Commits on Mar 13, 2013
  1. Merge pull request #352 from dotta/backport/3.0.x/issue/NPE-on-variab…

    dotta committed Mar 13, 2013
    [backport] Workaround for NPE in debugger variable view when using Eclipse Juno
  2. Workaround for NPE in debugger variable view when using Eclipse Juno

    dotta committed Mar 12, 2013
    Unfortunately, the way we keep track of the currently selected thread in the
    Scala debugger isn't robust enough. Currently, we rely on a _selectionChanged_
    event to be sent when the Debug Perspective is opened. This worked fine with
    Eclipse Indigo, but it's no longer working with Eclipse Juno.
    The workaround is to look for the first suspended thread if
    ``ScalaDebugger.currentThread`` is null. This is far from being a proper fix,
    but it doesn't look like we have many alternatives. The Java debugger
    implementation seem to be using a DebugContextProvider/Manager/Event/Listener
    to know the thread associated to each variable. But using the debug context
    classes would require some import work, so it would be preferable to delay this
    for after the 3.0.0 final release.
    The workaround was suggested by @skyluc, and he believes the only issue this
    could lead to is visibility issues due to invoking a method call on the wrong
    thread (in case more than one is suspended), and hence the executing thread may
    see some stale values. Not an issue to be taken lightly.
    Fix #1001585
    backport to _release/3.0.x_
    (cherry picked from commit 2de89fe)
Commits on Mar 12, 2013
  1. Merge pull request #350 from dotta/backport/3.0.x/issue/ISE-during-co…

    dotta committed Mar 12, 2013
    [backport] Don't add arguments templates for parameterless method's completion
  2. Don't add arguments templates for parameterless method's completion

    dotta committed Mar 12, 2013
    When asking for completion on a parameterless method we were trying to install
    arguments' templates. This operation in not allowed on parameterless methods,
    and it caused a ``java.lang.IllegalStateException: must specify at least one
    linked position`` during completion.
    ``explicitParamNames`` is a ``List[List[String]]`` (because of currying), hence
    a parameterless method is represented as ``List(List())``. The arguments'
    templates where added only if ``explicitParamNames`` is not empty, however a
    ``List(List())`` contains exactly one element (the empty list), hence it's
    never empty!  The fix is really simple, we just need to flatten the
    ``explicitParamNames`` before checking for emptiness.
    Fixes #1001591
    backport to _release/3.0.x_
    review by @dragos
    (cherry picked from commit d08e999)
Commits on Mar 11, 2013
  1. Merge pull request #348 from dotta/backport/issue/flaky-presentation-…

    dotta committed Mar 11, 2013
    [backport] Disabled ``implicitConversionFromPackageObjectShouldBeInScope`` PC test
  2. Disabled ``implicitConversionFromPackageObjectShouldBeInScope`` PC test

    dotta committed Mar 11, 2013
    Disabling the test ``implicitConversionFromPackageObjectShouldBeInScope`` from
    PresentationCompilerTest because it often fails on our Jenkins machine (and
    that's just plain annoying), e.g.,
    This test has been flaky for at least a month now.
    Re #1001588
    (cherry picked from commit 02fbfe2)
  3. Merge pull request #346 from dotta/backport/3.0.x/issue/expanding-var…

    dotta committed Mar 11, 2013
    [backport] Expanding variable in debugger resulted in NPE
  4. Expanding variable in debugger resulted in NPE

    dotta committed Mar 11, 2013
    The regression was introduced by SHA: be83fc8.
    The reason for the regression is that the name of the class
    ``ScalaLogicalStructureProvider`` was inadvertently changed to
    ``ScalaLogicalStructureProviders`` (note the final **s**). Since that class was
    linked to the *org.eclipse.debug.core.logicalStructureProviders* extension
    point, as soon as the debugger code needed to load that class an exception is
    What we should do to avoid this sort of errors in the future is setting the PDE
    compiler flag for "References to non-existing classes" to *Error*. Furthermore,
    it would be good if this check could be enforced in our Tycho build (I've asked
    in the tycho-user ML,
    [here]( is the
    link to the discussion).
    Fix #1001586
    Backport to _release/3.0.x_
    (cherry picked from commit ca3a67d)
Commits on Mar 8, 2013
  1. Return `OK_STATUS` from the semantic highlighting job when the editor…

