| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |\
| |
| | |
SI-3772 Fix detection of term-owned companions
|
| | | |
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Companion detection consults the scopes of enclosing Contexts during
typechecking to avoid the cycles that would ensue if we had to look
at into the info of enclosing class symbols. For example, this used
to typecheck:
object CC { val outer = 42 }
if ("".isEmpty) {
case class CC(c: Int)
CC.outer
}
This logic was not suitably hardened to find companions in exactly
the same nesting level.
After fixing this problem, a similar problem in `Namer::inCurrentScope`
could be solved to be more selective about synthesizing a companion
object. In particular, if a manually defined companion trails after
the case class, don't create an addiotional synthetic comanpanion object.
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
VarHandles bring a host of new "polymorphic signature" methods to
the Java 9 standard library. This commit updates the way we detect
such methods to look at the absense/presense of the
`PolymorphicSignature` annotation, so as to include these (and any
future additions.)
When we see applications of such methods, we disable adaptation
of argument and return types.
Tested manually with JDK9-ea:
```
Welcome to Scala 2.12.2-20161208-165912-3de1c0c (Java HotSpot(TM) 64-Bit Server VM, Java 9-ea).
Type in expressions for evaluation. Or try :help.
scala> import java.lang.invoke._, scala.runtime.IntRef
import java.lang.invoke._
import scala.runtime.IntRef
scala> val varHandle = MethodHandles.lookup().in(classOf[IntRef]).findVarHandle(classOf[IntRef], "elem", classOf[Int]);
varHandle: java.lang.invoke.VarHandle = java.lang.invoke.VarHandleInts$FieldInstanceReadWrite@7112ce6
scala> varHandle.getAndSet(ref, 1): Int
res5: Int = 0
scala> varHandle.getAndSet(ref, 1): Int
res6: Int = 1
```
Inspecting bytecode shows the absense of box/unboxing:
```
object Test {
import java.lang.invoke._, scala.runtime.IntRef
val varHandle = MethodHandles.lookup().in(classOf[IntRef]).findVarHandle(classOf[IntRef], "elem", classOf[Int]);
def main(args: Array[String]): Unit = {
val i : Int = varHandle.getAndSet(IntRef.zero, 1)
}
}
```
```
public void main(java.lang.String[]);
Code:
0: aload_0
1: invokevirtual #28 // Method varHandle:()Ljava/lang/invoke/VarHandle;
4: invokestatic #34 // Method scala/runtime/IntRef.zero:()Lscala/runtime/IntRef;
7: iconst_1
8: invokevirtual #40 // Method java/lang/invoke/VarHandle.getAndSet:(Lscala/runtime/IntRef;I)I
11: istore_2
12: return
```
|
| |\ \
| | |
| | | |
Remove existential infer hack enabled by #5376
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
It wasn't a good idea after all.
Also removed some tracing code that I cannot imagine
was ever used in a production compiler. It's still just
a recompile away.
Fixes scala/scala-dev#280
|
| |/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
http://openjdk.java.net/jeps/220 changes the layout of the
JDK to encapsulate the provided libraries with the new module
system.
This commit modifies the compiler's classpath implementation
to scan the new location of these, the `jrt://` virtual filesystem.
This might need to be adjusted once we provide a means for
users to specify the subset of modules that they want to
depend on, but for now reclaims the ground we lost.
```
⚡ (java_use 9-ea; qscala)
Welcome to Scala 2.12.0-20160908-223617-7e4ebda (Java HotSpot(TM) 64-Bit Server VM, Java 9-ea).
Type in expressions for evaluation. Or try :help.
scala> import StackWalker._, java.util.stream._, scala.collection.JavaConverters._
import StackWalker._
import java.util.stream._
import scala.collection.JavaConverters._
scala> (() => StackWalker.getInstance(java.util.EnumSet.of(Option.RETAIN_CLASS_REFERENCE)).walk[Seq[String]]((s: java.util.stream.Stream[StackFrame]) => s.iterator.asScala.take(3).map(_.toString).toList)).apply().mkString("\n")
res0: String =
.$anonfun$res0$1(<console>:21)
.<init>(<console>:21)
.<clinit>(<console>)
scala>
```
I've marked the new class, `NioFile` as `private[scala]` to justify
the forward compatibility whitelist entry.
