| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
| |
The underlying transformer has a by-name parameter for the
to provide the `to` tree, but this was strict in the layers
of API above.
Tree sharing is frowned upon in general as it leads to cross
talk when, e.g., the erasure typechecker mutates the
`tpe` field of the shared tree in different context.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Until now, concrete methods in traits were encoded with
"trait implementation classes".
- Such a trait would compile to two class files
- the trait interface, a Java interface, and
- the implementation class, containing "trait implementation methods"
- trait implementation methods are static methods has an explicit self
parameter.
- some methods don't require addition of an interface method, such as
private methods. Calls to these directly call the implementation method
- classes that mixin a trait install "trait forwarders", which implement
the abstract method in the interface by forwarding to the trait
implementation method.
The new encoding:
- no longer emits trait implementation classes or trait implementation
methods.
- instead, concrete methods are simply retained in the interface, as JVM 8
default interface methods (the JVM spec changes in
[JSR-335](http://download.oracle.com/otndocs/jcp/lambda-0_9_3-fr-eval-spec/index.html)
pave the way)
- use `invokespecial` to call private or particular super implementations
of a method (rather `invokestatic`)
- in cases when we `invokespecial` to a method in an indirect ancestor, we add
that ancestor redundantly as a direct parent. We are investigating alternatives
approaches here.
- we still emit trait fowrarders, although we are
[investigating](https://github.com/scala/scala-dev/issues/98) ways to only do
this when the JVM would be unable to resolve the correct method using its rules
for default method resolution.
Here's an example:
```
trait T {
println("T")
def m1 = m2
private def m2 = "m2"
}
trait U extends T {
println("T")
override def m1 = super[T].m1
}
class C extends U {
println("C")
def test = m1
}
```
The old and new encodings are displayed and diffed here: https://gist.github.com/retronym/f174d23f859f0e053580
Some notes in the implementation:
- No need to filter members from class decls at all in AddInterfaces
(although we do have to trigger side effecting info transformers)
- We can now emit an EnclosingMethod attribute for classes nested
in private trait methods
- Created a factory method for an AST shape that is used in
a number of places to symbolically bind to a particular
super method without needed to specify the qualifier of
the `Super` tree (which is too limiting, as it only allows
you to refer to direct parents.)
- I also found a similar tree shape created in Delambdafy,
that is better expressed with an existing tree creation
factory method, mkSuperInit.
|
|
|
|
|
|
|
|
|
|
| |
- Language imports are preceding other imports
- Deleted empty file: InlineErasure
- Removed some unused private[parallel] methods in
scala/collection/parallel/package.scala
This removes hundreds of warnings when compiling with
"-Xlint -Ywarn-dead-code -Ywarn-unused -Ywarn-unused-import".
|
|
|
|
|
| |
Keep owner for module (symbol of the tree) and module class (holds the members)
in synch while moving trees between owners (e.g., while duplicating them in specialization)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Typechecking a pattern that defines a pattern type variable
initially assigns abstract type symbol with open type bounds.
Later on, pattern type inference kicks in to sharpen the type
of the variable based on constraints imposed by the expected
type (ie, the type of scrutinee of the pattern.)
However, before inference does this, a `TypeRef` to the abstract
type symbol can be queried for its base type with respect to some
class, which leads to it populating an internal cache. This cache
becomes stale when the underlying symbol has its type mutated.
The repercussions of this meant that a subsequent call to `baseType`
gave the wrong result (`NoType`), which lead to an `asSeenFrom`
operation to miss out of substitution of a type variable. Note the
appearance of `A` in the old type errors in the enclosed test case.
