Skip to content


Subversion checkout URL

You can clone with
Download ZIP
Commits on Nov 20, 2015
  1. @Util

    Prepare for the 7.10.0 release.

    Util committed
Commits on Oct 20, 2015
  1. @rurban

    [config] add a cperldoc probe

    rurban committed
    in case perldoc was not found.
    Note that a perl5.6 perldoc will run properly until 5.18
    but will silently fail starting with 5.20.
    We could add a check for that also eventually.
  2. @rurban

    Update release_manager_guide

    rurban committed
    Move new ChangeLog skeleton down after the release.
    A release ChangeLog should look clean.
    Add link names to the github master links.
  3. @rurban

    Prepare new ChangeLog for 7.10

    rurban committed
  4. @rurban

    Supported Release 7.9.0 - "Sangihe hanging parrot"

    rurban committed
    - Core
        + StringBuilder.clone and ResizablePMCArray write barrier optimizations.
          1.14% faster, GH #1123
        + More StringBuilder optimizations with substr, set_pmc, push_pmc,
          i_concatenate, and do less overallocation. <1% faster. GH #1123
        + Support -D1 GC stats with --ccflags=-DMEMORY_DEBUG for --gc=ms
        + Added a Parrot_{un,}block_GC_move API for the 2 string compacting
          collectors ms and ms2
        + Deprecated ListChunk GC API and src/list.c GH #1179
        + Fixed a --gc=ms segv in Parrot_io_write_s GH #1196
        + const/ARGIN all read-only IO api arguments
        + Fixed --gc=inf string header allocation and failing tests
    - Build
        + --optimize: Replace -O2 from perl5 with -O3 on clang or gcc. GH #1185
          5% faster
        + Fixed make clean realclean and install which sometimes forgot pmc_default.h
          GH #1194, GH #1206
    - Tests
        + Use the faster travis docker infrastructure without sudo
Commits on Oct 19, 2015
  1. @rurban

    [config] unquote the make clean args

    rurban committed
    We use now /bin/rm for $(RM_F) in the Makefile
    which did not expand "$(INC_PMC_DIR)/pmc_*.h"
    and many more generated files.
    This led to leftovers, confusing make install.
    E.g. the file include/pmc/pmc_default.h was not removed, and so was not rebuilt,
    and so was not added to MANIFEST.generated, so t/distro/manifest_generated.t
    `make clean realclean` does now the same as
    `git clean -dxf`
    Fixes GH #1194 and GH #1206.
  2. @rurban

    [gc] Fixed --gc=inf string header allocation

    rurban committed
    add missing mark_str_header,
    do not use pre-buffer for string headers.
    clear the flags with allocate_string_header
  3. @rurban

    myconfig: add gc_type

    rurban committed
    and re-order it a bit.
  4. @rurban

    api.yaml: add deprecation notice for GH #1179

    rurban committed
    Deprecated ListChunk GC API and src/list.c
    Also update ChangeLog.
  5. @rurban

    [test] TODO fragile t/steps/auto/warnings-01.t

    rurban committed
    testing for -Wunreachable-code with --cage is fragile.
  6. @rurban
  7. @rurban
  8. @rurban

    [gc] Add Parrot_{un,}block_GC_move API #1197

    rurban committed
    and use it in the string_gc.c compaction code and the problem
    in #1196.
    run t/op/gc-non-recursive.t also with debugging, just with less loops
    to catch more gc errors.
    This does not work yet
  9. @rurban

    [gc] fix a MS segv in Parrot_io_write_s

    rurban committed
    When certain functions cause a GC, the local string buffers may get
    moved away, but local STRING* variables or worse, strings inside a
    StringHandle may still point to the old location.
    Lock the GC then for this section, mostly during IO writing to string handles,
    when the handle needs to realloc its buffer (the stringhandle).
    But reading into a buffer is also problematic. we really should know the
    size beforehand.
    One way to fix GH #1196, lock the GC sweep in those cases.
    The other variants would be:
    - lock the whole GC via Parrot_block_GC_mark
    - unset the PObj_is_movable flag in the STRING that it may not be moved,
      as it is still locally referenced.
    Remaining ms test failure: t/pmc/fixedstringarray.t
  10. @rurban

    [pmc] add missing fields to StringHandle.clone

    rurban committed
    read_offset, filename, read_buffer, write_buffer, record_separator
    also need to get cloned.
    Probably related to GH #1196
  11. @rurban
  12. @rurban
  13. @rurban

    Prepare deprecation of the ListChunk GC API, and src/list.c

    rurban committed
    See GH #1179.
    See #define PARROT_BUFFERLIKE_LIST in config.h to use it, still.
    imageiofreeze still uses it.
  14. @rurban
  15. @rurban

    [codingstd] for io changes

    rurban committed
  16. @rurban

    [cage] ARGMOD->ARGIN const io api getters

    rurban committed
    Harmonize ARGMOD vs ARGIN in all IO functions.
    Getters have a ARGIN handle, setters an ARGMOD.
    io.h, latform_interface.h, io/filehandle.c, io/io_private.h,
    io/pipe.c, io/socket.c, io/stringhandle.c, platform/generic/io.c
    Not tested on msvc, most ARGMOD annotations do look wrong and need to be
    improved.  this has only effect on msvc with sal
  17. @rurban
  18. @rurban

    [cage] const io api arguments

    rurban committed
    also add an easier str dumper for -D100 gc tracings,
    currenty only on ms2.
  19. @rurban

    [pmc] rpa gc optimizations

    rurban committed
    Re-enable the status from 6.11.0, not too many write barriers in
    ResizablePMCArray methods. Not in shift as we only move the offset
    and only when needed in resize.
    But it does not give us back the lost 13% in parrot-bench, from 6.11
    to 7.0.
  20. @rurban

    [test] un-todo t/src/extend.t 19 on darwin

    rurban committed
    not repro anymore, at least with yosemite.
    GH #856 SIGABRT with darwin threads
  21. @rurban

    [core] optimize io_stringhandle_write_b

    rurban committed
    Reuse write buffer if large enough. This happens very often.
    But surpringly no perf. improvement measurable.
  22. @rurban

    [pmc] Avoid StringBuilder copies on set_pmc, ...

    rurban committed
    and push_pmc, i_concatenate, when the arg is a StringBuilder
    PMC already. 0.5% faster.
  23. @rurban

    [pmc] StringBuilder.substr needs no WB

    rurban committed
    GH #1123. 1.14% faster
  24. @rurban

    [pmc] disable 2x StringBuilder.substr clone, GH #1123

    rurban committed
    STRING_substr already creates a copy for us, no need to copy it twice.
    no regression.
    Note that t/stress/gc.t started failing earlier, even in 7.0.0.
  25. @rurban

    [pmc] change StringBuilder overallocation

    rurban committed
    by 1.5, not 2. And for >8192 onto the next block size.
    The optimal overallocation is the golden ration 1.618, but we prefer int ops.
    A difference is not measurable.
  26. @rurban

    [test] fix t/steps/init/optimize-01.t for -O3

    rurban committed
    the optimization level needs to be the perl5 derived one
    or now also -O3
  27. @rurban

    [config] change perl5 derived -O2 to -O3 for gcc/clang

    rurban committed
    Faster --optimize. 5% faster in parrot-bench. Closes GH #1185
  28. @rurban

    travis: use the faster docker images

    rurban committed
    no sudo, use apt packages instead.
    also use perl 5.20 (5.22 would be best)
Commits on Sep 16, 2015
  1. @Util
  2. @Util

    Prepare for the 7.8.0 release.

    Util committed
Something went wrong with that request. Please try again.