add Data.Align.Text

[ocramz]

• Add 'text' dependency
• Add 'CHANGELOG.md'
• Moved common types to Data.Align.Types
• Add type signatures in Data.Align.Demo
• Small cosmetic refactorings in Data.Align
• bump package version to 0.2

[robinp]

Thanks for the changes! Minor comments, looks good otherwise.

Optional for bonus points (in separate PR): think up & add some property tests with hedgehog.

[robinp]

Please move to be a local helper at callsite, as mentioned in #2 .

[ocramz]

Good point, but we're now using this in Data.Align.Text as well. Perhaps an INLINE pragma would work instead?

[robinp]

If you have even anecdotal knowledge, please add a few lines of comment about expectable gains vs using stringy version.

[ocramz]

I think better still would be to set up a benchmark suite.

[ocramz]

Now I'm not even too sure about duplicating align, since the only differences in the implementations are the implementations of index and length. Perhaps using a typeclass with instances for Vector, String and Text would be better?

[robinp]

Is the duplication driven by an actual performance need? If not, can as well keep the listy module only, and convert at callsite.

Otherwise, doing benchmarks on the real problem (against the simply-wrapping) with criterion indeed sounds good.

Not sure about the typeclass. A bit hackier, but an often employed trick (see containers IIRC) is to have a header file contain the code (with some macro placeholders), and include it with CPP from the various implementations.

Personally I never liked the trick, but the typeclass move is something to be made only if benchmark data verifies its positive effect (passing around dictionaries and whatnot).