This commit takes an approach similar to 286dafbd to invalidate
caches after the mutation. I've routed both bandaids through the
same first aid kit: I'm sure over time we'll add additional calls
to this method, and additional cache invalidations within it.
|
|
|
|
|
|
|
|
| |
only trivial merge conflicts here.
not dealing with PR #4333 in this merge because there is a substantial
conflict there -- so that's why I stopped at
63daba33ae99471175e9d7b20792324615f5999b for now
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Added `since` to deprecation statement
- Added unit to parameter list
- Removed usage of deprecated method polyType
- Replaced deprecated `debugwarn` with `devWarning`
- Changed switch statement to if else in order to remove a warning
- Switched implementation of `init` and `processOptions` to prevent
warning
- Replaced deprecated `Console.readLine` with `scala.io.StdIn.readLine`
- Replaced deprecated `startOrPoint` with `start`
- Replaced deprecated `tpe_=` with `setType`
- Replaced deprecated `typeCheck` with `typecheck`
- Replaced deprecated `CompilationUnit.warning` with `typer.context.warning`
- Replaced deprecated `scala.tools.nsc.util.ScalaClassLoader` with `scala.reflect.internal.util.ScalaClassLoader`
- Replaced deprecated `scala.tools.ListOfNil` with `scala.reflect.internal.util.ListOfNil`
- Replaced deprecated `scala.tools.utils.ScalaClassLoader` with `scala.reflect.internal.util.ScalaClassLoader`
- Replaced deprecated `emptyValDef` with `noSelfType`
- In `BoxesRunTime` removed unused method and commented out unused values. Did not delete to keep a reference to the values. If they are deleted people might wonder why `1` and `2` are not used.
- Replaced deprecated `scala.tools.nsc.util.AbstractFileClassLoader` with `scala.reflect.internal.util.AbstractFileClassLoader`
|
|
|
|
|
|
|
|
| |
The community build discovered that #4252 introduced the possibility
for a NullPointerException. The tree with a null type was a synthetic
`Apply(<<matchEnd>>)` created by the pattern matcher.
This commit adds a null check.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
From the "Substitution is hard to do" department.
In 7babdab9a, TreeSymSubstitutor was modified to mutate the info
of symbols defined in the tree, if that symbol's info referred to
one of the `from` symbols in the substitution.
It would have been more principled to create a cloned symbol
with the updated info, and add that to the substitution. But I
wasn't able implement that correctly (let alone efficiently.)
The in-place mutation of the info of a symbol led to the crasher
in this bug: a singleton type over that symbol ends up with a stale
cached value of 'underlying'. In the enclosed test case, this leads
to a type error in the `SubstituteRecursion` of the extension
methods phase.
This commit performs a cleanup job at the end of `substituteSymbols`
by invalidating the cache of any `SingleType`-s in the tree that
refer to one of the mutated symbols.
|
|\
| |
| | |
SI-8947 Avoid cross talk between tag materializers and reify
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
As suggested in review:
- Use `abort` rather than `{error; EmptyTree} when we hit an
error in reification or tag materialization.
- Explicitly avoid adding the `MacroExpansionAttachment` to the
macro expansion if it an `EmptyTree`
- Emit a `-Xdev` warning if any other code paths find a way to
mutate attachments in places they shouldn't.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
After a macro has been expanded, the expandee are expansion are
bidirectionally linked with tree attachments. Reify uses the back
reference to replace the expansion with the expandee in the reified
tree. It also has some special cases to replace calls to macros
defined in scala-compiler.jar with `Predef.implicitly[XxxTag[T]]`.
This logic lives in `Reshape`.
However, the expansion of a macro may be `EmptyTree`. This is the case
when a tag materializer macro fails. User defined macros could do the
also expand to `EmptyTree`. In the enclosed test case, the error
message that the tag materializer issued ("cannot materialize
class tag for unsplicable type") is not displayed as the typechecker
finds another means of making the surrounding expression typecheck.
However, the macro engine attaches a backreference to the materializer
macro on `EmpytyTree`!
Later, when `reify` reshapes a tree, every occurance of `EmptyTree`
will be replaced by a call to `implicitly`.
This commit expands the domain of `CannotHaveAttrs`, which is mixed
in to `EmptyTree`. It silently ignores all attempts to mutate
attachments.
