Hi Nigel, Thanks for replying!
The real context of the problem is as following: In Fortran code, we write comment like: > ! This is a comment line. > use foo ! This is a comment following a use statement. then there are two scenarios: 1. A whole-line comment; 2. A inline comment. And I would like to reformat (or re-indent) the code into: > ! This is a comment line. > use foo ! This is a comment following a use statement. so I need to process the two kinds of comment differently, for the first one, it should follow the current indentation, for the second one, it should follow the previous statement with one space between them. Based on Kaspar's explanation, the lookahead can't span rules, so this can't be achieved by using lookahead to check what scenario is (I have a lot of rules). My potential solution would be that write two rules for the different comment scenario (currently they are parsed with the same rule). Cheers, Li 在 2013-6-29，下午1:24，Nigel Thorne <[email protected]> 写道： > I don't follow how you felt it would work that way... > > If you unroll your logic you get > > str('a').as(:a).maybe >> ( str('a').present? >> str('b').as(:b1) ) | > str('b').as(:b2) > > which says "consume an 'a' if there is one" then "if there is another 'a' > next, only match 'b'" or "read in a 'b'" > > str('a').present? >> str('b') will never match anything. > > so.. it sounds like you wanted to say.. > > Match "a" if there is one > If I matched "a" then matching a 'b' should mark it as 'b1' > otherwise matching a 'b' should mark it as 'b2' > > 'If I matched "a"' would be looking backwards not forwards so 'present?' > isn't going to work. > > Your goal however is the same as saying: > > "Either match "ab" (marking the b as 'b1') > Or match "b" (marking the b as "b2") > > Written like this... it becomes > > ( str('a') >> str('b').as(:b1) ) | ( str('b').as(:b2) ) > > The 'a' isn't optional. It's the existence of the 'a' that makes the parser > take that branch. > > Does this make clarify anything? > > > --- > "No man is an island... except Philip" > > > On Fri, Jun 28, 2013 at 5:17 PM, Kaspar Schiess <[email protected]> > wrote: > Hei Li, > > When parsing the string 'ab', the rule4 use of rule1 will consume 'a', > leaving 'b' in the input string. Then it will try to match rule1 (for > rule1.present?) again - but the input string at that point just contains > 'b', so :b2 matches and is consumed, yielding a successful match. > > > So 'present?' can not span rules? In my real application, the problem is > > more complicated, so I can not just merge 'rule3' and 'rule4'. > #present? protects subsequent matches in the same sequence from matching > unless the atom given to present is available at that point. So no, it > doesn't somehow span multiple rules, it has very local action. You will > need to formulate this differently. > > If you post a piece of your parser, maybe the list will help you > reformulate what you want so that present? works? > > cheers! > kaspar > > > >
