Bio.Iteratee.Builder

Buffer builder to assemble Bgzf blocks. (This will probably be renamed.) The plan is to serialize stuff (BAM and BCF) into a buffer, then Bgzf chunks from the buffer and reuse it. This should avoid redundant copying and relieve some pressure from the garbage collector. And I hope to plug a mysterious memory leak that doesn't show up in the profiler.

Exported functions with unsafe in the name resulting in a type of Push omit the bounds checking. To use them safely, an appropriate ensureBuffer has to precede them.

Documentation

data BB

The MutableByteArray is garbage collected, so we don't get leaks. Once it has grown to a practical size (and the initial 128k should be very practical), we don't get fragmentation either. We also avoid copies for the most part, since no intermediate ByteStrings, either lazy or strict have to be allocated.

Constructors

BB

Fields

buffer :: !(MutableByteArray RealWorld)

len :: !Int

mark :: !Int

mark2 :: !Int

newtype Push

Constructors

Push (BB -> IO BB)

Instances

Monoid Push

NullPoint Push

newBuffer :: IO BB

Creates a buffer with initial capacity of ~128k.

ensureBuffer :: Int -> Push

Ensures a given free space in the buffer by doubling its capacity if necessary.

expandBuffer :: BB -> IO BB

unsafePushByte :: Word8 -> Push

pushByte :: Word8 -> Push

unsafePushWord32 :: Word32 -> Push

unsafePushWord16 :: Word16 -> Push

pushWord32 :: Word32 -> Push

pushWord16 :: Word16 -> Push

unsafePushByteString :: ByteString -> Push

pushByteString :: ByteString -> Push

unsafePushFloat :: Float -> Push

pushFloat :: Float -> Push

pushBuilder :: Builder -> Push

unsafeSetMark :: Push

Sets a mark. This can later be filled in with a record length (used to create BAM records).

setMark :: Push

endRecord :: Push

Ends a record by filling the length into the field that was previously marked. Terrible things will happen if this wasn't preceded by a corresponding setMark.

endRecordPart1 :: Push

Ends the first part of a record. The length is filled in *before* the mark, which is specifically done to support the *two* length fields in BCF. It also remembers the current position. Horrible things happen if this isn't preceeded by *two* succesive invocations of setMark.

endRecordPart2 :: Push

Ends the second part of a record. The length is filled in at the mark, but computed from the sencond mark only. This is specifically done to support the *two* length fields in BCF. Horrible things happen if this isn't preceeded by *two* succesive invocations of setMark and one of endRecordPart1.

encodeBgzfWith :: MonadIO m => Int -> Enumeratee Push ByteString m b