Permalink
Commits on Apr 7, 2014
Commits on Feb 24, 2014
  1. Merge pull request #911 from Daniel15/fix-catchall-routing

    Fix routing of CatchAll tokens (Bug #17960)
    marek-safar committed Feb 24, 2014
  2. Fix routing of CatchAll tokens

    Daniel15 committed Feb 24, 2014
Commits on Feb 19, 2014
  1. Fix another mips32 warning.

    alexrp committed Feb 19, 2014
  2. Fix some warnings on mips32.

    alexrp committed Feb 19, 2014
  3. Revert "[xbuild] Define VisualStudioVersion, since F# projects can de…

    …pend on it"
    
    This reverts commit 0a425af.
    
    Taking this out of 3.2.8 as per Don's comment:
    28163fd#commitcomment-5195420
    alexrp committed Feb 19, 2014
Commits on Feb 18, 2014
  1. Include xbuild_4_0_dirs in dist; fixes 425fdc0

    In commit 425fdc0:
    
      [xbuild] Build & install MSBuild 12.0 assemblies
    
    The net_3_5_only_dirs containing the Build classes were moved to
    various xbuild_X_dirs variables, but an undefined $(xbuild_dirs)
    variable was appended to DIST_ONLY_SUBDIRS.
    
    This patch adds $(xbuild_4_0_dirs) instead.
    awinters-fvs committed with alexrp Jan 23, 2014
  2. [metadata] Fix a potential crasher.

    kumpera committed with alexrp Jan 16, 2014
  3. [runtime] Fix the DISABLE_JIT build.

    vargaz committed with alexrp Jan 21, 2014
  4. [runtime] Handle native size of structs with explicit layout and min …

    …alignment smaller than default packing. Fixes #4510
    
    The calculating the native size of structs with explicit layout we need to take into consideration the minimum alignment
    needed and the packing asked.
    
    If no packing is supplied, we should respect the minimum alignment. If packing is supplied we should use the minimum required.
    kumpera committed with alexrp Jan 22, 2014
  5. [runtime] Fix the computation of AssemblyName.ProcessorArchitecture (…

    …) to be compatible with MS.NET. Fixes #17632.
    vargaz committed with alexrp Feb 6, 2014
  6. Revert "[jit] Align the stack in mono_arch_restore_context () on x86 …

    …since it might be unaligned if it was captured asynchronously."
    
    This reverts commit 7798b94.
    
    Revert this as it breaks debugging on android/x86 (#17410).
    vargaz committed with alexrp Feb 12, 2014
  7. [jit] Avoid running class cctors during AOT when using the Aggressive…

    …Inlining attribute. Fixes #17558.
    vargaz committed with alexrp Feb 4, 2014
  8. [corlib] Fix implementation of Tuple hash generator

    The Tuples.cs code generator dies with:
    
    Unhandled Exception:
    System.NotImplementedException: The requested feature is not implemented.
      at TupleGen.computeHash (Int32 destVar, Int32 start, Int32 count) [0x00000] in <filename unknown>:0
      at TupleGen.WriteHash (Int32 destVar, Int32 start, Int32 count) [0x00000] in <filename unknown>:0
      at TupleGen.Main () [0x00000] in <filename unknown>:0
    
    This is caused by the incorrect definition of WriteHash(), which
    instead of recursively calling itself is calling computeHash().
    
    Thanks to Marek Safar for reporting the issue in
    fsharp/fsharp#188 (comment)
    ranma42 committed with alexrp Feb 9, 2014
  9. [numerics] Fix type in Cosh

    marek-safar committed with alexrp Feb 7, 2014
  10. [runtime] Fix the linux build.

    kumpera committed with alexrp Feb 18, 2014
  11. [runtime] Add two variants of mono_gc_bzero and mono_gc_memmove. They…

    … better convey the behavior we need.
    
    After reviewing all usage of mono_gc_bzero and mono_gc_memmove, I found that there are two usage patterns
    we care for:
    
    1)Aligned pointer, any size, have managed pointers.
    This happens when directly dealing with managed objects like when cloning of boxing.
    On this case, all pointers MUST be aligned.
    
    2)Maybe aligned pointer, any size, might have managed pointers.
    This happens for functions dealing with structs, specially around boxing and unboxing them.
    Our data layout code ensures that structs with managed layout are aligned to at least a word.
    This means that if the pointer is unaligned, no managed pointer can exists in the memory range
    been processed.
    
    Given those two scenarios, lets introduce two variants that enforce those constrains.
    
    _aligned:
    Pointers must be aligned, ensures atomicity of word copies;
    
    _atomic:
    If pointers are aligned, works like _aligned, otherwise it inherits the behavior of the
    system memmove&memset.
    
    The above split ensures we won't have to deal with any RISC unaligned load nonsense.
    
    It also adds some sanity checks to code that must respect alignment.
    
    We still need fast code for mono_gc_memmove_aligned and mono_gc_bzero_aligned.
    
    We still need to verify if a version of memcpy version is needed. Maybe the extra
    branch is not such a perf penalty.
    kumpera committed with alexrp Feb 17, 2014
  12. ARM: Back up VFP scratch registers before using them on hard float.

    This is because d0-d7 are argument registers and d8-d15 are
    preserved registers - we can't use any of them as scratch
    registers without restoring their previous values.
    alexrp committed Feb 17, 2014
  13. [arm] Use d0/d1 as fp scratch regs on non-hard float platforms. Hopef…

    …ully fixes #17597.
    vargaz committed with alexrp Feb 10, 2014
  14. Bump mono to 3.2.8

    kumpera committed Feb 18, 2014
  15. [xbuild] Fix PCL 4.5 targets

    mhutch committed Feb 17, 2014
  16. Fix 'make dist' broken since 3528787

    In commit 3528787:
    
        [xbuild] Reorganize xbuild targets/frameworks/bin data
    
    Several files were shuffled around, but the EXTRA_DISTFILES was
    not fully updated with the new file paths.
    
    This patch is required to build a 3.2.7 tarball.
    awinters-fvs committed with mhutch Jan 23, 2014