Unlike similar code that discards mutations of its type and position,
I have refrained from issuing a developer warning in this case, as
to silence this I would need to go and add a special case at any
places adding attachments.
|
|/ |
|
| |
|
| |
|
| |
|
|\
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
src/compiler/scala/reflect/macros/compiler/Resolvers.scala
src/compiler/scala/reflect/macros/contexts/Typers.scala
src/compiler/scala/tools/reflect/ToolBoxFactory.scala
src/reflect/scala/reflect/api/BuildUtils.scala
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
MemberDefs alone can't be typechecked as is, because namer only names
contents of PackageDefs, Templates and Blocks. And, if not named, a tree
can't be typed.
This commit solves this problem by wrapping typecheckees in a trivial block
and then unwrapping the result when it returns back from the typechecker.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
When I removed XXXDef(...) and XXXType(...) methods from the public API,
I put compatibility stubs in compat via implicit classes enriching
XXXExtractor traits with apply methods.
Unfortunately, this doesn't work, because if XXXDef or XXXType have
any kind of an apply method left in the public API, then implicit classes
won't even be considered when resolving calls to XXXDef(...) or XXXType(...).
Therefore I had to put those removed methods back and adorn them with an
implicit parameter that can only be resolved when "import compat._"
is in place. Quite extravagant, yes, but goes along the lines with the
design goals of the refactoring, which are:
1) Break source compatibility for users who are using methods that are
now moved to internal in order to attract attention.
2) Provide an easy way to fix the breakage by importing compat._, which
will pimp back all the missing APIs with deprecation warnings that are
going to guide migration.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Reflection API exhibits a tension inherent to experimental things:
on the one hand we want it to grow into a beautiful and robust API,
but on the other hand we have to deal with immaturity of underlying mechanisms
by providing not very pretty solutions to enable important use cases.
In Scala 2.10, which was our first stab at reflection API, we didn't
have a systematic approach to dealing with this tension, sometimes exposing
too much of internals (e.g. Symbol.deSkolemize) and sometimes exposing
too little (e.g. there's still no facility to change owners, to do typing
transformations, etc). This resulted in certain confusion with some internal
APIs living among public ones, scaring the newcomers, and some internal APIs
only available via casting, which requires intimate knowledge of the
compiler and breaks compatibility guarantees.
This led to creation of the `internal` API module for the reflection API,
which provides advanced APIs necessary for macros that push boundaries
of the state of the art, clearly demarcating them from the more or less
straightforward rest and providing compatibility guarantees on par with
the rest of the reflection API.
This commit does break source compatibility with reflection API in 2.10,
but the next commit is going to introduce a strategy of dealing with that.
|
| |
| |
| |
| |
| | |
Adds some clarifications to docs, introduces personally long-awaited
treeCopy.RefTree that lets us deal with RefTrees in a fully uniform fashion.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Back then when we planned to introduce type macros, we relaxed the type
of DefDef.name from TermName to Name in order to potentially be able to
accommodate type names for type macros.
Since then, type macros have been cancelled (and, for the record, my
implementation of type macros in paradise 1.0 didn’t involve DefDefs
with TypeNames), and we’ve rolled back the change to DefDef.name.
What we forgot to do, however, was to change the type of ValOrDefDef.name,
which is taken care of in this commit.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Makes sure that almost every abstract type declared in reflection API
erases to a unique class, so that they can be adequately used for
method overloading to the same extent that tags allow them to be used
in pattern matching.
The only two exceptions from this rule are the types whose implementations
we do not control: FlagSet that is implemented as Long and RuntimeClass
that is implemented as java.lang.Class[_].
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There’s been a conflation of two distinct meanings of the word “local”
in the internal symbol API: the first meaning being “local to this”
(as in has the LOCAL flag set), the second meaning being “local to block”
(as in declared in a block, i.e. having owner being a term symbol).
Especially confusing is the fact that sym.isLocal isn’t the same as
sym.hasFlag(LOCAL), which has led to now fixed SI-6733.
