| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
| |
|
|
|
|
| |
I swear this change didn't work last time I was in this neighborhood.
|
|
|
|
|
|
|
|
|
|
| |
For snapshots, it points to the exact commit, for releases it points to
the tag ("v" + maven version). The link now opens in a different tab,
as opening in the same frame is not compatible with github (the page
doesn't load for some reason).
Left the repo url in test/review untouched because it points to the
root of all LAMP repos. But... is anyone still using that script?!?
|
|
|
|
|
|
| |
Possibly the ugliest function in the compiler. Unfortunately
also effectively obscuring serious bugs like SI-3452. Still won't
be winning homecoming queen, but might be able to leave the house.
|
|\ |
|
| | |
|
| | |
|
| |
| |
| |
| |
| | |
sharing trees is bad
also, substing in untyped trees into typed outer trees is not good for your health
|
| |
| |
| |
| |
| |
| | |
we still do it in ifs when the packed types are equal,
but packing types in cases breaks a lot of stuff
(and I can't remember what it fixed -- we'll find out in the next test run)
|
| | |
|
| |
| |
| |
| | |
can't pass the scrutinee since its type has been widened
|
| |
| |
| |
| |
| | |
type-switch-based catches must not contain type tests on traits for some reason, so fall back to full-blown pattern match
TODO: can we improve this? this simply mimics the old pattern matcher's behavior
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
this came to light with the virtual pattern matcher, which
emits jumps like `matchEnd3(_test(Test.this, false))`, where _test is a tailcall
the nested jumping caused double-closing (the second time in ignore mode)
thus. when closing a closed block in ignore mode, simply do nothing
from genLoad for label-jumps:
note: when one of the args to genLoadLabelArguments is a jump to a label,
it will call back into genLoad and arrive at this case, which will then set ctx1.bb.ignore to true,
this is okay, since we're jumping unconditionally, so the loads and jumps emitted by the outer
call to genLoad (by calling genLoadLabelArguments and emitOnly) can safely be ignored,
however, as emitOnly will close the block, which reverses its instructions (when it's still open),
we better not reverse when the block has already been closed but is in ignore mode
(if it's not in ignore mode, double-closing is an error)
@dragos figured it out, all I did was write the comment and the `if`
test case to repro basic blocks crasher
the tailcall in the forward jump `matchEnd3(_test(Test.this, false))`
in the following program crashes the back-end (error below)
@scala.annotation.tailrec final def test(meh: Boolean): Boolean = {
<synthetic> val _$this: Test.type = Test.this;
_test(_$this,meh){
case <synthetic> val x1: Some[String] = new Some[String]("a");
case3(){
matchEnd2({
case <synthetic> val x1: Some[String] = x1;
case4(){
if (x1.ne(null))
matchEnd3(if (meh)
_test(Test.this, false)
else
false)
else
case5()
};
case5(){
matchEnd3(_test(Test.this, false))
};
matchEnd3(x){
x
}
})
};
matchEnd2(x){
x
}
}
};
The last instruction (of basic block 11) is not a control flow instruction: CONSTANT(false)
// methods
def test(meh: Boolean (BOOL)): Boolean {
locals: value meh, value _$this, value x1, value x, value x, value x1
startBlock: 1
blocks: [1,2,3,4,5,6,7,8,9,10,11,12,13]
1:
4 JUMP 2
2:
