Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
Commits on Jul 22, 2015
  1. @Util
Commits on Jun 15, 2015
  1. @rurban

    [core] optimize integer mod

    rurban authored
    for all positive int args use the HW mod insn,
    and do not call Parrot_util_intval_mod().
    This was profiled to be a bottleneck. 4% perf. win in parrot-bench.
Commits on May 25, 2015
  1. @Util

    Fix Coverity bug CID 743941 (NESTING_INDENT_MISMATCH).

    Util authored
    The statement is indented to column 9, as if it were nested
    within the preceding parent statement, but it is not.
  2. @Util

    Fix Coverity bug CID 61291: Insecure chroot (CHROOT).

    Util authored
    chdir must be called immediately after chroot.
Commits on May 20, 2015
  1. @Util
Commits on Apr 20, 2015
  1. @rurban

    OpenBSD has no RLIMIT_AS

    rurban authored
    as of 4.9
Commits on Mar 17, 2015
  1. @Util

    Ran `bootstrap-nci` to get changes introduced by 11611de ; only affec…

    Util authored
    …ted prelude comment.
Commits on Feb 8, 2015
  1. @rurban

    [gc] fix SIGBUS in GC trace_mem_block on alignment-strict CPUs (armv5)

    rurban authored
    We traced all mem and stack pointers +1 and not by +4 or +8. This is not
    only a huge performance loss, it is also illegal on certain RISC
    architectures.  ptrdiff_t is not a pointer, it is an integer.
    use size_t* instead.
    Note that this is not repro in qemu, even not with echo 5>/proc/cpu/alignment
    on armv5-linux.
    
    Should fix GH #1200.
Commits on Jan 31, 2015
  1. @rurban

    [test] Fix tests for --ccflags=-DSTRUCT_DEBUG. #1195

    rurban authored
    Skipped the failing tests due to stderr pollution,
    and added a # prefix to the debugging output.
    Closes GH #1195
Commits on Jan 28, 2015
  1. @rurban

    [native_pbc] finish cvt_num16_num8 fix #1193

    rurban authored
    move the old cvt_num16_num8 code for true __float128 to cvt_num16_num8_be,
    where it is really used in non-powerpc long doubles.
    
    Closes GH #1193.
    Also fixed a small codingstd problem.
    
    TODO: darwin ppc long-double, and many more conversions. See `native_pbc2` for that.
Commits on Jan 27, 2015
  1. @rurban

    [nativbe_pbc] fix win64 cvt_num16_num8

    rurban authored
    on intel assume we use intel 10-byte, not __float128. thus use cvt_num12_num8.
    Note that was already fixed in the branch native_pbc2
  2. @rurban

    win64: fix nci call regression, msvc only

    rurban authored
    On win64 with msvc the size of long is 4.
    Broken with 6.11.0
    Closes GH #1190
Commits on Jan 23, 2015
  1. @rurban

    Merge branch '7.0.1'

    rurban authored
    Fixed Conflict:
    	ChangeLog
  2. @rurban

    Release 7.0.1

    rurban authored
Commits on Jan 22, 2015
  1. @rurban

    [docs] alarm and threads

    rurban authored
    Update docs
  2. @rurban

    [threads] fix deadlock in gc_gms_mark_and_sweep

    rurban authored
    move the lock downwards to the guard. Fixes GH #1187
  3. @rurban

    [GC] optimize away a Parrot_pf_get_current_code_segment call

    rurban authored
    This was the call with the wrong CANNOT_RETURN_NULL attribute, causing GH #1186.
    We only need inter->code, and it can be const. The const_table also
  4. @rurban

    [GC] fix GC segfault with --optimize

    rurban authored
    A wrong PARROT_CANNOT_RETURN_NULL in Parrot_pf_get_current_code_segment
    (unchecked return of inter->code, which can be null) causes GC segfaults
    when inter->code == NULL.
    
