Commits on Sep 6, 2012
  1. @tsee
  2. @tsee

    Test more expected failures

    tsee committed Sep 6, 2012
  3. @tsee

    Add author-only random-data decoder torture test

    This is for making sure that the decoder won't segfault on invalid
    packets. Trading CPU time for brain time.
    tsee committed Sep 6, 2012
  4. @tsee

    Test encoding of more integers around type boundaries

    Inspired by Storable tests.
    tsee committed Sep 6, 2012
  5. @tsee

    Skip length tests that depend on hash ordering

    This could be considered a temporary measure since it is
    a THEORY (if a damn good one) that this is due to hash key
    ordering and compression. Once we have an option to sort
    hash keys in the encoder, we can re-enable these tests.
    tsee committed Sep 6, 2012
  6. @demerphq

    add a space

    demerphq committed Sep 6, 2012
  7. @demerphq

    "fix" off by one error in how we use snappy.

    Note that tests still fail, however these tests are dependent on hash
    ordering, which in turn affects how well snappy compressed the data.
    
    IOW, if the "complex hash" tests fail on length checks under snappy but
    nothing else fails then the test is probably broken.
    demerphq committed Sep 6, 2012
  8. @tsee

    Snappy option in round-trip tests

    tsee committed Sep 6, 2012
  9. @tsee

    Better snappy errors

    tsee committed Sep 6, 2012
  10. @tsee

    Fix various snappy-compression bugs in the encoder

    Not all is well. /me pulls hair.
    tsee committed Sep 6, 2012
  11. @tsee

    snappy-benchmarking support

    tsee committed Sep 6, 2012
  12. @tsee

    Add snappy support to hobodecoder

    tsee committed Sep 6, 2012
  13. @tsee

    Fix snappy docs

    tsee committed Sep 6, 2012
  14. @tsee

    Better debugging

    tsee committed Sep 6, 2012
  15. @tsee

    Run roundtrip tests with Snappy

    Hint: Not all is well!
    tsee committed Sep 6, 2012
  16. @tsee

    Fixup shady exception test

    tsee committed Sep 6, 2012
  17. @tsee

    OBIWAN strikes again

    I wish I got a penny for every time I've made a stupid off-by-one error.
    tsee committed Sep 6, 2012
Commits on Sep 5, 2012
  1. @demerphq

    allocate the remaining reserved tags generate new reference tables

    no more reserved tags, just unimplemented ones!
    demerphq committed Sep 5, 2012
  2. @demerphq

    improve autogenerated stuff

    demerphq committed Sep 5, 2012
  3. @tsee

    Snappy decompression WIP

    It's still a total fail, but I must be close. Just need a fresh look in
    the morning. (Or an Yves to point out my obvious mistake.)
    tsee committed Sep 5, 2012
  4. @tsee
  5. @tsee

    Back out the uncompressed varint from header

    This is because Snappy already stores that information for us. Other
    compression algorithms will break compatibility anyway, so we can add
    this feature then, if we must.
    tsee committed Sep 5, 2012
  6. @tsee

    Enable Snappy option in decoder

    Not well debugged/tested.
    
    It appears that Snappy output will quite commonly be somewhat bigger
    than uncompressed output. In principle, we could then just revert to
    storing the uncompressed output at the expense of some extra memcpying,
    but that's not implemented yet.
    tsee committed Sep 5, 2012
  7. @tsee

    Implement Snappy-compression option

    Including the respective flags on encoder and the Sereal header.
    tsee committed Sep 5, 2012
  8. @tsee

    Import csnappy library code

    tsee committed Sep 5, 2012
  9. @demerphq
  10. @demerphq

    Improve header, reduce number of "tags" allocated to ASCII strings.

    This defines a new header and changes some of our tag allocations.
    
    As of this release the magic string for serial is four bytes consisting
    of "=srl". The '=' byte is reserved as a tag so that it will never be
    used as a tag inside fo the dump.
    
    Additionally we now only use 32 tags for short strings, so we can
    reallocate the other 32 to hash and array references.
    demerphq committed Sep 5, 2012
  11. @demerphq
  12. @demerphq

    Zefram on p5p said we should guard against refcount issues

    So here is the patch. Maybe this could be improved.
    demerphq committed Sep 5, 2012
  13. @tsee
  14. @tsee

    Bulk testing also in the decoder

    tsee committed Sep 5, 2012
  15. @tsee

    Move bulk tests to shared module

    tsee committed Sep 5, 2012
  16. @tsee

    Share corpus

    tsee committed Sep 5, 2012
Commits on Sep 4, 2012
  1. @demerphq

    More tests, integrate hobo to test fails

    Note we skip the string test on the "complex" tests as they
    .
    demerphq committed Sep 5, 2012
  2. @demerphq