    dragos committed Mar 8, 2013
    … is dirty.
    Previously, we announced that the job finishes asynchronously (`ASYNC_FINISH`),
    but forgot to call `done` if the editor was dirty (and no UI asyncExec was
    performed at all). This lead in turn to the job manager thinking that the job
    is still running and not scheduling another run, ever, meaning the loss of
    semantic highlighting.
    Fixed #1001536.(cherry picked from commit 5e51152)
Commits on Mar 6, 2013
  1. Merge pull request #340 from dotta/backport/3.0.x/issue/handling-debu…

    dotta committed Mar 6, 2013
    [backport] Comply to the debugger interfaces by wrapping JDI runtime exceptions
  2. Comply to the debugger interfaces by wrapping JDI runtime exceptions

    dotta committed Mar 6, 2013
    Several of the debugger interfaces we implement expect a ``DebugException`` to
    be thrown when the called method fails to execute. Failure can occur, for
    instance, when a debugger session is terminated by the user. Failing to wrap
    the JDI runtime exception in a ``DebugException`` can prevent the debugger to
    correctly work and the user is sometime forced to restart Eclipse.
    ``DebugException`` is a platform exception that is specially treated by
    If you wonder why do we have calls to ``safeStackFrameCalls`` in, for instance,
    ``ScalaStackFrame.getLineNumber``, the answer is: I don't think we need it. However,
    since we are in RCs cycle for the 3.0 release, I want to minimize the changes in
    this commit to avoid regressions.
    Fix #1001531
    (cherry picked from commit 60df6af)
  3. Lazy retrieval of Java parameter names in completions.

    dragos committed Mar 4, 2013
    When a completion is selected we need to show the parameter names as placeholders.
    If the method comes from Java, we need to retrieve the Java element, a potentially
    long-running operation that can trigger the structure builder. Instead of retrieving
    all parameter names eagerly, for each completion proposal, we delay it to the
    moment a completion is selected and inserted in the editor.
    Fixed #1001560, #1001497.(cherry picked from commit ffcf62a)
Commits on Feb 25, 2013
  1. Merge pull request #333 from dotta/backport/3.0.0/cannot-enable-conti…

    dotta committed Feb 25, 2013
    Backport/3.0.0/cannot enable continuations plugin 1001030
  2. Testing presentation compiler when project depends on continuations.jar

    dotta committed Feb 25, 2013
    Added a couple of tests for making sure that the build compiler and the
    presentation compiler behave the same when compiling projects that depend on
    the continuations plugin.
    (cherry picked from commit 514c2a5)
  3. Do not force default location of -XpluginsDir in Scala Compiler Prefe…

    dotta committed Feb 24, 2013
    Trying to force the default location to the plugins directory containing the
    continuations.jar plugin turned out to be a very bad idea (this was done as
    part of commit SHA-cbfd82d874f998b7726fec895ad4d83194ad1e5b).
    The issue is that the default location where the continuations.jar is stored
    changes *each time a user updates the Scala IDE* and, when updating, the
    *former location is deleted*. The net effect of this is that after upgrading,
    all projects using the continuations plugin would report the following error
    	`bad option: -P:continuations:enable` is reported
    Simply because the continuations.jar can no longer be located.
    The fix is to somewhat restore the former implementation wrt how we handle the
    `-Xpluginsdir` compiler setting. Basically, the solution consists in not
    showing the default value of `-Xpluginsdir` to the user in the Preferences
    dialog, and pass this default to the different compiler instance we create
    unless the user specifices a different location (in which case, we do use the
    location provided by the user).
    There is one more catch due to the current way `ScalaProject.scalacArguments`
    is implemented (the method is used to pass the compiler arguments to the Sbt
    builder). In short, the returned arguments are only the ones forces by the user
    via the Preference dialog. Since the `-Xpluginsdir` default directory is
    **not** shown in the Preference dialog, the Sbt builder would fail to compiler
    projects with continuations enabled because it wouldn't know where to look for
    the continuations.jar. Hence, I had to add some ad-hoc code to handle this
    correctly in `ScalaProject.scalacArguments`.
    This whole fix is far from ideal, but I want to keep the changeset as small as
    possible to avoid regressions, since this commit should be included in the
    upcoming 3.0.0 release.
    By the way, all existing projects that were affected by the error should
    restore the defaults (Preferences > Scala > Compiler and hit the Restore
    Defaults button).
    Fixes #1001030, #1000700
    (cherry picked from commit aa2ecb9)
  4. Merge pull request #330 from dragos/isssue/tasks-disappear-1001401

    dragos committed Feb 25, 2013
    Use Scala specific task markers.(cherry picked from commit f248290)
Commits on Feb 23, 2013
  1. Uses a specific scala version for the version updater

    skyluc committed Feb 22, 2013
    (cherry picked from commit 24a421a)
Commits on Feb 22, 2013
  1. More robust against VM exceptions.

