Leex regex problem

kokolegorille

Hello everyone,

I am having some trouble understanding leex behaviour. To give some context, I would like to create a parser for pgn files.

Here is some code to explain my problem.

I created the lexer file pgn_lexer.xrl with

% -- Definitions.

Definitions.

TAG        = (\[[^\]]*\])+
COMMENT    = ({[^}]*})+
WHITESPACE = [\s\t\n\r]

Rules.

{TAG}         : {token, {tag, TokenLine, TokenChars}}.
{COMMENT}     : {token, {comment, TokenLine, TokenChars}}.
{WHITESPACE}+ : skip_token.

Erlang code.

The important rule is for TAG

( \[ [^\]] * \] )+

I cannot understand why

iex> :pgn_lexer.string '[Tag1 "Value1"] [Tag2 "Value2"]{comment}'                                                        
{:ok,
 [
   {:tag, 1, '[Tag1 "Value1"]'},
   {:tag, 2, '[Tag2 "Value2"]'},
   {:comment, 2, '{comment}'}
 ], 2}
# This works has expected, it detects 2 tags

iex> :pgn_lexer.string '[Tag1 "Value1"][Tag2 "Value2"]{comment}'  
{:ok, [{:tag, 1, '[Tag1 "Value1"][Tag2 "Value2"]'}, {:comment, 1, '{comment}'}],
 1}
# This does not work, tag1 and tag2 are merged

Why do the tags need to be separated by a space?

Thanks for enlightments.

5 comments

#leex #regex

0 1146 5

2018-04-10 22:00:43 UTC

Showing Posts 5 to 1

Show Best Posts
Show All (oldest first)
Show All (newest first)

kip

ex_cldr Core Team

Arrrggghhhhh sorry for my misleading reply.

Post #5

kokolegorille

That did the trick, thank You very much.

Here is the modified version…

TAG        = (\[[^\]]*\])

iex> :pgn_lexer.string '[Tag1 "Value1"][Tag2 "Value2"]{comment}' 
{:ok,
 [
   {:tag, 1, '[Tag1 "Value1"]'},
   {:tag, 1, '[Tag2 "Value2"]'},
   {:comment, 1, '{comment}'}
 ], 1}

And as mentionned, the short version works as well

TAG        = \[[^\]]*\]

Post #4

NobbZ

Yes, they are greedy, but thats not the cause here.

TAG is defined as (\[[^\]]*\])+ and thus we demand at least one but allow many repitions of square-bracket-pairs-with-stuff-inbetween and consolidate them into one token.

If the second example is expected to spit out 2 tag tokens, then TAG should be just (\[[^\]]*\]) (without the +), now we should even be able to remove the grouping parens without changing anything.

Post #3

kokolegorille

Thank You for the answer, now I know why, and I know why I am spending hours too

Post #2

kip

ex_cldr Core Team

The reason is that regexes in Leex are greedy. Your regex in this case matches the longest string possible. Here’s a example using your regex and data in IEx:

iex> r = ~r/(\[[^\]]*\])+/
~r/(\[[^\]]*\])+/

iex> String.split "[Tag1 \"Value1\"] [Tag2 \"Value2\"]{comment}", r, include_captures: true
["", "[Tag1 \"Value1\"]", " ", "[Tag2 \"Value2\"]", "{comment}"]

iex> String.split "[Tag1 \"Value1\"][Tag2 \"Value2\"]{comment}", r, include_captures: true 
["", "[Tag1 \"Value1\"][Tag2 \"Value2\"]", "{comment}"]

In general I find these issues become prevalent when you’re using Leex to parse instead of just tokenize. (I know this because I’ve spent hours on similar cases myself). My learning is to use Leex to tokenise and Yecc to parse.

Or look at nimble_parsec or ex_spirit

Post #1

— All posts loaded —