I have a 4000 line text file which is parsing slowly, taking perhaps 3 minutes. I am running the Intellij Antlr plugin. When I look at the profiler, I see this:
The time being consumed is the largest of all rules, by a factor of 15 or so. That's ok, the file is full of things I actually don't care about (hence 'trash'). However, the profiler says words_and_trash is ambiguous but I don't know why. Here are the productions in question. (There are many others of course...):
I have no idea why this is ambiguous. The parser isn't complaining about so_much_trash and I don't think word, trash, and OPEN_PAREN overlap.
What's my strategy for solving this ambiguity?
It's ambiguous because, given your two alternatives for words_and_trash, anything that matches the first alternative, could also match the second alternative (that's the definition ambiguity in this context).
It appears you might be using a technique common in other grammar tools to handle repetition. ANTLR can do this like so:
words_and_trash: so_much_trash+;
so_much_trash: word
| trash
| OPEN_PAREN words_and_trash CLOSE_PAREN
;
You might also find the following video, useful: ANTLR4 Intellij Plugin -- Parser Preview, Parse Tree, and Profiling. It's by the author of ANTLR, and covers ambiguities.
Related
I need to match the values of key = value pairs in BibTeX files, which can contain arbitrarily nested braces, delimited by braces. I've got as far as matching at most two deep nested curly braces, like {some {stuff} like {this}} with the kludgey:
token brace-value {
'{' <-[{}]>* ['{' <-[}]>* '}' <-[{}]>* ]* '}'
}
I shudder at the idea of going one level further down... but proper parsing of my BibTeX stuff needs at least three levels deep.
Yes, I know there are BibTeX parsers around, but I need to grab the complete entry for further processing, and peek at a few keys meanwhile. My *.bib files are rather tame (and I wouldn't mind to handle a few stray entries by hand), the problem is that I have a lot of them, with much overlap. But some of the "same" entries have different keys, or extra data. I want to consolidate them into a few master files (the whole idea behind BibTeX, right?). Not fun by hand if bibtool gives a file with no duplicates (ha!) of some 20 thousand lines...
After perusing Lenz' "Parsing with Perl 6 Regexes and Grammars" (Apress, 2017), I realized the "regex" machinery (based on backtracking) might actually be a lot more capable than officially admitted, as a regex can call another, and nowhere do I see a prohibition on recursive calls.
Before digging in, a bit of context free grammars: A way to describing nested braces (and nothing else) is with the grammar:
S -> { S } S | <nothing>
I.e., nested braces are either an opening brace, nested braces, a closing brace, more nested braces; or nothing whatsoever. This translates more or less directly to Raku (there is no empty regex, fake it by making the construction optional):
my regex nb {
[ '{' <nb> '}' <nb> ]?
}
Lo and behold, this works. Need to fix up to avoid captures, kill backtracking (if it doesn't match on the first try, it won't ever match), and decorate with "anything else" fillers.
my regex nested-braces {
:ratchet
<-[{}]>*
[ '{' <.nested-braces> '}' <.nested-braces> ]?
<-[{}]>*
};
This checks out with my test cases.
For not-so-adventurous souls, there is the Text::Balanced module for Perl (formerly Perl 5, callable from Raku using Inline::Perl5). Not directly useful to me inside a grammar, unfortunately.
Solution
A way to describe nested braces (and nothing else)
Presuming a rule named &R, I'd likely write the following pattern if I was writing a quick small one-off script:
\{ <&R>* \}
If I was writing a larger program that should be maintainable I'd likely be writing a grammar and, using a rule named R the pattern would be:
'{' ~ '}' <R>*
This latter avoids leaning toothpick syndrome and uses the regex ~ operator.
These will both parse arbitrarily deeply nested paired braces, eg:
say '{{{{}}}}' ~~ token { \{ <&?ROUTINE>* \} } # ｢{{{{}}}}｣
(&?ROUTINE refers to the routine in which it appears. A regex is a routine. (Though you can't use <&?ROUTINE> in a regex declared with / ... / syntax.)
regex vs token
kill backtracking
my regex nested-braces {
:ratchet
The only difference between patterns declared with regex and token is that the former turns ratcheting off. So using it and then immediately turning ratcheting on is notably unidiomatic. Instead:
my token nested-braces {
Backtracking
the "regex" machinery (based on backtracking)
The grammar/regex engine does include backtracking as an optional feature because that's occasionally exactly what one wants.
But the engine is not "based on backtracking", and many grammars/parsers make little or no use of backtracking.
Recursion
a regex can call another, and nowhere do I see a prohibition on recursive calls.
This alone is nothing special for contemporary regex engines.
PCRE has supported recursion since 2000, and named regexes since 2003. Perl's default regex engine has supported both since 2007.
Their support for deeper levels of recursion and more named regexes being stored at once has been increasing over time.
Damian Conway's PPR uses these features of regexes to build non-trivial (but still small) parse trees.
Capabilities
a lot more capable
Raku "regexes" can be viewed as a cleaned up take on the unfolding regex evolution. To the degree this helps someone understand them, great.
But really, it's a whole new deal. For example, they're turing complete, in a sensible way, and thus able to parse anything.
than officially admitted
Well that's an odd thing to say! Raku's Grammars are frequently touted as one of Raku's most innovative features.
There are three major caveats:
Performance The primary current caveat is that a well written C parser will blow the socks off a well written Raku Grammar based parser.
