--HG--
rename : tests/tests/.ghci => tests/.ghci
rename : tests/tests/Makefile => tests/Makefile
rename : tests/tests/src/Data/Text/Tests.hs => tests/Tests.hs
rename : tests/tests/src/Data/Text/Tests/IO.hs => tests/Tests/IO.hs
rename : tests/tests/src/Data/Text/Tests/Properties.hs => tests/Tests/Properties.hs
rename : tests/tests/src/Data/Text/Tests/QuickCheckUtils.hs => tests/Tests/QuickCheckUtils.hs
rename : tests/tests/src/Data/Text/Tests/Regressions.hs => tests/Tests/Regressions.hs
rename : tests/tests/src/Data/Text/Tests/SlowFunctions.hs => tests/Tests/SlowFunctions.hs
rename : tests/tests/src/Data/Text/Tests/Utils.hs => tests/Tests/Utils.hs
rename : tests/tests/scripts/cover-stdio.sh => tests/scripts/cover-stdio.sh
rename : tests/tests/text-tests.cabal => tests/text-tests.cabal

--HG--
rename : tests/benchmarks/.gitignore => benchmarks/.gitignore
rename : tests/benchmarks/Setup.hs => benchmarks/Setup.hs
rename : tests/benchmarks/cbits/time_iconv.c => benchmarks/cbits/time_iconv.c
rename : tests/benchmarks/src/Data/Text/Benchmarks.hs => benchmarks/haskell/Benchmarks.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Builder.hs => benchmarks/haskell/Benchmarks/Builder.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/DecodeUtf8.hs => benchmarks/haskell/Benchmarks/DecodeUtf8.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/EncodeUtf8.hs => benchmarks/haskell/Benchmarks/EncodeUtf8.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Equality.hs => benchmarks/haskell/Benchmarks/Equality.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/FileRead.hs => benchmarks/haskell/Benchmarks/FileRead.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/FoldLines.hs => benchmarks/haskell/Benchmarks/FoldLines.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Programs/BigTable.hs => benchmarks/haskell/Benchmarks/Programs/BigTable.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Programs/Cut.hs => benchmarks/haskell/Benchmarks/Programs/Cut.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Programs/Fold.hs => benchmarks/haskell/Benchmarks/Programs/Fold.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Programs/Sort.hs => benchmarks/haskell/Benchmarks/Programs/Sort.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Programs/StripTags.hs => benchmarks/haskell/Benchmarks/Programs/StripTags.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Programs/Throughput.hs => benchmarks/haskell/Benchmarks/Programs/Throughput.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Pure.hs => benchmarks/haskell/Benchmarks/Pure.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/ReadNumbers.hs => benchmarks/haskell/Benchmarks/ReadNumbers.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Replace.hs => benchmarks/haskell/Benchmarks/Replace.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Search.hs => benchmarks/haskell/Benchmarks/Search.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/Stream.hs => benchmarks/haskell/Benchmarks/Stream.hs
rename : tests/benchmarks/src/Data/Text/Benchmarks/WordFrequencies.hs => benchmarks/haskell/Benchmarks/WordFrequencies.hs
rename : tests/benchmarks/python/.gitignore => benchmarks/python/.gitignore
rename : tests/benchmarks/python/cut.py => benchmarks/python/cut.py
rename : tests/benchmarks/python/sort.py => benchmarks/python/sort.py
rename : tests/benchmarks/python/strip_tags.py => benchmarks/python/strip_tags.py
rename : tests/benchmarks/python/utils.py => benchmarks/python/utils.py
rename : tests/benchmarks/ruby/cut.rb => benchmarks/ruby/cut.rb
rename : tests/benchmarks/ruby/fold.rb => benchmarks/ruby/fold.rb
rename : tests/benchmarks/ruby/sort.rb => benchmarks/ruby/sort.rb
rename : tests/benchmarks/ruby/strip_tags.rb => benchmarks/ruby/strip_tags.rb
rename : tests/benchmarks/ruby/utils.rb => benchmarks/ruby/utils.rb
rename : tests/benchmarks/text-benchmarks.cabal => benchmarks/text-benchmarks.cabal

This improves performance by 10% on small strings, and reduces the
amount of memory allocated by 17%.

For a simple benchmark filtering many very small strings, this
allocates 7.8% less memory and 2.4% less time than unstream in
0.11.2.1.  The difference slowly disappears for larger strings.

The "use a specialised pair" approach can't match the reduction in
memory use, and for reasons unknown to me, seems to be slower than
the code in 0.11.2.1.

The specialised Run tuple avoids boxing and indirection for both
the MArray and Int parameters in the predecessor code.

I think I can do better, though.

Previously, every Text literal generated a big wad of inefficient
code:

    {-# LANGUAGE OverloadedStrings #-}
    foo :: Text
    foo = "foo"

This would first convert to a String, then to a Text.  To make
matters worse, the code for conversion from String to Text was
inlined at every site where a string literal occurred (expected,
but undesired, behaviour).

In this change, we introduce a direct conversion from Addr# to each
of the Text types, and we ensure that uses of these never result in
excessive code generation.

The effect of the rule was to make isPrefixOf infloop; see [GHC bug
5783](http://hackage.haskell.org/trac/ghc/ticket/5783) for details.

This fixes gh-16.

Simpler restreaming state

Add a span_ function, using unboxed tuples, to Text.Private.

Use span_ in a few places where it can help a little.

Relax the constraint on rational to Fractional.

Specialize over many more integral types.

My assertion that it was safe to skip the "do I have 1 byte available?" check
was incorrect.

This didn't actually work - it slowed down aeson encoding by almost 2x!

This drops a single allocation of a boxed integer.