In principle we could use NioFile more widely rather than `PlainFile`
I tried this out in https://github.com/retronym/scala/commit/b2d0a17a
which passed CI. But to be conservative, I'm not submitting that change
at this point.
|
| |\ \
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
SI-7046 reflection doesn't see all knownDirectSubclasses
This appears to do the right thing in the most typical scenarios in which `knownDirectSubclasses` would be used. The missing 5% is that subclasses defined in local scopes might not be seen by `knownDirectSubclasses` (see `Local` and `Riddle` in the test below). In mitigation, though, it is almost certain that a local subclass would represent an error in any scenario where `knownDirectSubclasses` might be used.
Errors for such situations are reported by recording (via a symbol attachment) that `knownDirectSubclasses` has been called and reporting an error if any additional children are added subsequently.
Despite these limitations and caveats, I believe that this represents a huge improvement over the status quo, and would eliminate 100% of the failures that I've seen in practice with people using shapeless for type class derivation.
|
| | | | |
|
| |\ \ \
| | | |
| | | | |
Clean up of code guarded by bare -Xexperimental
|
| | | | | |
|
| | |/ / |
|
| |\ \ \
| | | |
| | | | |
SD-275 Further harden against refs to absentee classes
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
AFAICT, this was only needed to support pickle compatibility after
the fix for SI-1591. We don't need to maintain the compatibility
after incrementing our major version.
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
- Limit the strategy of unpickling an external reference to a
module class to a lookup of the module var to non-stub owners
in order to enable fall through to stub symbol creation.
Fixes scala/scala-dev#275
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
- Rework previous fixes for SI-8502 to move the creation of
a term or type stub symbol during unpickling to the initial
point of stub creation, based on the tag.
- Just set the PACKAGE flag on class stub symbols created during
unpickling `ThisType`, rather than bothering with a different
subclass of `StubSymbol` for (assumed) packages.
|
| |\ \ \ \
| | | | |
| | | | | |
SI-10009 Fields survive untypecheck/retypecheck
|
| | | |_|/
| |/| |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Some places in the compiler, and many places in macros, use
`untypecheck` (aka `resetAttrs`) to strip types and local symbols
from a tree before retypechecking it under some different context.
The refactoring of the desugaring of vals and vars in Scala 2.12.0
broke an assumption in this facility.
When a ValDef must be split into multiple members (e.g. a field and
a getter, or a perhaps also a setter), the ValDef that was parsed
assumes the role of the `field`, and the trees for other members are
stached by `Namer` to the `synthetics` map of the compilation unit,
in order to spliced into the right statement list by typechecking.
See `enterGetterSetter` for more details.
However, the parsed ValDef is now used verbatim, carrying the meaning
(ie, the symbol) of the `private[this]` field. This tree now had
an inconsistency between the flags in `tree.mods.flags` and
`tree.symbol.flags`. `tree.name` also differed from `tree.symbol.name`
(the latter was renamed to be a local name, ie one with a trailing space.)
When `ResetAttrs` stripped off the symbol and we retypechecked, we'd
end up with two symbols in scope with the same name.
In the first from the `run` test:
```
================================================================================
{
class a extends scala.AnyRef {
def <init>(): a = {
a.super.<init>();
()
};
private[this] val x: Int = 42;
<stable> <accessor> def x: Int = a.this.x
};
new a()
}
{
class a extends scala.AnyRef {
def <init>() = {
super.<init>();
()
};
val x = 42; // oops, the name is "x" rather than "x " and we've missing `private[this]`!
<stable> <accessor> def x: Int = a.this.x
};
new a()
}
scala.tools.reflect.ToolBoxError: reflective typecheck has failed: x is already defined as value x
```
This commit uses the flags and name of the symbol in `typedValDef`.
