| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Tree reification fails for ExistentialTypeTree. The reason is that the
tree passed for reification (see reifyTree at GetTrees.scala) must have
not null tpe (see reifyBoundType at GenTrees.scala), which is not true
in the case of ExistentialTypeTree.
Why is it so? The tree passed to reifyTree was obtained in the reshape
phase of reificationusing using original TypeTrees that reporesent pre-
typer representation of a type. The problem is that original's tpe for
ExistentialTypeTree is not set.
So the solution to the issue is to create ExistentialTypeTree's original
in a such way that is has actual tpe set.
|
|\
| |
| | |
SI-7321 Memory leak in specialize on multiple compiler runs.
|
| |
| |
| |
| |
| |
| |
| |
| | |
Currently the map is declared LinkedHashMap, which doesn't align with
other caching maps that are cleared on every run. Linkedness is only
needed to ensure deterministic order on the generated specialized
classes. The same can be accomplished by sorting generated classes
a-posteriori.
|
|\ \
| |/
|/| |
Make build even more regular
|
| | |
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
To use Zinc with the ant build:
- install zinc and symlink the installed zinc script to ${basedir}/tools/zinc
(${basedir} is where build.xml and the rest of your checkout resides)
- make sure to set ZINC_OPTS to match ANT_OPTS!
- invoke ant as `ant -Dstarr.version="2.10.1" -Dlocker.skip=1`
(zinc does not work if locker is only classfiles, needs jars
TODO rework the build to pack locker and build using that when using zinc?)
Mostly to enable dog fooding of incremental compilation work for now.
|
| |
| |
| |
| |
| |
| |
| | |
Untangle actors, msil builds from library/compiler builds.
For further regularity, always build java files.
TODO: update IDE projects
|
| | |
|
|\ \
| | |
| | | |
SI-6900 Tail call elimination + dependent method type crasher
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Uncurry's info transformer could generate a MethodType
with cloned parameter symbols. This type was used
for the LabelDef generated in the TailCalls phase.
But, the RHS of the method still contains types that
refer to the original parmameter symbol. Spurious
type errors ensued.
I've spent a good chunk of time pursuing a more
principled fix, in which we keep the symbols in the
tree in sync with those in the MethodType. You can
relive the procession of false dawns:
https://github.com/scala/scala/pull/2248
Ultimately that scheme was derailed by a mismatch
between the type parameter `T` and the skolem `T&`
in the example below.
trait Endo[A] { def apply(a: => A): A }
class Test {
def foo[T] =
new Endo[(T, Unit)] {
def apply(v1: => (T, Unit)) = v1 // no bridge created
}
}
Interestingly, by removing the caching in SingleType,
I got past that problem. But I didn't characterize it
further.
This commit sets asides the noble goal of operating in
the world of types, and sledgehammers past the crash by
casting the arguments to and the result of the label jump
generated in TailCalls.
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Now, the InfoTransformer is responsible for erasing the
path dependent types of formal parameters, at the same
place where it flattens nested method types.
This is preferable to having the tree transformer overwrite
the result of the info transformer, as used to be the case
after my previous work on SI-6135 / 493197fc.
|
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
For imminent reuse in the subsequent commit.
The binary compatibility whitelist is updated to
ignore these, as they live in reflect.internal.
|
| | | |
|
|\ \ \
| | | |
| | | | |
Take the N^2 out of the compiler's TreeSet.
|
|/ / /
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
The code responsible for this performance bug lives on somewhere
in the combination of Iterator.single and Iterator.++ and is
tracked by SI-7316. What this commit does is bypass the creation
and composition of iterators entirely in favor of applying foreach
to walking the tree.
The important lesson of a bug like this: the occurrence depends on
the existence of multiple implementations of basic structures like
Trees. For each redundant implementation, scrutiny and testing
are divided and "bug diversity" is increased. We should labor
hard to structure collections in such a way that people have no
good reason not to take advantage of the basic and hopefully
battle-tested logic - especially when those people are us.
I hope to remove util.TreeSet entirely. Until then, here is the
impact of this commit on the time to compile a piece of generated
test code.
% time scalac3 ./target/generated/src.scala
Mar 28 13:20:31 [running phase parser on src.scala]
...
Mar 28 13:21:28 [running phase lazyvals on src.scala]
Mar 28 13:21:28 [running phase lambdalift on src.scala] <-- WHOA
Mar 28 13:25:05 [running phase constructors on src.scala]
...
Mar 28 13:25:19 [running phase jvm on icode]
316.387 real, 438.182 user, 8.163 sys
To this:
97.927 real, 211.015 user, 8.043 sys
% time pscalac ./target/generated/src.scala
Mar 28 13:18:47 [running phase parser on src.scala]
...
Mar 28 13:19:44 [running phase lazyvals on src.scala]
Mar 28 13:19:44 [running phase lambdalift on src.scala]
Mar 28 13:19:46 [running phase constructors on src.scala]
...
Mar 28 13:19:57 [running phase jvm on icode]
99.909 real, 223.605 user, 7.847 sys
That's lambdalift dropping from 217 seconds to 2 seconds.
|
|\ \ \
| |/ /
|/| | |
SI-7147 Diagnostic for unexplained assertion in presentation compiler.
|
| |/
| |
| |
| |
| | |
We don't have a reproducible test for this, so the best we can
do is beef up the assertion to shine a little light on the problem.
|
|\ \
| | |
| | | |
SI-6793 Don't use super param accessors if inaccessible.
