Close # 27.
Backtracking user state can be achieved via combination of ‘StateT’
monad transformer and ‘ParsecT’:
StateT StateType (ParsecT s m a)
This user state can be more flexible. This fact renders current built-in
user state redundant.
To help work with this new approach (combining monad transformers more
freely) we introduce ‘MonadParsec’ MTL-style type class. All tools that
come with Megaparsec library were modified to work smoothly with any
instance of ‘MonadParsec’, not only ‘ParsecT’.
Now all the combinators in ‘Text.Megaparsec.Combinator’ are defined for
any instance of ‘Control.Alternative’ (sometimes ‘Control.Applicative’).
Some combinators are inlined.
Closes#29.
Now testing function can return ‘Either [Message] a’ so it can construct
full list of error messages. This may be useful in some cases when
tokens are more complex than simple characters.
Multi-character operators should use ‘try’ in order to be reported
correctly (as “operator”). I've mentioned it in doc-string of
‘makeExprParser’.
It's tempting to include ‘try’ directly in expression parsing code, but
following general spirit of Parsec toward ‘try’, I think current
solution is the best.
This code benchmarks the "string" primitive and various non-primitive combinators.
The code coverage of these benchmarks is not 100%: new benchmarks should be added
as relevant performance questions are discovered.
Various languages may vary in how hexadecimal and octal literals should
be prefixed. Following the spirit of the new lexer we leave this to
programmer to decide.
Eliminated ‘Text.Megaparsec.Language’ module because at this point it is
clear that already existing definitions are of little use in
Megaparsec. I started writing “default” language definition in
‘Text.Megaparsec.Lexer’.
At this point it should be possible to parse languages where indentation
matters, although we will need to provide more helpers to make it
easier.
The single test covers 100 % of the module's code. However it doesn't
check quality of error messages, so we still have room for improvement.
Manual tests show that error messages are good.
Obviously order does matter here, since ‘Monoid’ instance for ‘Hints’ is
derived from [], so (<>) is the same as (++) and we should be careful
to keep things in the right order.
These parsers are considered deprecated:
* ‘chainl’
* ‘chainl1’
* ‘chainr’
* ‘chainr1’
* ‘sepEndBy’
* ‘sepEndBy1’
Apart from this, the commit includes various cosmetic changes in
module ‘Text.Megaparsec.Combinator’.
The following functions and data types have been renamed:
* ‘permute’ → ‘makePermParser’
* ‘buildExpressionParser’ → ‘makeExprParser’
* ‘GenLanguageDef’ → ‘LanguageDef’
* ‘GenTokenParser’ → ‘Lexer’
* ‘makeTokenParser’ → ‘makeLexer’