| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Top level modules in Scala currently desugar as:
```
class C; object O extends C { toString }
```
```
public final class O$ extends C {
public static final O$ MODULE$;
public static {};
Code:
0: new #2 // class O$
3: invokespecial #12 // Method "<init>":()V
6: return
private O$();
Code:
0: aload_0
1: invokespecial #13 // Method C."<init>":()V
4: aload_0
5: putstatic #15 // Field MODULE$:LO$;
8: aload_0
9: invokevirtual #21 // Method java/lang/Object.toString:()Ljava/lang/String;
12: pop
13: return
}
```
The static initalizer `<clinit>` calls the constructor `<init>`, which
invokes superclass constructor, assigns `MODULE$= this`, and then runs
the remainder of the object's constructor (`toString` in the example
above.)
It turns out that this relies on a bug in the JVM's verifier: assignment to a
static final must occur lexically within the <clinit>, not from within `<init>`
(even if the latter is happens to be called by the former).
I'd like to move the assignment to <clinit> but that would
change behaviour of "benign" cyclic references between modules.
Example:
```
package p1; class CC { def foo = O.bar}; object O {new CC().foo; def bar = println(1)};
// Exiting paste mode, now interpreting.
scala> p1.O
1
```
This relies on the way that we assign MODULE$ field after the super class constructors
are finished, but before the rest of the module constructor is called.
Instead, this commit removes the ACC_FINAL bit from the field. It actually wasn't
behaving as final at all, precisely the issue that the stricter verifier
now alerts us to.
```
scala> :paste -raw
// Entering paste mode (ctrl-D to finish)
package p1; object O
// Exiting paste mode, now interpreting.
scala> val O1 = p1.O
O1: p1.O.type = p1.O$@ee7d9f1
scala> scala.reflect.ensureAccessible(p1.O.getClass.getDeclaredConstructor()).newInstance()
res0: p1.O.type = p1.O$@64cee07
scala> O1 eq p1.O
res1: Boolean = false
```
We will still achieve safe publication of the assignment to other threads
by virtue of the fact that `<clinit>` is executed within the scope of
an initlization lock, as specified by:
https://docs.oracle.com/javase/specs/jvms/se8/html/jvms-5.html#jvms-5.5
Fixes scala/scala-dev#SD-194
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
One step towards teasing apart the mixin phase, making
each phase that adds members to traits responsible for
mixing in those members into subclasses of said traits.
Another design tenet is to not emit symbols or trees
only to later remove them. Therefore, we model a
val in a trait as its accessor. The underlying field
is an implementation detail. It must be mixed into
subclasses, but has no business in a trait (an interface).
Also trying to reduce tree creation by changing less in subtrees
during tree transforms.
A lot of nice fixes fall out from this rework:
- Correct bridges and more precise generic signatures for
mixed in accessors, since they are now created before erasure.
- Correct enclosing method attribute for classes nested in trait fields.
Trait fields are now created as MethodSymbol (no longer TermSymbol).
This symbol shows up in the `originalOwner` chain of a class declared
within the field initializer. This promoted the field getter to
being the enclosing method of the nested class, which it is not
(the EnclosingMethod attribute is a source-level property).
- Signature inference is now more similar between vals and defs
- No more field for constant-typed vals, or mixed in accessors
for subclasses. A constant val can be fully implemented in a trait.
TODO:
- give same treatment to trait lazy vals (only accessors, no fields)
- remove support for presuper vals in traits
(they don't have the right init semantics in traits anyway)
- lambdalift should emit accessors for captured vals in traits,
not a field
Assorted notes from the full git history before squashing below.
Unit-typed vals: don't suppress field
it affects the memory model -- even a write of unit to a field is relevant...
unit-typed lazy vals should never receive a field
this need was unmasked by test/files/run/t7843-jsr223-service.scala,
which no longer printed the output expected from the `0 to 10 foreach`
Use getter.referenced to track traitsetter
reify's toolbox compiler changes the name of the trait
that owns the accessor between fields and constructors (`$` suffix),
so that the trait setter cannot be found when doing mkAssign in constructors
this could be solved by creating the mkAssign tree immediately during fields
anyway, first experiment: use `referenced` now that fields runs closer
to the constructors phase (I tried this before and something broke)
Infer result type for `val`s, like we do for `def`s
The lack of result type inference caused pos/t6780 to fail
in the new field encoding for traits, as there is no separate accessor,
and method synthesis computes the type signature based on the ValDef tree.