    dragos committed Feb 21, 2013
    * clearly report an actor shutting down because of an uncaught exception
      using the Eclipse Error log
    * gracefully handle the common VM exceptions.
    This is not exhaustive (a handful of FIXME notes are still in the codebase),
    but it is a step in the right direction and fixes a number of issues I've
    been seeing lately.
    Fixed #1001328, Refs #1001487.(cherry picked from commit a07bd6b)
  2. Possible deadlock fix.

    dragos committed Feb 14, 2013
    Two actors sending each-other synchronous messages
    may deadlock. Break the cycle by making one of them
    The cycle is:
    * a classPrepare event is fired, and listeners are
    * a breakpoint is deleted, and the breakpoint support
      actor de-registers from load events.
    I couldn't write a test to exercise this scenario,
    but I think it's worth discussing and see if we can
    ship it as is.
    I moved most sync message sends to `send with timeout`.
    Two sync message sends are still in the code base:
    one for debugging and one for initialising the
    breakpoint manager.
    Fixed #1001512(cherry picked from commit 4fc53e3)
  3. Implement 'Search test methods' in the Scala JUnit4 test runner.

    dragos committed Feb 12, 2013
    Hook into the JUnit test configuration tab and implement the
    Search Methods button functionality. We use aspects to intercept
    the right call, and route it to our `ITestKind` implementation.
    Fixed #1001474(cherry picked from commit c390cee)
Commits on Feb 21, 2013
  1. Merge pull request #322 from dragos/issue/revert-parse-tree

    huitseeker committed Feb 21, 2013
    Revert "Merge pull request #269 from mads379/parsetree-1001326"
  2. Merge pull request #325 from dotta/backport/3.0.x/project-settings-sh…

    dotta committed Feb 21, 2013
    [backport] Clean projects in a background job when project settings' change
  3. Merge pull request #324 from dotta/backport/3.0.x/dont-use-annotation…

    dotta committed Feb 21, 2013
    [backport] issue/dont-use-annotation-model-for-semantic-highlighting-1001156
  4. Clean projects in a background job when project settings' change

    dotta committed Feb 20, 2013
    Cleaning the projects in the UI Thread was causing UI freeze. The
    fix is really simple: clean the projects in a background job.
    Note: The implemented solution is inspired by the Eclipse
          ``CleanDialog`` component.
    Fixes #1001527
    (cherry picked from commit aec8edd)
  5. Deprecated ``SemanticHighlightingReconciliationPartecipant``s

    dotta committed Feb 1, 2013
    (cherry picked from commit 87e20d5)
  6. Java Semantic Highlighting can be disabled without using aspects

    dotta committed Feb 1, 2013
    (cherry picked from commit 05956d4)
  7. Disabled Java semantic highlighting in Scala source attachments

    dotta committed Feb 19, 2013
    Preventing the Java semantic highlighting engine to be installed
    on Scala sources attachments.
    By doing so, ``ScalaPresentationReconciler`` can now inherit from
    (cherry picked from commit 4ba2ba9)
  8. Faster and slicker Semantic Highlighting