I've also had to modify the internals of `CodePrinter` to use the
implicit, override, and deferred flags from the modifiers of an
accessor when recovering pre-typer tree for a ValDef.
|
| |\ \ \ \
| |_|/ /
|/| | | |
Avoid name table pollution with fresh existentials
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
During large compilations runs, the large numbers of globally unique
fresh names for existentials captured from prefixes of `asSeenFrom`.
is a) somewhat wasteful (all these names are interned in the name table)
, and, b) form a pathological case for the current implementation of
`Names#hashValue`, which leads to overfull hash-buckets in the name table.
`hashValue` should probably be improved, but my attempts to do so have
shown a small performance degradation in some benchmarks. So this commit
starts by being more frugal with these names, only uniquely naming
within an `asSeenFrom` operation.
References scala/scala-dev#246
|
| | |_|/
|/| |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
The REPL has a long running instance of Global which outputs
classfiles by default to a VirtualDirectory. The inliner did not find
any of these class files when compiling calls to methods defined in
previous runs (ie, previous lines of input.)
This commit:
- Adds a hook to augment the classpath that the optimizer searches,
and uses this in the REPL to add the output directory
- Fixes the implementation of `findClassFile` in VirtualDirectory,
which doesn't seem to have been used in anger before. I've factored out
some common code into a new method on `AbstractFile`.
- Fixes a similar problem getSubDir reported by Li Haoyi
- Adds missing unit test coverage.
This also fixes a bug in REPL autocompletion for types defined
in packages >= 2 level deep (with the `:paste -raw` command).
I've added a test for this case.
|
| | | |
| | |
| | |
| | |
| | |
| | | |
Cloning the original symbol in its entirety, rather than cloning
its type/value parameters individually. `cloneSymbol` takes care of
all the tricky substitutions for us!
|
| | | | |
|
| | |/
|/|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Make sure that methods annotated with varargs are properly mixed-in. This commit
splits the transformation into an info transformer (that works on all symbols, whether
they come from source or binary) and a tree transformer.
The gist of this is that the symbol-creation part of the code was moved to the UnCurry
info transformer, while tree operations remained in the tree transformer. The newly
created symbol is attached to the original method so that the tree transformer can still
retrieve the symbol.
A few fall outs:
- I removed a local map that was identical to TypeParamsVarargsAttachment
- moved the said attachment to StdAttachments so it’s visible between reflect.internal
and nsc.transform
- a couple more comments in UnCurry to honour the boy-scout rule
|
| |\ \
| | |
| | | |
SI-9885 Don't return offset past EOF
|
| | | |
| | |
| | |
| | |
| | | |
On bad line number, `lineToOffset` should not return
an offset past EOF (which was sentinel, internally).
|
| | | | |
|
| |\ \ \
| | | |
| | | | |
avoid boxing
|
| | | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
scala.runtime.Rich{Double, Float} has `isNaN` and these are value class.
Also java.lang.{Double, Float} has `isNaN`.
- https://docs.oracle.com/javase/8/docs/api/java/lang/Double.html#isNaN--
- https://docs.oracle.com/javase/8/docs/api/java/lang/Float.html#isNaN--
We can't call `RichDouble#isNaN` because
`implicit def double2Double(x: Double): java.lang.Double`
is higher priority than
`implicit def doubleWrapper(x: Double): RichDouble`
```
$ scala -version
Scala code runner version 2.11.8 -- Copyright 2002-2016, LAMP/EPFL
$ scala -Xprint:jvm -e "1.0.isNaN"
[[syntax trees at end of jvm]] // scalacmd616162202928036892.scala
package <empty> {
object Main extends Object {
def main(args: Array[String]): Unit = {
new <$anon: Object>();
()
};
def <init>(): Main.type = {
Main.super.<init>();
()
}
};
final class anon$1 extends Object {
def <init>(): <$anon: Object> = {
anon$1.super.<init>();
scala.this.Predef.double2Double(1.0).isNaN();
()
}
}
}
```
|
| | |_|/
|/| |
| | |
| | |
| | | |
to avoid the same kind of slowdowns that Vector was experiencing due
to the less aggressive inlining by scalac.
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Tighten some types (Symbol -> ClassSymbol / ModuleSymbol), use NonFatal
instead of catching Throwable.
