How many recursive calls to findMember are performed before logging names starts Note: this threshold has to be chosen carefully. Too large, and programs like tests/pos/IterableSelfRec go into polynomial (or even exponential?) compile time slowdown. Too small and normal programs will cause the compiler to do inefficient operations on findMember. The current value is determined so that (1) IterableSelfRec still compiles in reasonable time (< 10sec) (2) Compiling dotty itself only causes small pending names lists to be generated (we measured at max 6 elements) and these lists are never searched with contains.
How many recursive calls to isSubType are performed before logging starts.
How many recursive calls to NamedType#underlying are performed before logging starts.
If this flag is set, and we compute
T2[X2] as a new
upper bound of a constrained parameter, try to align the arguments by computing
S1 =:= S2 (which might instantiate type parameters).
This rule is contentious because it cuts the constraint set.
For more info, see the comment in
Check that Name#toString is not called directly from backend by analyzing the stack trace of each toString call on names. This is very expensive, so not suitable for continuous testing. But it can be used to find a problem when running a specific test.
When updating a constraint bound, check that the constrained parameter does not appear at the top-level of either of its bounds.
Check that each constraint is fully propagated. i.e. If P <: Q then the upper bound of P is a subtype of the upper bound of Q and the lower bound of Q is a subtype of the lower bound of P.
Check that each constraint resulting from a subtype test is satisfiable.
Make sure none of the bounds of a parameter in an OrderingConstraint contains this parameter at its toplevel (i.e. as an operand of a combination of &'s and |'s.). The check is performed each time a new bound is added to the constraint.
If this flag is set, higher-kinded applications are checked for validity
Check that variances of lambda arguments match the variance of the underlying lambda class.
If this flag is set, method types are checked for valid parameter references
Enable noDoubleDef checking if option "-YnoDoubleDefs" is set. The reason to have an option as well as the present global switch is that the noDoubleDef checking is done in a hotspot, and we do not want to incur the overhead of checking an option each time.
Check that no type appearing as the info of a SymDenotation contains skolem types.
Check positions for consistency after parsing
Check that typed trees don't point to untyped ones
If this flag is set, it is checked that class type parameters are only references with NoPrefix or ThisTypes as prefixes. This option is usally disabled, because there are still some legitimate cases where this can arise (e.g. for pos/Map.scala, in LambdaType.integrate).
If this flag is set, it is checked that TypeRefs don't refer directly to themselves.
Check that certain types cannot be created in erasedTypes phases. Note: Turning this option on will get some false negatives, since it is possible that And/Or types are still created during erasure as the result of some operation on an existing type.
If this flag is set, we check that types assigned to trees are error types only
if some error was already reported. There are complicicated scenarios where this
is not true. An example is TestNonCyclic in posTwice. If we remove the
first (unused) import
import dotty.tools.dotc.core.Types.Type in
we end up assigning a CyclicReference error type to an import expression
before the cyclic reference is reported. What happens is that the error was reported
as a result of a completion in a not-yet committed typerstate. So we cannot enforce
this in all circumstances. But since it is almost always true it is useful to
keep the Config option for debugging.
Check that constraints of globally committable typer states are closed. NOTE: When enabled, the check can cause CyclicReference errors because it traverses all elements of a type. Such failures were observed when compiling all of dotty together (source seems to be in GenBCode which accesses javac's settings.)
It is recommended to turn this option on only when chasing down a TypeParamRef instantiation error. See comment in Types.TypeVar.instantiate.
Type comparer will fail with an assert if the upper bound of a constrained parameter becomes Nothing. This should be turned on only for specific debugging as normally instantiation to Nothing is not an error consdition.
If this flag is set, take the fast path when comparing same-named type-aliases and types
When in IDE, turn StaleSymbol errors into warnings instead of crashing
Initial capacity of uniques HashMap. Note: This MUST BE a power of two to work with util.HashSet
If set, prints a trace of all symbol completions
If this flag is on, always rewrite an application
S is an alias for
[Xs] -> U to
[Xs := Ts]U.
Turning this flag on was observed to give a ~6% speedup on the JUnit test suite.
(S & T)#A --> S#A & T#A (S | T)#A --> S#A | T#A
Not sure whether this is useful. Preliminary measurements show a slowdown of about 7% for the build when this option is enabled.
The recursion depth for showing a summarized string
Show subtype traces for all deep subtype recursions
If set, enables tracing
When explaining subtypes and this flag is set, also show the classes of the compared types.