5 NEW REF(class Some)
5 DUP(REF(class Some))
5 CONSTANT("a")
5 CALL_METHOD scala.Some.<init> (static-instance)
5 STORE_LOCAL(value x1)
5 SCOPE_ENTER value x1
5 JUMP 3
3:
5 LOAD_LOCAL(value x1)
7 STORE_LOCAL(value x1)
7 SCOPE_ENTER value x1
7 JUMP 4
4:
7 LOAD_LOCAL(value x1)
7 CZJUMP (REF(class Object))NE ? 5 : 6
5:
8 LOAD_LOCAL(value meh)
8 CZJUMP (BOOL)NE ? 8 : 9
6:
? JUMP 11
7:
7 DROP BOOL
7 JUMP 11
8:
8 CONSTANT(false)
8 STORE_LOCAL(value meh)
8 JUMP 2
9:
8 CONSTANT(false)
8 JUMP 10
10:
8 STORE_LOCAL(value x)
8 JUMP 12
11:
9 JUMP 2
9 STORE_LOCAL(value meh)
9 CONSTANT(false)
12:
7 LOAD_LOCAL(value x)
7 SCOPE_EXIT value x1
7 STORE_LOCAL(value x)
7 JUMP 13
13:
5 LOAD_LOCAL(value x)
5 SCOPE_EXIT value x1
5 RETURN(BOOL)
|
| |
| |
| |
| |
| | |
the following commit deals with the fall-out in basicblocks
(double closing of blocks in ignore mode)
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
need to set up symbols with approximated info for the method and its owner
before typing the method's body; when the body has been typed,
we know what its result type will be, so update info of class and method to reflect that
better detection of synthesized matches: annotate selector rather than relying on symbol info
encapsulate CASE | SYNTHETIC flags (setting and querying)
|
|\ \ |
|
| | |
| | |
| | |
| | |
| | | |
type-switch-based catches must not contain type tests on traits for some reason, so fall back to full-blown pattern match
TODO: can we improve this? this simply mimics the old pattern matcher's behavior
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
this came to light with the virtual pattern matcher, which
emits jumps like `matchEnd3(_test(Test.this, false))`, where _test is a tailcall
the nested jumping caused double-closing (the second time in ignore mode)
thus. when closing a closed block in ignore mode, simply do nothing
from genLoad for label-jumps:
note: when one of the args to genLoadLabelArguments is a jump to a label,
it will call back into genLoad and arrive at this case, which will then set ctx1.bb.ignore to true,
this is okay, since we're jumping unconditionally, so the loads and jumps emitted by the outer
call to genLoad (by calling genLoadLabelArguments and emitOnly) can safely be ignored,
however, as emitOnly will close the block, which reverses its instructions (when it's still open),
we better not reverse when the block has already been closed but is in ignore mode
(if it's not in ignore mode, double-closing is an error)
@dragos figured it out, all I did was write the comment and the `if`
test case to repro basic blocks crasher
the tailcall in the forward jump `matchEnd3(_test(Test.this, false))`
in the following program crashes the back-end (error below)
@scala.annotation.tailrec final def test(meh: Boolean): Boolean = {
<synthetic> val _$this: Test.type = Test.this;
_test(_$this,meh){
case <synthetic> val x1: Some[String] = new Some[String]("a");
case3(){
matchEnd2({
case <synthetic> val x1: Some[String] = x1;
case4(){
if (x1.ne(null))
matchEnd3(if (meh)
_test(Test.this, false)
else
false)
else
case5()
};
case5(){
matchEnd3(_test(Test.this, false))
};
matchEnd3(x){
x
}
})
};
matchEnd2(x){
x
}
}
};
The last instruction (of basic block 11) is not a control flow instruction: CONSTANT(false)
// methods
def test(meh: Boolean (BOOL)): Boolean {
