Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
branch: nom
Commits on Apr 27, 2015
  1. @FROGGS
  2. @lizmat
  3. @lizmat
  4. @lizmat
  5. @TimToady
  6. @masak

    create RolePunning.pun method

    masak authored
  7. @masak

    CArray.^parameterize made fresh puns every time

    masak authored
    It shouldn't -- this commit fixes that, with a regression test.
    
    Bug discovered while working commit
    1c916bc. FROGGS++ confirmed that
    the code in CArray was based on Pointer, and provided the failing
    test case (FROGGS++).
  8. @masak

    Pointer.^parameterize made fresh puns every time

    masak authored
    It shouldn't -- this commit fixes that, with a regression test.
    
    Bug discovered by DrForr++. This takes care of RT #124321.
  9. @lizmat
  10. @lizmat
  11. @lizmat

    Move module loading logic to World

    lizmat authored
    Step #1 towards making module loading more S11/S22 compatible.
  12. @TimToady
  13. @TimToady
  14. @TimToady

    min/max/minmax deflatting

    TimToady authored
  15. @TimToady
  16. @TimToady

    remove commented code

    TimToady authored
Commits on Apr 26, 2015
  1. @TimToady
  2. @jnthn
  3. @lizmat

    Merge pull request #411 from skids/xnyi

    lizmat authored
    Add typed exception X::NYI::Available
  4. @lizmat
  5. @peschwa

    Simplifiy JavaHOW.

    peschwa authored
    Instead of reimplementing partial functionality from MethodContainer we might
    as well just C<does> it.
  6. @lizmat

    Add some ChangeLog entries

    lizmat authored
  7. @lizmat
  8. @lizmat
  9. @lizmat

    Merge pull request #371 from paultcochrane/pr/hyphenate_test_functions

    lizmat authored
    Rename test functions to use hyphens instead of underscores
  10. @lizmat
Commits on Apr 25, 2015
  1. @FROGGS
  2. @lizmat
  3. @lizmat
  4. @lizmat
  5. @TimToady

    distinguish join from _join

    TimToady authored
    for now, _join is the old flattening join, until we decide the best notation
  6. @TimToady
  7. @lizmat

    Use eager pick(*) where it makes sense

    lizmat authored
    This makes at least EARLIEST a bit faster, and hopefully more stable.
  8. @TimToady
Commits on Apr 24, 2015
  1. @lizmat

    Make a faster List.pick(*,:eager) version

    lizmat authored
    Well, that was the idea.  Turns out the eager version is about 20x *slower*.
    Added here for further research.
    
    What has been ruled out:
    - anything inside the loop
      That part of the code is mostly the same as the non-eager version.  If
      the code inside the while loop is reduced to just the $elems = $elems - 1,
      even then the eager variant is about 15x slower.
    - something in the MMD dispatch
      If the code inside the "eager" variant is the same as the non-eager variant,
      there is no difference in execution noticeable either.
    - not SPESH related
      With MVM_SPESH_DISABLE=1 the difference in execution is even larger: the
      optimizer did its best!
    
    What *is* different?
    - the number of frames in a profile
      example code: my @a = ^100; for ^500 { @a.pick(*) } # or *, :eager
      non-eager / eager: 10618 / 176614 frames
      So the difference in number of frames is in same order of magnitude as the
      difference in CPU.
    
    So the question is really: why does the eager version create so many frames?
Something went wrong with that request. Please try again.