Also don't run the classfile parser enteringPhase(phaseBeforeRefchecks)
anymore. This was added in 0ccdb15 but seems no longer required.
|
| | | | |
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
When unpickling a class, if the name and owner matches the current
`classRoot` of the unpickling Scan, that `classRoot` symbol is used
instead of creating a new symbol.
If, in addition, the class being unpickled has the MODULE flag, the
unpickler should use the `moduleRoot.moduleClass` symbol (instead of
creating a new one).
To identify the module class, the current implementation compares the
name and owner to the `classRoot`. This fails in case the `classRoot`
is `NoSymbol`, which can happen in corner cases (when a type alias
shadows a class symbol, scala-dev#248).
In this patch we identify the module class by comparing the name and
owner to the `moduleRoot` symbol directly (using a `toTypeName`).
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
In SymbolLoaders, when seeing a classfile `Foo.class`, we always
(unconditionally) create 3 symbols: a class, a module and a module
class. Some symbols get invalidated later (`.exists`).
Until now, the classfile parser (and unpickler) received the "root"
symbol as argument, which is the symbol whose type is being completed.
This is either the class symbol or the module symbol.
The classfile parser would then try to lookup the other symbol through
`root.companionClass` or `root.companionModule`. Howver, this lookup can
fail. One example is scala-dev#248: when a type alias (in a package
object) shadows a class symbol, `companionClass` will fail.
The implementations of the classfile parser / unpickler assume that
both the `clazz` and the `staticModule` symbols are available. This
change makes sure that they are always passed in explicitly.
Before this patch, in the example of scala-dev#248, the `classRoot` of
the unpickler was NoSymbol. This caused a bug when unpickling the
module class symbol, causing a second module class symbol to be created
mistakingly. The next commit cleans up this logic, more details there.
This second symbol would then cause the crash in the backend because it
doesn't have an `associatedFile`, therefore `isCoDefinedWith` would
spuriously return `true`.
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
This makes getClassByName fail / getClassIfDefined return NoSymbol
when querying an alias.
The current behavior can confuse the classfile parser: when parsing a
class, a cross-check verifies that `pool.getClassSymbol(nameIdx)`
returns the symbol of the class currently being parsed. If there's a
type alias that shadows the linked class, following the alias would
return an unrelated class.
(The cross-check didn't fail because there are some other guards
around it)
The logic to follow aliases was was added in ff98878, without a clear
explanation. Note that `requiredClass[C]` works if `C` is an alias, it
is expanded by the compiler.
|
| |/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This fixes scala/scala-dev#248, where a type alias reached the backend
through this method.
This is very similar to the fix for SI-5031, which changed it only in
ModuleSymbol, but not in Symbol.
The override in ModuleSymbol is actually unnecessary (it's identical),
so it's removed in this commit. It was added for unclear reasons in
296b706.
|
| |\ \
| | |
| | | |
SI-9943 sealed class does not yield SAM type
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Cannot subclass such a class. (Well, we could subclass a sealed
class in the same compilation unit. We ignore this for simplicity.)
This is a bit of a sneaky fix for this bug, but our hand
is pretty much forced by other constraints, in this intersection
of overload resolution involving built-in function types and SAMs,
and type inference for higher-order function literals (#5307).
Luckily, in this particular issue, the overloading clash seems
accidental. The `sealed` `<:<` class is not a SAM type as it
cannot be subclassed outside of `Predef`. For simplicity,
we don't consider where the SAM conversion occurs and exclude
all sealed classes from yielding SAM types.
Thanks to Miles for pointing out that `final` was missing in my
first iteration of this fix.
|
| |/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
If a non-transient lazy val is the only user of a private field in
a class, the field is nulled out at the end of the lazy initializer.
This is tested in the existing test `run/lazy-leaks.scala`.
The analysis of which fields could be nulled out was recently moved
from `mixin` to the new `fields` phase. This introduced a regression
as a reference from an inner- or companion-classes had not yet
been processed by `explicitouter` to publicise private fields.
This commit delays the analysis to mixin (after explicit outer has
done its work.)