    dotta committed Feb 1, 2013
    The former implementation of semantic highlighting was slow becuase of the
    following two reasons:
    First, the annotations created by the semantic highlighting component, where
    invalidated and recomputed every time the compilation unit was reconciled.
    Second, for the current implementation of the AnnotationModel, adding/removing
    an annotation is O(n^ 2). (The quadratic behavior is due to the implementation
    of ``CompilationUnitDocumentProvider#ReverseMap``, which is used by
    Basically, each time an annotation is added/removed, the ``ReverseMap`` class -
    which is backed by a linked list - is accessed to verify if the element exist.)
    The proposed solution is to apply semantic colorings directly on the editor's
    ``TextPresentation`` and hence no longer use the ``AnnotationModel``. This
    approach is both lightweight and orders of magnitude more efficient than the
    The way semantic highlighting works now is relatively simple. At a high level,
    there are only three key elements:
    * ``Presenter$SemanticHighlightingJob``: Given a compilation unit, it
      classifyies and collects all positions that need to be colored in the editor.
    * ``PositionsTracker``: Keeps track of currently highlighted positions in the
    * ``TextPresentationEditorHighlighter$ApplyHighlightingTextPresentationChange``:
      Converts them into   ``StyleRange``s, and applies the styles to the editor's
    Of course the actual implementation contains a bit more than just the above
    mentioned classes, but most of it is needed to correctly handle platform's
    events such as the swapping of the document input (see
    There is however one important optimization that is worth mentioning, which is
    done to reduce to a minimum the editor's region whose styles need to be
    refreshed. Basically, whenever the user is typing in the editor, all positions
    that are **after** the cursor should be invalidated. If you are editing the
    beginning of a big file, almost the whole editor's text presentation would need
    to be refreshed. Furthermore, the mentioned approach would cause colors to
    flicker because all highlighted positions **after** the edit would be deleted
    (and hence the colors would be removed), and the new positions would get
    colored only **after** the semantic highlighting job completes. This is clearly
    not ideal, and it turns out it can be easily avoided by registering a
    ``IPositionUpdater`` on the editor's document. The concrete implementation of
    the listener (i.e., ``Presenter$PositionUpdater``) takes care of shifting -
    through side-effect - all currently highlighted positions that are **after**
    the edited region. The net effect is that colors never flicker.
    Bottom line, with the current implementation semantic highlighting is:
    * Considerably faster.
    * Never flickers.
    * Testable.
    If you wonder why the semantically highlighted positions are not stored in the
    document's model, the answer is that I simply could not see any benefit of
    doing so (why? because the document synchronizes access to the underlying
    position's map when adding/removing positions).  Furthermore, it is really easy
    to add the missing logic for storing the positions in the document's model, if
    in the future it turns out we need/want to do so.
    Fix #1001156
    Other facts about this commit:
    * Added the needed infrastructure for registering reconciling listeners (i.e.,
      ``IJavaReconcilingListener``) to the editor.
    * Created a ``UiThread`` trait needed for testing code that needs to be run
      within the UI Thread.
    Other tickets fixed by this commit:
    * Disabling semantic highlighting in the preferences now refreshes currently
      opened editors. Fixes #1001507
    * Changing semantic highlighting style's preferences now refreshes the
      currently opened editors. Fixes #1001508
    * Cache highlighting styles instead of creating the during
      This optimization has major consequence on the user's experience. I should also
      point out that ``ScalaSyntaxClass`` and ``SymbolTypes``  are quite messy and
      should be improved. However, I think this should be tackled in a different PR.
      Fixes #1001493, #1001489.
    (cherry picked from commit 9bfeb02)
  9. Uses sbinary '-pretending-SNAPSHOT' version

    skyluc committed Feb 18, 2013
    (cherry picked from commit b33b664)
Commits on Feb 20, 2013
  1. Revert "Merge pull request #269 from mads379/parsetree-1001326"

    dragos committed Feb 20, 2013
    The parse-tree changes have been merged in Scala [pull 1982](scala/scala#1982),
    so there is no need for it anymore. Moreover, the presentation compiler must *not* be interrupted
    while parsing on a foreign thread, and that can't be ensured in this commit (but is correctly ensured
    in the Scala changeset). Since this branch is used with 2.9.3, there is a potential for race conditions.
    This reverts commit 712e6bf, reversing
    changes made to 4f7b391.
  2. Make stepping as fast as in the JDT debugger.

    dragos committed Feb 15, 2013
    Two performance improvements in the way we handle stack frames:
    * make `variables` lazy. This way we don't retrieve variables values
      unless they are needed (most of the times, they are needed for only one
      frame in a full thread stack). Retrieving values requires a roundtrip
      to the debugged VM, hitting the network stack.
    * don't fire DebugEvents for creation of stack frames. This is the
      time consuming part: each frame (can be hundreds) fires one CREATE
      event that is put in a queue and a job-dispatch is scheduled to run for
      *each* frame in part. This floods the event queue and leads to
      lag times in the order of seconds (depending on the stack depth).
    I checked (in the debugger) that the JDT debugger does not fire CREATE
    events for stack frames, so I'm suggesting we do the same thing.(cherry picked from commit 96de8f0)
Commits on Feb 19, 2013
  1. Adds an extractor to check incompatible versions

    skyluc committed Feb 11, 2013
    The code is more self-contained.
  2. Adds a preference for the classpath validator

    skyluc committed Feb 8, 2013
    The preference allows to disable the search for jars not compatible with the current version of Scala.
    Limits the check to possible version of Scala >=2.8.0.
    Fixes #1001482
Commits on Feb 13, 2013
  1. Initialize symbols for primitive types on compiler startup.

    dragos committed Feb 11, 2013
    The nightly build uncovered a race condition in the indexer, that stepped on
    a case where `definitions.ShortClass` was not yet loaded, triggering the
    symbol loader.
    Needs back port. Should fix flaky presentation compiler tests.(cherry picked from commit bed8bfa)