locals: value meh, value _$this, value x1, value x, value x, value x1
startBlock: 1
blocks: [1,2,3,4,5,6,7,8,9,10,11,12,13]
1:
4 JUMP 2
2:
5 NEW REF(class Some)
5 DUP(REF(class Some))
5 CONSTANT("a")
5 CALL_METHOD scala.Some.<init> (static-instance)
5 STORE_LOCAL(value x1)
5 SCOPE_ENTER value x1
5 JUMP 3
3:
5 LOAD_LOCAL(value x1)
7 STORE_LOCAL(value x1)
7 SCOPE_ENTER value x1
7 JUMP 4
4:
7 LOAD_LOCAL(value x1)
7 CZJUMP (REF(class Object))NE ? 5 : 6
5:
8 LOAD_LOCAL(value meh)
8 CZJUMP (BOOL)NE ? 8 : 9
6:
? JUMP 11
7:
7 DROP BOOL
7 JUMP 11
8:
8 CONSTANT(false)
8 STORE_LOCAL(value meh)
8 JUMP 2
9:
8 CONSTANT(false)
8 JUMP 10
10:
8 STORE_LOCAL(value x)
8 JUMP 12
11:
9 JUMP 2
9 STORE_LOCAL(value meh)
9 CONSTANT(false)
12:
7 LOAD_LOCAL(value x)
7 SCOPE_EXIT value x1
7 STORE_LOCAL(value x)
7 JUMP 13
13:
5 LOAD_LOCAL(value x)
5 SCOPE_EXIT value x1
5 RETURN(BOOL)
|
| | |
| | |
| | |
| | |
| | | |
the following commit deals with the fall-out in basicblocks
(double closing of blocks in ignore mode)
|
| | | |
|
| |/
| |
| |
| |
| |
| |
| |
| |
| |
| | |
need to set up symbols with approximated info for the method and its owner
before typing the method's body; when the body has been typed,
we know what its result type will be, so update info of class and method to reflect that
better detection of synthesized matches: annotate selector rather than relying on symbol info
encapsulate CASE | SYNTHETIC flags (setting and querying)
|
| |
| |
| |
| | |
Review by: @jsuereth
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
All replacement methods use dolar signs to identify groups in the
matched string, and backslashes to escape characters. Document this
behavior, and provide a method that can be used to properly quote
replacement strings when this behavior is not desired.
Closes SI-4750.
|
|/
|
|
|
|
|
|
|
|
|
| |
This adds findAllMatchIn to Regex to mirror other similar methods.
It also overloads StringLike's "r", adding a version that accepts
group names.
It includes test cases for both methods.
Closes SI-2460.
|
|
|
|
|
| |
reverts 3db29dde051614d976bca92a1cdeb109c9c0ab01 and 5af2bf54d21ac63236cd6e68586b2c38fa0f28c3
restores 19a48510c2e18430a35319c04dfe3bad7119f23f
|
|
|
|
| |
This reverts commit 19a48510c2e18430a35319c04dfe3bad7119f23f.
|
|\ \ \ \ \
| | | | | |
| | | | | |
| | | | | | |
'VladUreche/issue/5593', 'dragos/master', 'VladUreche/issue/5599', 'adriaanm/ticket/treeannot' and 'heathermiller/issue/5291' into develop
|
| | | |_|/
| | |/| |
| | | | |
| | | | |
| | | | | |
Makes the full signature of a use case member available, but
still hidden under a drop-down menu.
|
| | | |/
| | |/|
| | | |
| | | |
| | | |
| | | | |
not tested for performance, but shouldn't add any memory overhead
see https://github.com/adriaanm/scala/commit/44362c557f for an example usage
|
| | |/
| | |
| | |
| | | |
Eliminated the spurious warnings and made the code easier to follow.
|
| | |\ |
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Initially motivated by SI-5580, then just motivated. I broke up
the opaquely named JavaConversions and JavaConverters into the following
traits encapsulating some permutation of
{ to java, to scala, bidirectional }
{ wrappers, decorators }
I named everything consistently in terms of either Wrappers
or Decorators. Decorators install those asJava/asScala methods
onto collections of the right kind; Wrappers hide the process.