This caused a cyclic error in implicit search, because now the
implicit val's result type was not inferred from the super member,
and inferring it from the RHS would cause implicit search to consider
the member in question, so that a cycle is detected and type checking fails...
Regardless of the new encoding, we should consistently infer result types
for `def`s and `val`s.
Removed test/files/run/t4287inferredMethodTypes.scala and test/files/presentation/t4287c,
since they were relying on inferring argument types from "overridden" constructors
in a test for range positions of default arguments. Constructors don't override,
so that was a mis-feature of -Yinfer-argument-types.
Had to slightly refactor test/files/presentation/doc, as it was relying
on scalac inferring a big intersection type to approximate the anonymous
class that's instantiated for `override lazy val analyzer`.
Now that we infer `Global` as the expected type based on the overridden val,
we make `getComment` private in navigating between good old Skylla and Charybdis.
I'm not sure why we need this restriction for anonymous classes though;
only structural calls are restricted in the way that we're trying to avoid.
The old behavior is maintained nder -Xsource:2.11.
Tests:
- test/files/{pos,neg}/val_infer.scala
- test/files/neg/val_sig_infer_match.scala
- test/files/neg/val_sig_infer_struct.scala
need NMT when inferring sig for accessor
Q: why are we calling valDefSig and not methodSig?
A: traits use defs for vals, but still use valDefSig...
keep accessor and field info in synch
|
| | |
|
| |
|
|
|
|
| |
This helps keeping ClassTag serialization stable under accidental
changes (like changing the order of definitions, which would change the
name of the anonymous classes).
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We compile FunctionN to Java 8's idea of a function now,
so no need to target the artisanal JFunction and friends,
except when the function is specialized, as I don't yet
see how we can use LMF with the way specialization handles
FunctionN:
First, the working status quo -- the hand-crafted specialized
versions of JFunction0. Notice how `apply$mcB$sp` is looking
pretty SAMmy:
```
@FunctionalInterface
public interface JFunction0$mcB$sp extends JFunction0 {
@Override
public byte apply$mcB$sp();
@Override
default public Object apply() {
return BoxesRunTime.boxToByte(this.apply$mcB$sp());
}
}
```
Contrast this with our specialized standard FunctionN:
```
public interface Function0<R> {
public R apply();
default public byte apply$mcB$sp() {
return BoxesRunTime.unboxToByte(this.apply());
}
}
public interface Function0$mcB$sp extends Function0<Object> { }
```
The single abstract method in `Function0$mcB$sp` is `apply`, and
the method that would let us avoid boxing, if it were abstract,
is `apply$mcB$sp`...
TODO (after M4):
- do same for specialized functions (issues with boxing?)
- remove scala/runtime/java8/JFunction* (need new STARR?)
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
These manual mixins were forwarding to the impl classes have
just been removed. We can now rely on default methods instead.
Update Tests:
- Fix test/files/pos/t1237.scala, we can't have an outer field
in an interface, always use the outer method.
- Don't crash on meaningless trait early init fields
test/files/neg/t2796.scala
- Remove impl class relate parts of inner class test
- Remove impl class relate parts of elidable test
- Remove impl class related reflection test.
- Remove test solely about trait impl classes renaming
- Update check file with additional stub symbol error
- Disable unstable parts of serialization test.
- TODO explain, and reset the expectation
|
| | |
|
| |
|
|
|
|
|
|
| |
The previous implementation of `mutable.TreeSet` uses a mutable reference to an immutable red-black tree as its underlying data structure. That leads to unnecessary objects being created, which can be a problem in systems with limited resources. It also has reduced performance when compared with common mutable implementations.
In this commit `mutable.TreeSet` is changed so that it uses the recently created `mutable.RedBlackTree` as its underlying data structure. Specialized red-black tree methods were created for working with keys for efficiency reasons. The new implementation is source-compatible with the previous one, although its serialized representation obviously changes.