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
"Alias replacement" has been with us since 2005 (13c59adf9).
Given:
package a {
class C1(val v0: String)
class C2(v1: String) extends a.C1(v1) { v1 }
}
The reference to `v1` is rewritten as `C2.super.v0()`, and
no field is generated in `C2`.
(Oddly, this optimization doesn't seem to kick in if
these classes are in the empty package. That's probably
a distinct bug.)
However, this rewriting is done without consideration of
the accessibility of `v0` from `C2`.
This commit disables this optimization if there if `v0` is
not accessible.
|
|\ \ \
| | | |
| | | | |
SI-6715 Scaladoc: Use an encoded name if a decoded name is ""
|
| | | | |
|
|/ / /
| | |
| | |
| | |
| | | |
Because this implementation is more clear than 2.10.x's and it will
simplify a further commit to fix SI-6715.
|
|\ \ \
| |_|/
|/| | |
Correct sorting example for Ordering in scaladoc
|
| |/
| |
| |
| |
| |
| |
| |
| | |
Below code snippet,
Sorting.quickSort(pairs)(Ordering.by[(String, Int, Int), Int](_._2)
should be
Sorting.quickSort(pairs)(Ordering.by[(String, Int, Int), Int](_._2))
|
|\ \
| | |
| | | |
Scaladoc: Load scripts at the bottom, and with a defer attribute
|
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
To improve latency on modern browsers (which supports defer) and old browsers:
* https://www.webkit.org/blog/1395/running-scripts-in-webkit/
* http://developer.yahoo.com/blogs/ydn/posts/2007/07/high_performanc_5/
|
|\ \ \
| | | |
| | | | |
Use java-diff-utils for diffing in partest.
|
| | | | |
|
|/ / /
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
We now use the unified diff format, hence the updated check files.
It's not clear to me how partest's classpath is managed,
but the approach in this commit works for the ant task and script invocation.
The diffutils jar is injected in the parent classloader.
|
|\ \ \
| | | |
| | | | |
further tweaks to the ant build
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Target pack.@{project} is considered up to date iff
classes emitted by quick.@{project} are.
TODO: encapsulate "classes emitted by quick.@{project}".
For now, they are duplicated as references
to `${build-@{stage}.dir}/classes/@{destproject}`
(in staged-scalac and *.build.path)
|
| | | | |
|
| | | | |
|
| | | | |
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Added starr.number, so that `ant -propertyfile starr.number -Dlocker.skip=1`
works across branches.
Introduced target `test.core`, which skips `test.stability`, since that requires
`locker == quick`, whereas `locker == starr` with the above settings.
(It's interesting to see for which files it fails when that assumption is broken.)
Stability will still be tested on a nightly basis. It's rarely broken.
|
| | | | |
|
| | | | |
|
| | | | |
|
| | | |
| | | |
| | | |
| | | | |
also added test.bc-opt target as binary compatibility differs between (non)optimized builds...
|
| |/ / |
|
|\ \ \
| | | |
| | | | |
SI-7285 Fix match analysis with nested objects
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
The fix for SI-6146 introduced `nestedMemberType` to
enumerate sealed subtypes based on the (prefixed) type
of the scrutinee and the symbols of its sealed subclasses.
That method needed to widen `ThisType(modSym)`s to
`ModuleTypeRef(modSym)` before calling `asSeenFrom`.
However, this could lead to confused in the match analysis,
which sees `ModuleTypeRef` as distinct from singleton types
on the same modules (after all, they aren't =:=). Spurious
warnings ensued.
This commit makes two changes:
- conditionally re-narrow the result of `asSeenFrom` in `nestedMemberType`.
- present `a.b.SomeModule.type` as `SomeModule` in warnings emitted
by the pattern matcher.
|
| | | | |
|
|\ \ \ \
| | | | |
| | | | | |
SI-7290 Discard duplicates in switchable alternative patterns.
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
- make a def a val, we only need to compute it once
- add a clarifying comment
- only report the first duplicate
|
| |/ / /
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
The pattern matcher must not allow duplicates to hit the
backend when generating switches. It already eliminates then
if they appear on different cases (with an unreachability warning.)
This commit does the same for duplicated literal patterns in an
alternative pattern: discard and warn.
|
|\ \ \ \
| |_|/ /
|/| | | |
SI-6387 Clones accessor before name expansion
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
When a symbol's name is expanded due to a conflict during
composition (e.g. multiple traits with same-named members, but
which are not both visible at the language level in the concrete
class) the compiler renames some symbols with expanded names which
embed the full name of the declaring class to avoid clashes.
In the rare cases when the accessor overrides the member in base
class, such expansion either results in AbstractMethodError when
the base method is abstract, or, even worse, can change the
semantics of the program.
To avoid such issues, we clone the accessor symbol, clear its
ACCESSOR flag and enter the symbol with an unchanged name.
|
|\ \ \ \
| | | | |
| | | | | |
[backport] SI-7237 Always choose ForkJoinTaskSupport
|