop, class_pointer_ban_classname_static_meth, class_pointer_ban_classname_class_const, class_pointer_ban_class_array_key
  It is now an error to use a classname<T> in places where a string is expected. You will see this typechecker error: Using `SomeType::class` in this position will trigger an implicit runtime conversion to string, please use `nameof SomeType`. nameof is not valid in the HHVM 4 series.
• Commit by viratyosin: Add error for recursive dependencies in where clauses of case types This only affects case type X = Y, not the classic type X = Y or newtype X = Y type aliases. The typechecker will now report an error if a where clause depends on its own type. This code now reports an error.
• Commit by Mistral Contrastin: Warn on calling static methods on traits The typechecker will now issue warnings when calling a private static method on the name of a trait from within the body of the trait. This will promote to a typechecker error in a future release.
• Commit by Mistral Contrastin: Ban accessing static properties on traits Similarly to the methods case above, but this is a hard typechecker error, rather than a warning.
• Commit by Mistral Contrastin : Warn against static property redeclaration The typechecker will now warn on static property declarations which repeat a static property from a base class. These properties were not harmless redeclarations but actually created a second property with the same name. This led to strange behavior where methods would operate on the static property of the superclass or the subclass depending on the late static bound type.

Highlights

• Commit by vassilmladenov: Model class<T> like classname<T> in xhp attributes You are now allowed to use class<SomeType> and classname<SomeType> as xhp attributes.
• Commit by vassilmladenov: Make serialize builtin use lazy class flavors for KindOf[Lazy]Class Values of the LazyClass and Class type now roundtrip correctly through \serialize(). Serializations made on previous versions of hhvm will continue to deserialize as they did before. If you serialize on HHVM 25.7.0 and try to deserialize on 25.6.0, this may result in unexpected behavior.
• Commit by vassilmladenov: Make APC and internal serialize use lazy classes for class pointers Similarly to the commit above, but now for \apc_* functions.
• Commit by jtwarren: Fix refcount bug in PBWH Fixes a missing incref in PriorityBridgeWaitHandle. This type is used under the hood of HH\Lib\Async\LowPri.
• Commit by Bin Liu: Allow LowPri<void> objects It is now possible to use a HH\Lib\Async\LowPri<void> object at runtime. This was already allowed by the typechecker, but failed with an internal as nonnull type assertion.
• Commit by jano: Reduce is_object() and is_scalar() to IsTypeC The JIT will now replace direct calls to \is_object() and \is_scalar() with bytecodes. This is similar to how the function call to \strlen() is optimized away. Owen Yamauchi did write a blog post on this in 2012.
• Commit by geralt-encore: Move ProtectedInternal experimental feature to preview The unstable feature protected internal has been promoted to Preview. The HIP for this feature is probably still internal to Meta.
• Commit by Andrew Kennedy : Move OpenTuples and TypeSplat to OngoingRelease The experimental features 'open_tuples' and 'type_splat' have been moved from Preview to OngoingRelease.
• Commit by mszabo-wikia: Build the mcrouter extension The ext_mcrouter extension, which provides the MCRouter class is now available again.
• Commit by mszabo-wikia: Consistently check connection validity in AsyncMysqlConnection Calling toString__FOR_DEBUGGING_ONLY on an HH\Lib\Query object when the database connection is busy on a query no longer segfaults hhvm. Instead, an InvalidOperationException is raised. This behavior matches performing a query on a busy database connection.
• Commit by Bin Liu: Use NUMA API to optimize for NUCA architectures A low-level performance optimization. This is disabled in Docker images, because the HHVM_DISABLE_NUMA environment variable is set.

Breaking Changes

• Commit by Niharika Devanathan: Add a runtime check that a non-async method cannot override an async method The runtime now enforces the belief of the typechecker that public async function fooAsync(): Awaitable<int> may not be overriden by a method without the async keyword. The reason being, the inner type of the awaitable is enforced at runtime if your method is an async method. Whereas an overriding non-async method returning an awaitable will not enforce its inner type. This meant that code that suppressed Hack errors could return an awaitable with a string inside instead of an int.
• Commit by vassilmladenov: Expand function pointer check, unify with special_function emitter, refactor a bit You are no longer allowed to create a function reference, using the function_name<> syntax to some special built-in functions: HH\classname_to_class, exit, HH\ReifiedGenerics\get_class_from_type, HH\global_get, HH\global_set, HH\global_unset, HH\tag_provenance_here, HH\FIXME\UNSAFE_CAST, HH\FIXME\UNSAFE_NONNULL_CAST
• Commit by vassilmladenov: Update Facts sub/supertypes HHI The functions HH\Facts\subtypes, HH\Facts\transitive_subtypes, and HH\Facts\supertypes now return class<T> instead of string.

Changes to Experimental Features

• Commit by Andrew Kennedy: Do not reject method overriding method with type-splat parameter This affects users using the experimental features: 'type_splat', 'open_tuples'. HHVM now correctly allows the (int $a, string $b) parameter list to override a (...TupleIntString $args) parameter list. This code is now accepted where it previously wasn't.
• Commit by dlreeves: Change How We Virtualize Operation Typing in Expression Trees This affects users using the experimental features: 'expression_trees'. The public static function operationType() method is now required on DSL classes.
• Commit by potomak: Remove nullType This affects users using the experimental features: 'expression_trees'. The public static function nullType() method is now ignored. null will be Hack's null in your DSL.