Closes [SI-6938](https://issues.scala-lang.org/browse/SI-6938).
|
| |
|
|
|
|
|
|
| |
Added `with Serializable` to `MapWrapper` and `SetWrapper`.
Test verifies that serialization works in the simplest case.
Also updated tests in t8549 to check that serialization works and doesn't change.
|
| |
|
|
|
|
|
|
|
|
| |
This commit contains an implementation of a mutable red-black tree with focus on performance. It also contains a new `mutable.TreeMap` Scala collection that is backed by the aforementioned tree. The common generic factories and traits related to mutable sorted maps didn't exist yet, so this commit also adds them.
Regarding performance, `TreeMap` overrides (from `MapLike` and `SortedMapLike`) all of the most common methods for maps and also those whose default implementations are asymptotically worse than direct red-black tree algorithms (e.g. `last`, `clear`).
The `rangeImpl` method of `TreeMap` returns an instance of `TreeMapView`, an inner class of `TreeMap`. This view is backed by the same `RedBlackTree.Tree` instance, and therefore changes to the original map are reflected in the view and vice-versa. The semantics of mutating a view by adding and removing keys outside the view's range are the same of the current `mutable.TreeSet`. A bit less focus was given on the performance of views - in particular, getting the `size` of a `TreeMapView` is O(n) on the number of elements inside the view bounds. That can be improved in the future.
In a future commit, `mutable.TreeSet` can be changed to be backed by this red-black tree implementation.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
the structure of Option.class generated by delambdafy:method is
slightly different. For example, lambdas declared within Option are
not emitted as nested classes, so under delambdafy:method there's no
inner class entry for anonfun classes.
The test failed because serializing a ClassTag involves serializing an
Option. Option did not have a `@SerialVersionUID`, and the classfile
generated by delambdafy:method has a different value.
The annotation is required on the parent class (Option) as well as the
subclasses (Some / None). De-serializing a Some will fail if Option
has a different SerialVersionUID.
Relates to SI-8576. We should probably have more SVUID annotations in
the library.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Parts of this test fail if testing a library built with -Xcheckinit.
The failures seem to be in two categories:
- A component of the serialized structure does not have a
declared SerialVersionUID, meaning that the extra field
added to track initialization results in a different ID.
This manifests as a `java.io.InvalidClassException` when
deserializing the blobs of data saved in the test case.
- Spurious `UnitializedFieldErrors` when calling methods on
the object that has been serialized and then deserialized.
Until we figure out the right course of action (more @SerialVersionUID
annotations / weaker tests / ...), this commit disabled those tests.
|
|
|
To date, we've been hesidant to offer any guarantees about
Java serialization of standard library types among heteregenous
Scala versions.
Nonetheless, we have added `SerialVersionUID` annotations to
parts of the standard library, to offer some stability. This
protects against two winds of change: automatic calculation of
this UID might differ between JVM versions, or it might differ
due to otherwise immaterial changes to the library in Scala
releases.
With this commit, we strengthen the guarantees. Classes
marked with `SerialVersionUID` will be serialization compatible
within minor releases of Scala. This is backed up by the
enclosed test.
After major releases, we reserve the right to break this.
But the test will serve to avoid *accidental* changes.
Specifically, the test case checks:
- deserialize(serialize(x)) == x
- serialize(x) is stable over time
I have included values of all types marked with `@SerialVersionUID`
in the library. For some types, I've added variations in the
values to exercise different subclasses, such as `Set1` / `Set2`.
This found that that the serialized form of predefined `ClassTags`
included the cached identity hash code and failed the stability
test. This wasn't an issue for correctness as they also provide
`readResolve`, but I marked those fields as `@transient` in any
case to comply with the test expectations.
That whole area is good example of a serialization worst-practice:
using anonymous classes in code like:
val Object: Manifest[java.lang.Object] = new PhantomManifest[...](...) {
private def readResolve(): Any = Manifest.AnyVal
}
... will lead to instability if these declarations are shifted around
in the file. Named classes would be preferred. I've noted this in a
TODO comment for 2.12.
|