This commit fixes the semantic mess by deprecating both Symbol.isLocal and
Symbol.hasLocalFlag (that we were forced to use, because Symbol.isLocal
had already been taken), and replacing them with Symbol.isLocalToThis
and Symbol.isLocalToBlock. Unfortunately, we can’t remove the deprecated
methods right away, because they are used in SBT, so I had to take small
steps.
|
|
|
|
|
|
|
|
|
|
|
| |
Today’s flight back to Lausanne wasn’t as productive as the recent flight
to Minsk (https://github.com/scala/scala/pull/3305), but I noticed
one minor thingie: ExistentialTypeTree had an imprecise type specified
for its whereClauses. This is now fixed.
I didn’t increment PickleFormat.*Version numbers, because this change
introduces only a miniscule incompatibility with what would have been
a meaningless and most likely crash-inducing pickle anyway.
|
|
|
|
|
|
|
|
| |
Avoids calling `thisType` on the owner if it is a term symbol,
which doesn't make much sense.
This method is used internally in tree factory methods that
create, e.g, a `DefDef` based on the info of a `Symbol`.
|
|
|
|
|
|
|
|
|
| |
- Use tree factories that accept symbols and encapsulate ValDef
creation
- Use `gen.mkForwarder` to handle the conditional addition of
`: _*` for varargs functions. We don't need to predicate this
on `etaExpandKeepsStar`; the only place that need to do that
is EtaExpansion.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit puts a real body on the Delambdafy phase.
From a lambda, Delambdafy will create
1) a static forwarder at the top level of the class that contained
the lambda
2) a new top level class that
a) has fields and a constructor taking the captured environment
(including possbily the "this" reference)
b) an apply method that calls the static forwarder
c) if needed a bridge method for the apply method
3) an instantiation of the newly created class which replaces the
lambda
Trees.scala is modified to add two more convenient factories
for templates and classdefs.
A few basic tests are included to verify that it works as expected.
Further commits will have additional tests.
|
|\
| |
| | |
Traverser and Pickler improvements.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
There's a huge amount of tree traversal related duplication
which is hard to eliminate only because trees aren't properly
traversed. The not-quite-tree bits of key trees are ignored
during traversals, making it impossible to write e.g. a
pretty printer without duplicating almost the entire traversal
logic (as indeed is done for printing purposes in more than one
place.) And almost the entire pickler logic is redundant with
Traverser, except since it's all duplicated of course it diverged.
The pickler issue is remedied in the commit to follow.
The not-quite-trees not quite being traversed were Modifiers, Name,
ImportSelector, and Constant. Now every case field of every tree is
traversed, with classes which aren't trees traversed via the following
methods, default implementations as shown:
def traverseName(name: Name): Unit = ()
def traverseConstant(c: Constant): Unit = ()
def traverseImportSelector(sel: ImportSelector): Unit = ()
def traverseModifiers(mods: Modifiers): Unit = traverseAnnotations(mods.annotations)
|
|/
|
|
|
|
| |
As long as it's a block of pure boilerplate we have to
navigate around all the time, it may as well be the most
beautiful boilerplate it knows how to be.
|
|
|
|
|
|
|
| |
Most of this was revealed via -Xlint with a flag which assumes
closed world. I can't see how to check the assumes-closed-world
code in without it being an ordeal. I'll leave it in a branch in
case anyone wants to finish the long slog to the merge.
|
|\
| |
| | |
Updating Position call sites.
|
| |
| |
| |
| |
| |
| | |
Calling position factories rather than instantiating these
particular classes. Not calling deprecated methods. Added a few
position combinator methods.
|
|\ \
| |/
|/| |
transformers no longer ignore UnApply.fun
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Second time's the charm. I remember trying to do exactly the same somewhen
around 2.10.0-M4, but then some continuations tests were failing.
Luckily, today everything went smoothly.
Please note that this fix changes the way that SI-5465 manifests itself.
Previously it produced type errors, now it simply crashes the compiler.
Therefore I had to attach the try/catch FatalError clause to invocations
of toolbox methods, so that compiler crashes get caught and translated to
ToolBoxErrors.
