arrow preprocessor

root

A prototype preprocessor for arrow notation.

Ross Paterson <ross@soi.city.ac.uk>

Note that recent versions of GHC support this notation directly, and
give better error messages to boot.

RUNNING THE ARROW PREPROCESSOR

The program supports the following options:

	--pretty=STYLE	pretty print in STYLE
			[offside|semicolon|inline|none] (default = offside)
	--help		display usage information

Typically you create a Haskell module using the extended syntax, and
call it Foo.as (for arrow script).  To convert this into a corresponding
Haskell module, you say

	arrowp Foo.as >Foo.hs
or
	arrowp Foo.as Foo.hs

and then use Foo.hs with your favourite Haskell implementation.
You can also run it with no arguments at all:

	arrowp <Foo.as >Foo.hs

but then it won't have a file name to put in the error messages.
Sometimes the input has already been preprocessed, e.g. with unlit
(from the GHC distribution).  For this situation, you can specify both
the original file name (for error messages) and the input file name, as
well as the output file name (which is obligatory in this situation), e.g.

	unlit Foo.las Foo.as
	arrowp Foo.las Foo.as Foo.hs

or equivalently as

	unlit Foo.las - | arrowp Foo.las - Foo.hs

(both unlit and arrowp allow filename arguments of "-", meaning standard
input or output as appropriate.)

WARNING: if you give the program 2 (or 3) arguments, the last one will
be overwritten.

DIAGNOSING ERRORS IN YOUR CODE

Syntax errors will be detected by the preprocessor, and the position in
the source where they were detected is reported.  (As usual, the actual
error may occur earlier.)  The preprocessor understands Haskell 98 plus
multiparameter type classes and the arrow extensions, but not some of
the other extensions supported by various Haskell implementations.
The preprocessor doesn't handle operator precedence, so it may copy
some erroneous expressions like `x == y == z'.  Otherwise, the output
of the preprocessor should not generate any syntax errors -- if it does,
please report it.

Semantic errors are more difficult for simple preprocessors like this one.
They will be detected by your Haskell implementation, whether they occur
in the Haskell part or the arrows part, and are thus rather hard to
trace back to the original source.  If you use --pretty=none and feed
the output to GHC, the error it reports will refer to the first line
of the declaration containing the error.  But if you need to read the
preprocessor output to find the problem, omit the --pretty option.

You can reduce the above problems by putting the parts of your program
that need arrow syntax (and thus the preprocessor) in separate modules.

BUGS

- there is no type checking until the output is compiled,
  resulting in incomprehensible error messages.
- failable patterns are not handled as promised yet.
  (And I'm starting to think they shouldn't be.)

SOURCE NOTES

This is based on hsparser, by Sven Panne, Simon Marlow and Noel Winstanley.
It is now the haskell-src package distributed with GHC.

My changes are

	Lexer -- a few lines adding extra symbols (all marked)
	Parser -- a few marked lines and several rules at the end,
		mostly modified versions of existing rules.
	ArrSyn, ArrCode, Utils -- new modules for arrow syntax.

The parser gives 2 shift/reduce conflicts, inherited from the original.

There was a reduce/reduce conflict:

	qual -> 'let' declist .
	qualA -> 'let' declist .

but I've kludged the grammar (see stmtA) to avoid it.