aboutsummaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
* Added note about mem leak with a bandaid fixSteve Slaven2009-11-031-10/+8
|
* Use a cStringBuffer for IO which appears to be faster than using join() forSteve Slaven2009-11-032-1/+32
| | | | very large lists
* Move chunkbuffer out a class level so we can try a couple different thingsSteve Slaven2009-11-032-34/+41
| | | | to see if we can optimize string() which is the slowest part now
* Fix wrong block count for cacheSteve Slaven2009-11-031-1/+1
|
* Keep track of length as long as possible so we don't have to loop overSteve Slaven2009-11-031-8/+5
| | | | chunk lengths or rejoin our string all the time
* Handle truncateSteve Slaven2009-11-021-0/+4
|
* Handle some of the string stuff more efficiently by not splittingSteve Slaven2009-11-021-5/+20
| | | | everything in to characters but trying to keep a string list around instead
* Use a stringbuffer list instead of appending strings, apparently pythonSteve Slaven2009-11-022-39/+88
| | | | string handling when appending/using large strings is very bad
* Back to shaSteve Slaven2009-08-071-2/+2
|
* Update chunk references on chunk flush too, so we don't lose partially usedSteve Slaven2009-08-071-37/+44
| | | | | chunks that never get locked or unlocked because we haven't sync'd the whole file yet
* A note about refcount testing as a TODOSteve Slaven2009-08-071-1/+2
|
* Turned off profilingSteve Slaven2009-08-071-1/+1
|
* Don't re-read the entire file in case we are extending it when we don'tSteve Slaven2009-08-071-2/+7
| | | | | have enough blocks, just start from the last block. This increases performace significantly for large files
* More notes about profilingSteve Slaven2009-08-071-0/+5
|
* Switched to cProfileSteve Slaven2009-08-071-5/+4
|
* Trying out md5 to see if it speeds up someSteve Slaven2009-08-071-2/+3
|
* Added some loggingSteve Slaven2009-08-071-1/+9
|
* Digest wrapperSteve Slaven2009-08-071-0/+7
|
* Add swapable digest for key generationSteve Slaven2009-08-072-4/+6
|
* Kill the logging.debug function if we're not debugging, otherwise it showsSteve Slaven2009-08-071-0/+4
| | | | | up in the top 3 functions in profiling, making it mostly nop speeds things up a bit
* Moved everything to main() so we can run the profiler againSteve Slaven2009-08-071-34/+38
|
* Added note about open/unlink for temp file probably not working (but notSteve Slaven2009-08-071-0/+2
| | | | tested though)
* Added a method to free blocks before unlink, just doing ftruncate thenSteve Slaven2009-08-072-2/+9
| | | | | unlink left the "empty" file and "empty" block with an extra ref for every unlink since we never freed those blocks
* Changed the way we look for the debug flag and actually turn the flag onSteve Slaven2009-08-071-2/+1
| | | | ONLY when we see debug enabled
* Turn on debugging with -d instead of having to edit the scriptSteve Slaven2009-08-071-9/+13
|
* Updated some things that are doneSteve Slaven2009-08-071-3/+3
|
* Don't gzip serialized data anymore since it's basically just a single keySteve Slaven2009-08-071-5/+5
| | | | | | | in the physical file, and the chunked data that has the actual file chunk contents is already compressed. Compressing the single key data in the tree actually makes it bigger (50+ bytes instead of 34) which can be seen by gzipping them manually to test
* Added note about wanting to delete empty directoriesSteve Slaven2009-08-071-1/+4
|
* Store a key in the phyiscal file that pulls the data from a chunk toSteve Slaven2009-08-072-4/+35
| | | | minimize 100% file duplicates to be nothing more than a key length
* Update usage, took out xmp.py sample copyrights since almost all of xmp.pySteve Slaven2009-08-071-4/+4
| | | | has been replaced
* Pre-init original_chunks because we might need to fflush early in theSteve Slaven2009-08-062-1/+14
| | | | | constructor, use ftruncate since truncate doesn't exist, and handle unlocking unused chunks due to ftruncate
* Truncate files on unlink so we free their blocksSteve Slaven2009-08-061-0/+4
|
* Update chunk ref counts when we update the on-disk chunk listSteve Slaven2009-08-061-23/+30
|
* Add reference counting so we can delete unused chunksSteve Slaven2009-08-065-24/+93
|
* Another thoughtstable2009080601Steve Slaven2009-08-061-0/+2
|
* Debugging utilitiesSteve Slaven2009-08-042-0/+32
|
* More thoughts/notesSteve Slaven2009-08-041-0/+7
|
* Made header length a variable so we can change it later if neededSteve Slaven2009-08-041-4/+6
|
* Added more logging, fixed error with storage size for block lengthSteve Slaven2009-08-041-3/+18
|
* Added note about fsckSteve Slaven2009-08-041-0/+3
|
* Don't use a stream class for nowSteve Slaven2009-08-041-3/+5
|
* Fix deserialize nameSteve Slaven2009-08-041-1/+1
|
* Added binary format chunk serializerSteve Slaven2009-08-042-1/+74
|
* Namespace/module fixesSteve Slaven2009-08-041-3/+3
|
* Namespace/module fixesSteve Slaven2009-08-042-2/+8
|
* Added storage switch back inSteve Slaven2009-08-041-3/+8
|
* Fix namespace problemSteve Slaven2009-08-043-5/+5
|
* Fix some missing symbolsSteve Slaven2009-08-041-2/+4
|
* Use the new chunk storageSteve Slaven2009-08-041-1/+4
|
* Chunk storage handlersSteve Slaven2009-08-043-0/+197
|