Also fixes SI-7871, and that clears the way for implementing quasiquotes
with conventional macros rather than relying on a special case in typer.
|
|/
|
|
|
| |
Looks like emptyValDef.isEmpty was already changed to return
false, so now all that's left is a name which means something.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Rather than localizing the fix to the outerAccessor, this
commit pushed the call to `memberType` into *all* usages of
`newValDef` and `newDefDef`.
The TPT of `applyOrElse` in synthetized partial functions must
be set explicitly to pass the pos/t7853-partial-function.scala.
Otherwise, the as-seen-from ends up cloning the type parameter
`B1` of `applyOrElse` as it transforms (questionably)
its bound from `List[Int @unchecked]` to `List[Int]`.
Partial Function synthesis was already a delicate area, and this
makes things more explicit which could be counted as an improvement.
|
|\
| |
| | |
SI-1909 SI-3832 SI-7007 SI-7223 Improved handling of larval objects
|
| |
| |
| |
| | |
In order to reduce the noise in OuterPathTransformer.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
TreeDSL has no future - it was always a temporary measure
waiting for something like quasiquotes to come along. In this
commit I cull as much of it as I can, especially the delicate
matter of creating new DefDefs and ValDefs, which I completely
turn over to the old style creators.
I unified all the symbol-based DefDef and ValDef creators under
a single method, since it was yet another place where ctrl-C and
ctrl-V were being punched with glee. Was beaten to the punch on
adding copyTypeDef to fill out the *Def creators.
Eliminated as many redundant positioning calls as I could find.
If you are creating a DefTree tree based on a symbol, it will
always have an atPos(sym.pos) { ... } wrapped around it. You
don't need another one.
All of this is motivated by positions work: positions are
assigned in so many places and in such an ad hoc fashion that
it is impossible to bring consistency to that without first
bringing some consistency to tree creation.
|
|/
|
|
|
|
|
|
|
|
|
|
| |
The implementation had come to depend on finalResultType
accidentally doing things beyond its charter - in particular,
widening types. After hunting down and fixing the call sites
depending on the bugs, I was able to rewrite the method to do
only what it's supposed to do.
I threw in a different way of writing it entirely to suggest how
some correctness might be obtained in the future. It's a lot
harder for a method written like this to break.
|
| |
|
|\
| |
| |
| |
| |
| | |
Conflicts:
src/compiler/scala/tools/nsc/Global.scala
src/compiler/scala/tools/nsc/transform/patmat/MatchTranslation.scala
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Historically calling NoSymbol.owner has crashed the compiler.
With this commit, NoSymbol owns itself. This is consistent with
the way ownership chains are handled elsewhere in the compiler
(e.g. NoContext.owner is NoContext, NoSymbol.enclClass is
NoSymbol, and so on) and frees every call site which handles
symbols from having to perform precondition tests against
NoSymbol.
Since calling NoSymbol.owner sometimes (not always) indicates
a bug which we'd like to catch sooner than later, I have
introduced a couple more methods for selected call sites.
def owner: Symbol // NoSymbol.owner is self, log if -Xdev
def safeOwner: Symbol // NoSymbol.owner is self, ignore
def assertOwner: Symbol // NoSymbol.owner is fatal
The idea is that everyone can call sym.owner without undue anxiety
or paranoid null-like tests. When compiling under -Xdev calls to
`owner` are logged with a stack trace, so any call sites for which
that is an expected occurrence should call safeOwner instead to
communicate the intention and stay out of the log. Conversely, any
call site where crashing on the owner call was a desirable behavior
can opt into calling assertOwner.
This commit also includes all the safeOwner calls necessary to
give us a silent log when compiling scala.
|
|/ |
|
|
|
|
|
|
|
|
|
|
|
| |
Previously setPos, pos_=, setType, tpe_= all behaved inconsistently
between each other even though they all represent similar
attributes that cannot be changed on CannotHaveAttrs trees.
In order to simplify handling of such trees in compiler code each
of these fields now supports assignment to its current default
value: NoType for tpe and NoPosition for pos. Such assignments
don't mutate underlying trees.
|
|
|
|
|
| |
Now when there's no hope left for type macros, it's reasonable to provide
a more specific type for DefDef.name.
|