Make a spec-looking reference implementation of outer grammar.

[phadej]

• The grammar is written in

  cabal/Cabal/Distribution/Parsec/Parser.hs

  Lines 159 to 206 in 76183b4

  	-- $grammar
  	--
  	-- @
  	-- CabalStyleFile ::= SecElems
  	--
  	-- SecElems ::= SecElem* '\n'?
  	-- SecElem ::= '\n' SecElemLayout | SecElemBraces
  	-- SecElemLayout ::= FieldLayout | FieldBraces | SectionLayout | SectionBraces
  	-- SecElemBraces ::= FieldInline | FieldBraces | SectionBraces
  	-- FieldLayout ::= name ':' line? ('\n' line)*
  	-- FieldBraces ::= name ':' '\n'? '{' content '}'
  	-- FieldInline ::= name ':' content
  	-- SectionLayout ::= name arg* SecElems
  	-- SectionBraces ::= name arg* '\n'? '{' SecElems '}'
  	-- @
  	--
  	-- and the same thing but left factored...
  	--
  	-- @
  	-- SecElems ::= SecElem*
  	-- SecElem ::= '\n' name SecElemLayout
  	-- | name SecElemBraces
  	-- SecElemLayout ::= ':' FieldLayoutOrBraces
  	-- | arg* SectionLayoutOrBraces
  	-- FieldLayoutOrBraces ::= '\n'? '{' content '}'
  	-- | line? ('\n' line)*
  	-- SectionLayoutOrBraces ::= '\n'? '{' SecElems '\n'? '}'
  	-- | SecElems
  	-- SecElemBraces ::= ':' FieldInlineOrBraces
  	-- | arg* '\n'? '{' SecElems '\n'? '}'
  	-- FieldInlineOrBraces ::= '\n'? '{' content '}'
  	-- | content
  	-- @
  	--
  	-- Note how we have several productions with the sequence:
  	--
  	-- > '\n'? '{'
  	--
  	-- That is, an optional newline (and indent) followed by a @{@ token.
  	-- In the @SectionLayoutOrBraces@ case you can see that this makes it
  	-- not fully left factored (because @SecElems@ can start with a @\n@).
  	-- Fully left factoring here would be ugly, and though we could use a
  	-- lookahead of two tokens to resolve the alternatives, we can't
  	-- conveniently use Parsec's 'try' here to get a lookahead of only two.
  	-- So instead we deal with this case in the lexer by making a line
  	-- where the first non-space is @{@ lex as just the @{@ token, without
  	-- the usual indent token. Then in the parser we can resolve everything
  	-- with just one token of lookahead and so without using 'try'.

  • it uses ad-hoc formalism
  • something like Indentation-Sensitive PEG from Indendation Sensitive Parsing for Parsec* https://dl.acm.org/citation.cfm?doid=2633357.2633369 should work better
  • https://github.com/phadej/cabal-refact/blob/master/src/Distribution/Refact/Parser.hs is one my experiment, it doesn't support { } (cabal files could have braces) IIRC.
  • https://github.com/haskell/cabal/blob/master/boot/Lexer.x is useful too
  • I think both trifecta-indentation and megaparsec should work.

• This is better done as a separate package in this repo cabal-trifecta or cabal-megaparsec, as there are technical limitations of having this as test-suite in Cabal package.

cc @sboosali

[hvr]

Could we use that as basis for a exactprint-style parser?

[phadej]

@hvr, Field omits comments. I'd rather not to deal with them in this. exactprint-style parser can of course be a fork, if it's suitable, but it's not a goal.

[sboosali]

since it's a separate package, can there be a larger number of dependencies? (like `megaparsec`, `frisby` or `indentation`).

…

On Dec 20, 2017 4:39 AM, "Oleg Grenrus" ***@***.***> wrote: @hvr <https://github.com/hvr>, Field omits comments. I'd rather not to deal with them in this. exactprint-style parser can of course be a fork, if it's suitable, but it's not a goal. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#4965 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACNoMUNeXl5r0YqSIeOBIqqjaw7rM95Kks5tCQARgaJpZM4RH2bw> .