…p_filtering' branch.

…r Hashbits.

Add KHMER_THREADED macro for conditional compilation with multi-threading support.
Wrap serial and parallel parsers in a namespace so that they can coexist (for easier porting).

…confused when linking it into a driver program.

…ones for debugging purposes. Fix bug with 'seq' variable not being declared thread-private.

…will build properly.

Allow 'load_graph' to test counting hashes or hashbits.
Add parallel reading support to 'consume_fasta' in 'hashtable.cc'.
Replace a constant literal with a named constant for parallel read chunk size.
Prevent build of incomplete driver program 'consume_prof'.

… modify.

Add TAU profiling support to Makefile.
Rename 'test.cc' to 'ktable_test.cc' so that the 'test' target doesn't confuse 'make'.

Add flag to compile code with extra sanity checking (with about a 15% to 20% performance loss).
Create high performance macro version of the 'toupper' function and use for normalization.
Use quicker code when extra sanity checking disabled: reduce number of calls to the 'toupper' function.

…c operations.

(EXPERIMENTAL) Allow slightly sloppy high counts in multi-threaded operation.
Reduce MAX_COUNT threshold for counting hash bins to accomodate some slop from multiple threads crossing the threshold.

…ogram.

Create Python wrapper for 'khmer:: Config'.
Create methods to get and set the number of threads in the configuration interface.
Create methods to get hash bin counting thresholds in the configuration interface.
Only allow number of threads to be set when codes are built with 'KHMER_THREADED' defined.
Note: C++ API allows multiple config objects (with one active at any given time), but Python wrapper effectively works with a singleton for now. The bookkeeping required to keep the wrapper objects synchronized with the underlying API does not seem worth the dubious benefit of multiple config objects at this time.

Tweak Python wrapper slightly for more cosmetic output in interpreter.
Ensure that driver programs are linked with khmer_config.o.

Refine interaction between lib/Makefile and the build of the bundled zlib.
Dynamically generate python/setup.py according to user-tweakable knobs which are set. (Automatically maintains consistency with what is built in lib.)

… version control.

Breakout the automatic generation of setup.py into a separate make rule.

Update Python wrapper and tests accordingly.

… force. (I.e., beware of consuming lowercase characters.)

…-level makefile.

…directory.

… has not been generated. (Handle multiple successive cleanings without breaking.)

…ned int so that attempts to set negative numbers of threads can be caught and handled properly.

(Doh! It helps to use the atomic operation you intended (compare-and-swap) rather than the one your fingers actually wrote (add).)
(Adding 0 to a variable having a value of 1 is not the same as setting that variable to 0. :-)
The entire automated test suite now passes with the threaded code.

…was found if not at end of file. (Technically, one could question rather this should be an assert or whether it should throw an exception.)

… logic.

…c operations.

Fuse testing for and setting kmer presence bits in Bloom filter hash tables into a single function for reasons of hopefully-improved efficiency and adding thread safety.
Cleanup and simplify declaration of OpenMP parallel region in 'consume_fasta_and_tag'.
Note: More performance testing and tuning must still be done. Some pieces of code related to this have been left in, wrapped in conditional compilation constructs.

Update tests of config interface accordingly.

… API or with Cython, but not everything tests out.

Merge branch 'merge_cython' into mt-reader

Conflicts:
	lib/parsers.cc
	lib/parsers.hh
	python/Makefile
	python/setup.py.in

NOTE: The 'get_config' method is not quite right, due to the way Cython handles C++ references.

…ch is disabled in Cython interface.

…le' in Cython merge.

…rude try-except hack during imports.

NOTE: The 'test_count_overlap_cpp' test may be broken. Rest of commit should be good though.

Conflicts:
	lib/hashbits.cc
	lib/hashbits.hh
	tests/test_scripts.py

Note: All tests now pass. However, one test requires 'matplotlib' (which requires 'NumPy'). We may wish to make the execution of this test depedent on the availability of 'matplotlib' on the user's system or else make 'matplotlib' a prerequistite for the 'khmer' package.

Conflicts:
	tests/test_scripts.py

… reads files and the stride length to be used by threaded reads parsers.

…ugh.)

Rework 'lib/Makefile' for a simpler zlib dependency graph.

…nes.

Make changes associated with the rename.
Improve the setup.py template to be more readable and hackable.

Resolve conflicts.

Conflicts:
	lib/Makefile
	lib/hashtable.cc
	lib/hashtable.hh

…s for new stream readers.

…ers.

Note: This commit includes a bug when appending a wrapped-around ssegment to a psegment.
Note: This implementation is not efficient across multiple NUMA nodes. Needs to be rewritten so that cache segments are allocated separately on demand rather than as part of one large pool which will be biased to the physical memory of the master thread.

…iendly and also does not have any bugs I am aware of from thinking about the code. Now on to finding and fixing those of which I am not aware....

… from the stream.

Instrument cache manager debug tracing and performance measurement logic.
Fix enough logic bugs for multi-threaded tests to successfully complete.

Fix a bug where a byte reader function could be using the wrong cache segment size.

…d probably store in a metrics object instead.)

… Still need to test.

…cache managers. (Still need to test.)

…n of cache manager segments. (Note: Still need to test.)

Add performance metrics and execution tracing to the parser interface.
(Still need to test.)

…uld not be regenerated on change of a header which the corresponding source used.

Fix bug where data was being lost if a cache segment fill occurred in the midst of copying memory to a parser buffer.

… being the Python built-in 'type'. Having a metaclass allows the object attributes to be listed and type to be determined. (Imagine that....)

Add parameter for tracking segment and setaside buffer boundaries to 'get_bytes' method in CacheManager.
Temporarily add deeper tracing to parser for debugging purposes.
Fix bug which caused overlapping reads and a segmentation fault from an invalid split position.

…er a fil l segment boundary had been crossed.

…s being flagged prematurely (before the parser buffer was fully consumed).

…the final parser bug.

Fix bug where first seqment fill would make a setaside buffer of size 0, causing fill segment 1 to erroneously believe that it did not have an extension. (I.e., fix an off-by-one error.)

…bytes' function was not checking if a cache segment was still valid and was recopying bytes from the start of the segment after the cursor was reset to there during a segment maintenance in which the segment was invalidated (due to end of stream being reached).

(No more known bugs in the cache manager and parser logic as of this commit.)

… 'TLVL_ALL'. Shuffle some includes from the read parsers implementation file to the read parsers header file.

Fix from QP to remove dependency upon the 'matplotlib' package.

Conflicts:
	tests/test_scripts.py

… fix a possible bug, but code executes without crashing.)

…fined.

…sting purposes. (The hashtables produced by the original parser and single-threaded operation of the new parser are currently indentical, btw.)

…igcounts and for sloppy counting.)