JavaConversions then reduces to an object which (ill-advisedly)
extends both WrapAsJava and WrapAsScala. And JavaConverters is
an object extending DecorateAsScala and DecorateAsJava. However
other more clearly named vals exist in the newly created
scala.collection.convert package object.
val decorateAsJava = new DecorateAsJava { }
val decorateAsScala = new DecorateAsScala { }
val decorateAll = new DecorateAsJava with DecorateAsScala { }
val wrapAsJava = new WrapAsJava { }
val wrapAsScala = new WrapAsScala { }
val wrapAll = new WrapAsJava with WrapAsScala { }
So for instance to import asScala decorators, and only those:
scala> import scala.collection.convert.decorateAsScala._
import scala.collection.convert.decorateAsScala._
scala> new java.util.ArrayList[String].asScala groupBy (x => x)
res0: scala.collection.immutable.Map[String,scala.collection.mutable.Buffer[String]] = Map()
I propose we put those vals or a subset of them in the scala
package object rather than way down in scala.collection.convert.
|
| | | |
| | | |
| | | |
| | | |
| | | | |
Lub explosions mean we can't expect sane types in error messages.
Defend against the insane.
|
| | | |
| | | |
| | | |
| | | |
| | | | |
Tried to paint a picture of how one might synthesize an implicit
method to accompany an implicit class.
|
| | | | |
|
| | |/ |
|
| |/
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
After hours of debugging and understanding all the root causes why
scaladoc is rummaging through the current directory for classes, I
decided to treat the symptoms rather than the cause. I'll document
the decision here in case anyone is interested:
- in order to find documentable entities, scaladoc exhaustively
looks at all the symbols starting from the root package and going
all the way to the leaves, visiting all classes in the classpath
- if the classpath is empty, scaladoc and scalac add "." as the
default classpath. This leads to scaladoc going on a class hunting
spree in the current directory. If it happens that the current
directory contains package objects and they're not in the expected
path (like scala.tools.nsc in build/quick/classes/.../nsc) errors
start to pop up
- scalac is also affected by this, but to a more limited scope,
since it will lazily expand packages whenever necessary. To show
this, run the repl in the root directory of the scala source code,
after compiling scala:
scala> import build.quick.classes.compiler.scala.tools.nsc._
error: error while loading package, class file './build/quick/classes/
compiler/scala/tools/nsc/package.class' contains wrong package package
The cleanest solution I can see now is to look at the tree resulting
from compilation and use it for guidance -- I will look into this.
|
| |
| |
| |
| | |
Feel the bugs falling away.
|
| |
| |
| |
| | |
This reverts commit d8ba5d091e5641553b438ef9930a6023a2709dcd.
|
| |
| |
| |
| |
| |
| |
| | |
Pattern matcher! Totally unrelated to irrefutability, the pattern matcher
at some point stopped sorting its lookup switch cases, and the butterfly's
wings flapped enough to swap two cases. Now they're sorted in ascending
order like they're supposed to be.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Fewer deprecation warnings, prettier trees, prettier
symbols, more polished error messages.
Oh the interesting people you meet handling warnings, I
feel sorry for you all that I get to do it all the time.
One of the characters I met invited me into the "Dead Code
Society" and that's what I'm doing on Tuesdays now. No of
course you haven't, it's a SECRET society.
|
| |
| |
| |
| | |
Concluded String->String would be more useful more often.
|
| |
| |
| |
| |
| | |
Temporary reversion of irrefutability commits in interests
of stable milestone. Expect to restore shortly.
|
| |\ \
| | | |
| | | |
| | | | |
'adriaanm/topic/virtpatmat' and 'vjovanov/actor-hierarchy' into develop
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Internal nodes added so methods relevant to akka can be overridden.
Review by: @phaller
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
this way we don't end up with fresh, incompatible, skolems for every branch, but a neatly packed existential
you get much nicer lubs this way -- threw in a fast path (I hope) that avoids lubbing when unnecessary (same types)
since lub isn't robust enough to deal with the result of packedType,
and we only really need the trivial case (S =:= T), special case it for now
... don't pack in ifs after typer, also only when virtPatmat
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
no need for the clunky repackExistential
pack the type of each case instead, since the skolems we've created shouldn't last beyond the case anyway
this way we don't end up with fresh, incompatible, skolems for every case, but a neatly packed existential
|