    Fixes GH #1186.
    This wrong PARROT_CANNOT_RETURN_NULL was added with 3.6.0
    in 7dc0e22 at Wed Jun 29 20:52:24 2011
    but started being fatal after the new packfile refactor with 7.0,
    when the cc optimizer started optimizing away the bc != NULL check
    in mark_code_segment()
  5. @rurban

    [pmc] rpa: fix custom_mark_destroy confusion

    rurban authored
    custom_mark_destroy is only needed in the init method, which is fpa.set_integer_native.
    optimize PObj_custom_mark_destroy_SETALL to use only one setter call.
  6. @rurban

    [threads] fix deadlock in gc_gms_mark_and_sweep

    rurban authored
    move the lock downwards to the guard. Fixes GH #1187
  7. @rurban

    [GC] optimize away a Parrot_pf_get_current_code_segment call

    rurban authored
    This was the call with the wrong CANNOT_RETURN_NULL attribute, causing GH #1186.
    We only need inter->code, and it can be const. The const_table also
  8. @rurban

    [GC] fix GC segfault with --optimize

    rurban authored
    A wrong PARROT_CANNOT_RETURN_NULL in Parrot_pf_get_current_code_segment
    (unchecked return of inter->code, which can be null) causes GC segfaults
    when inter->code == NULL.
    
    Fixes GH #1186.
    This wrong PARROT_CANNOT_RETURN_NULL was added with 3.6.0
    in 7dc0e22 at Wed Jun 29 20:52:24 2011
    but started being fatal after the new packfile refactor with 7.0,
    when the cc optimizer started optimizing away the bc != NULL check
    in mark_code_segment()
  9. @rurban

    [pmc] rpa: fix custom_mark_destroy confusion

    rurban authored
    custom_mark_destroy is only needed in the init method, which is fpa.set_integer_native.
    optimize PObj_custom_mark_destroy_SETALL to use only one setter call.
  10. @rurban

    [atomic] cannot mix PARROT_EXPORT + PARROT_INLINE

    rurban authored
    remove the INLINE
Commits on Jan 21, 2015
  1. @rurban

    [config] Fix cpu config values for gcc_cmpxchg #1173

    rurban authored
    On amd64 atomic/gcc_x86.o is now included.
    The HAS_\U${platform}_$feature config hash entries are now harmonized.
  2. @rurban

    internal: remove interp arg from pf_register_standard_funcs()

    rurban authored
    This is not exported, so no deal. Closes GH #1183
Commits on Jan 20, 2015
  1. @moritz
Commits on Jan 19, 2015
  1. @rurban

    [nci] fix new libffi code for big-endian

    rurban authored
    wrong pointer for nci sigs 2, 3. See GH #601.
  2. @rurban

    [core] silence Unknown codeset `', defaulting to ASCII

    rurban authored
    On small or old libc's nl_langinfo will just return an empty string,
    where we used to default to ASCII, in Parrot_init_platform_encoding().
    
    This is a regression on ppc. Closes #1180.
Commits on Jan 18, 2015
  1. @rurban

    [nci] fix 234 call_ffi calculation for cornercases

    rurban authored
    failed on cygwin 32bit with INTVAL_SIZE=8, but probably also on
    all other special --intval and pointer sizes.
    Calculate the slack for INTVAL dynamically.
    Use it internally +1 for the switch fallthru. Improves GH #601
  2. @rurban

    [pmc] this should be the last splice fix GH #1176

    rurban authored
    Do not enlarge the array on overlarge offset. Rather fix the offset, as in perl5.
    Add a proper testcase also
  3. @rurban

    [pmc] fix more rpa splice cases #1174 + #1175 + #1176

    rurban authored
    - First, for debugging only:
      track if no TRACE_RPA was ever emitted and do it at the end to see a splice
      noop.
    - Second, on splice off > size, enlarge the array as in perl5, see GH #1176.
    - Third, disallow the splice shrink fast, just adjusting the offset if we have
      to keep one element at the left.
    - Forth, fix splice shrink followed by a grow or fill, by not adjusting the
      offset twice.
    - Fifth, on splice shrink fast, also adjust the new size, as the newsize branch
      below is not always taken (jump to splice_copy)
    
    Add an optional argument to t/stress/rpa-splice.t for the maxsize, default 7.
    We still have 'splice off > size' errors > 7.
  4. @rurban

    [test] enable the rpa-splice offset past end warning

    rurban authored
    See GH #1176. In parrot we have to manually enable the warning, in perl5 not. Whatever.
Commits on Jan 17, 2015
  1. @rurban

    [pmc] rpa: add splice: offset past end of array exception

    rurban authored
    conformant to perl5 semantics. Closes GH #1176
  2. @rurban

    [test] add more rpa-splice stress tests

    rurban authored
    this time with explicit result checks, with perl5 semantics.
    All tests pass, the new #1175 assertion in splice grow with (5,4,8) off=0 count=0
    not repro yet
Something went wrong with that request. Please try again.