Navigating from `foo$lzycompute()` to `foo()` to `foo` is a little
dirty now. I'm not sure whether there is a more robust way to
structure things.
|
| |\ \
| | |
| | | |
Emit local module like lazy val
|
| | | |
| | |
| | |
| | |
| | |
| | | |
There's still a lot of duplication,
as well as plenty of opportunities
for constant folding / simplification.
|
| |/ /
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
When a class captures an outer value, a field for that value is created
in the class. The class also gets a constructor parameter for the
captured value, the constructor will assign the field.
LambdaLift re-writes accesses to the local value (Ident trees) to the
field. However, if the statement accessing the local value will end up
inside the constructor, the access is re-written to the constructor
parameter instead. This is the case for constructor statements:
class C {
{
println(capturedLocal)
}
}
If C extends DelayedInit, the statement does not end up in C's
constructor, but into a new synthetic method. The access to
`capturedLocal` needs to be re-written to the field instead of the
constructor parameter.
LambdaLift takes the decision (field or constructor parameter) based on
the owner chain of `currentOwner`. For the constructor statement block,
the owner is a local dummy, for which `logicallyEnclosingMember` returns
the constructor symbol.
This commit introduces a special case in LambdaLift for local dummies
of DelayedInit subclasses: instead of the constructor, we use a
temporary symbol representing the synthetic method holding the
initializer statements.
|
| | | |
|
| |\ \
| | |
| | | |
Fixes to mixin forwarders
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
JUnit 4 does not support default methods. For better user experience,
this commit makes the compiler generate mixin forwarders for inherited
trait methods that carry a JUnit annotation.
The -Yjunit-trait-methods-no-forwarders flag disables this behavior.
This supersedes the scala-js/scala-2.12-junit-mixin-plugin compiler
plugin.
|
| |\ \ \
| |/ /
|/| | |
Fields: expand lazy vals during fields, like modules
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
The goal is to avoid emitting unneeded `BoxedUnit` values,
which are the result of adapting a `Unit`-typed expression
inside a `synchronized(...)` to the erased type of
`synchronized`'s argument -- `Object`.
The proposed solution gives `synchronized` a polymorphic
type (the info of the type param is still erased so that
bounds checking works in the erased type system), so that
an application `synchronized(println("boo"))` erases to
`synchronized[Unit])(println("boo"))`, and no boxing is
performed on the `println("boo")` argument, whose expected
type is now `Unit` instead of `Object`.
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Essentially, we fuse mixin and lazyvals into the fields phase.
With fields mixing in trait members into subclasses, we
have all info needed to compute bitmaps, and thus we can
synthesize the synchronisation logic as well.
By doing this before erasure we get better signatures,
and before specialized means specialized lazy vals work now.
Mixins is now almost reduced to its essence: implementing
super accessors and forwarders. It still synthesizes
accessors for param accessors and early init trait vals.
Concretely, trait lazy vals are mixed into subclasses
with the needed synchronization logic in place, as do
lazy vals in classes and methods. Similarly, modules
are initialized using double checked locking.
Since the code to initialize a module is short,
we do not emit compute methods for modules (anymore).
For simplicity, local lazy vals do not get a compute method either.
The strange corner case of constant-typed final lazy vals
is resolved in favor of laziness, by no longer assigning
a constant type to a lazy val (see widenIfNecessary in namers).
If you explicitly ask for something lazy, you get laziness;
with the constant-typedness implicit, it yields to the
conflicting `lazy` modifier because it is explicit.
Co-Authored-By: Lukas Rytz <lukas@lightbend.com>
Fixes scala/scala-dev#133
Inspired by dotc, desugar a local `lazy val x = rhs` into
```
val x$lzy = new scala.runtime.LazyInt()
def x(): Int = {
x$lzy.synchronized {
if (!x$lzy.initialized) {
x$lzy.initialized = true
x$lzy.value = rhs
}
x$lzy.value
}
}
```
Note that the 2.11 decoding (into a local variable and a bitmap) also
creates boxes for local lazy vals, in fact two for each lazy val:
```
def f = {
lazy val x = 0
x
}
```
desugars to
```
public int f() {
IntRef x$lzy = IntRef.zero();
VolatileByteRef bitmap$0 = VolatileByteRef.create((byte)0);
return this.x$1(x$lzy, bitmap$0);
}
private final int x$lzycompute$1(IntRef x$lzy$1, VolatileByteRef bitmap$0$1) {
C c = this;
synchronized (c) {
if ((byte)(bitmap$0$1.elem & 1) == 0) {
x$lzy$1.elem = 0;
bitmap$0$1.elem = (byte)(bitmap$0$1.elem | 1);
}
return x$lzy$1.elem;
}
}
private final int x$1(IntRef x$lzy$1, VolatileByteRef bitmap$0$1) {
return (byte)(bitmap$0$1.elem & 1) == 0 ?
this.x$lzycompute$1(x$lzy$1, bitmap$0$1) : x$lzy$1.elem;
}
```
An additional problem with the above encoding is that the `lzycompute`
method synchronizes on `this`. In connection with the new lambda
encoding that no longer generates anonymous classes, captured lazy vals
no longer synchronize on the lambda object.
The new encoding solves this problem (scala/scala-dev#133)
by synchronizing on the lazy holder.
Currently, we don't exploit the fact that the initialized field
is `@volatile`, because it's not clear the performance is needed
for local lazy vals (as they are not contended, and as soon as
the VM warms up, biased locking should deal with that)
Note, be very very careful when moving to double-checked locking,
as this needs a different variation than the one we use for
class-member lazy vals. A read of a volatile field of a class
does not necessarily impart any knowledge about a "subsequent" read
of another non-volatile field of the same object. A pair of
volatile reads and write can be used to implement a lock, but it's
not clear if the complexity is worth an unproven performance gain.
(Once the performance gain is proven, let's change the encoding.)
- don't explicitly init bitmap in bytecode
- must apply method to () explicitly after uncurry
|
| | | | |
|
| | | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
They remain ValDefs until then.
- remove lazy accessor logic
now that we have a single ValDef for lazy vals,
with the underlying machinery being hidden until the fields phase
leave a `@deprecated def lazyAccessor` for scala-refactoring
- don't skolemize in purely synthetic getters,
but *do* skolemize in lazy accessor during typers
Lazy accessors have arbitrary user code, so have to skolemize.
We exempt the purely synthetic accessors (`isSyntheticAccessor`)
for strict vals, and lazy accessors emitted by the fields phase
to avoid spurious type mismatches due to issues with existentials
(That bug is tracked as https://github.com/scala/scala-dev/issues/165)
When we're past typer, lazy accessors are synthetic,
but before they are user-defined to make this hack less hacky,
we could rework our flag usage to allow for
requiring both the ACCESSOR and the SYNTHETIC bits
to identify synthetic accessors and trigger the exemption.
see also https://github.com/scala/scala-dev/issues/165
ok 7 - pos/existentials-harmful.scala
ok 8 - pos/t2435.scala
ok 9 - pos/existentials.scala
previous attempt: skolemize type of val inside the private[this] val
because its type is only observed from inside the
accessor methods (inside the method scope its existentials are skolemized)
- bean accessors have regular method types, not nullary method types
- must re-infer type for param accessor
some weirdness with scoping of param accessor vals and defs?
- tailcalls detect lazy vals, which are defdefs after fields
- can inline constant lazy val from trait
- don't mix in fields etc for an overridden lazy val
- need try-lift in lazy vals: the assign is not seen in uncurry
because fields does the transform (see run/t2333.scala)
- ensure field members end up final in bytecode
- implicit class companion method: annot filter in completer
- update check: previous error message was tangled up with unrelated
field definitions (`var s` and `val s_scope`),
now it behaves consistently whether those are val/vars or defs
- analyzer plugin check update seems benign, but no way to know...
- error message gen: there is no underlying symbol for a deferred var
look for missing getter/setter instead
- avoid retypechecking valdefs while duplicating for specialize
see pos/spec-private
- Scaladoc uniformly looks to field/accessor symbol
- test updates to innerClassAttribute by Lukas
|