Pay off It's often not worth the effort it takes to write a fully correct parser for a non-trivial format if there's an existing parser.
Left recursion Raku does not automatically rewrite left recursion (infinite loops).
Using existing parsers
I know there are BibTeX parsers around, but I need to grab the complete entry for further processing, and peek at a few keys meanwhile.
Using a foreign module in Raku can be a bit of a revelation. It is not necessarily like anything you'll have experienced before. Raku's foreign language adaptors can do smart marshaling for you so it can be like you're using native Raku features.
Two of the available foreign language adaptors are already sufficiently polished to be amazing -- the ones for Perl and for C.
I'm pretty sure there's a BibTeX package for Perl that wraps a C BibTeX parser. If you used that you'd hopefully get parsing results all nicely wrapped up into Raku objects as if it was all Raku in the first place, but retaining much of the high performance of the C code.
A Raku BibTeX Grammar?
Perhaps your needs do call for creating and using a small Raku Grammar.
(Maybe you're doing this partly as an exercise to familiarize yourself with Raku, or the regex/grammar aspect of Raku. For that it sounds pretty ideal.)
As soon as you begin to use multiple regexes together -- even just two -- you are closing in on grammar territory. After all, they're just an easy-to-use construct for using multiple regexes together.
So if you decide you want to stick with writing parsing code in Raku, expect to write it something like this:
grammar BiBTeX {
token TOP { ... }
token ...
token ...
}
BiBTeX.parse: my-bib-file
For more details, see the official doc's Grammar tutorial or read Moritz's book.
OK, just (re) checked. The documentation of '{' ~ '}' leaves a whole lot to desire, it is not at all clear it is meant to handle balanced, correctly nested delimiters.
So my final solution is really just along the lines:
my regex nested-braces {
:ratchet
'{' ~ '}' .*
}
Thanks everyone! Lerned quite a bit today.
I'm working on a new antlr grammar which is similar to nattys and should recognize date expressions, but I have problem with skip rules. In more detail I want to ignore useless "and"s in expressions for example:
Call Sam, John and Adam and fix a meeting with Sarah about the finance on Monday and Friday.
The first two "and"s are useless. I wrote the rule bellow to fix this problem but it didn't work, why? what should I do?
NW : [~WeekDay];
UselessAnd : AND NW -> skip;
"Useless AND" is a semantic concept.
Grammars are about syntax, and handle semantic issues poorly. Don't couple these together.
Suggestion: when you write a grammar for a language, make your parser accept the language as it is, warts and all. In your case, I suggest you "collect" the useless ANDs. That way you can get the grammar "right" more easily, and more transparently to the next coder who has to maintain your grammar.
Once you have the AST, it is pretty easy to ignore (semantically) useless things; if nothing else, you can post-process the AST and remove the useless AND nodes.
If I have a grammar where a certain expression can match two productions, I will obviously have a reduce/reduce conflict with yacc. Specifically, say I have two productions (FirstProduction and SecondProduction) where both of them could be TOKEN END.
Then yacc will not be able to know what to reduce TOKEN END to (FirstProduction or SecondProduction). However, I want to make it so that yacc prioritises FirstProduction in this situation. How can I achieve that?
Note that both FirstProduction and SecondProduction could be a great deal of things and that Body is the only place in the grammar where these conflict.
Also, I do know that in these situations, yacc will choose the first production that was declared in the grammar. However, I want to avoid having any reduce/reduce warnings.
You can refactor the grammar to not allow the second list to start with something that could be part of the first list:
Body: FirstProductionList SecondProductionList
| FirstProductionList
;
FirstProductionList: FirstProductionList FirstProduction
| /* empty */
;
SecondProductionList: SecondProductionList SecondProduction
| NonFirstProduction
;
NonFirstProduction is any production that is unique to SecondProduction, and marks the transition from reducing FirstProdutions to SecondProductions
Bison has no way to explicitly mark one production as preferred over another; the only such mechanism is precedence relations, which resolve shift/reduce conflicts. As you say, the file order provides an implicit priority. You can suppress the warning with an %expect declaration; unfortunately, that only lets you tell bison how many conflicts to expect, and not which conflicts.
I'm currently attempting to write a UCUM parser using ANTLR4. My current approach has involved defining every valid unit and prefix as a token.
Here's a very small subset of the defined tokens. I could make a cut-down version of the grammar as an example, but it seems like it shouldn't be necessary to resolve this problem (or to point out that I'm going about this entirely the wrong way).
MILLI_OR_METRE: 'm' ;
OSMOLE: 'osm' ;
MONTH: 'mo' ;
SECOND: 's' ;
One of the standard testcases is mosm, from which the lexer should generate the token stream MILLI_OR_METRE OSMOLE. Unfortunately, because ANTLR preferentially matches longer tokens, it generates the token stream MONTH SECOND MILLI_OR_METRE, which then causes the parser to raise an error.
Is it possible to make an ANTLR4 lexer try to match using shorter tokens first? Adding lookahead-type rules to MONTH isn't a great solution, as there are all sorts of potential lexing conflicts that I'd need to take account of (for example mol being lexed as MONTH LITRE instead of MOLE and so on).
EDIT:
StefanA below is of course correct; this is a job for a parser capable of backtracking (eg. recursive descent, packrat, PEG and probably various others... Coco/R is one reasonable package to do this). In an attempt to avoid adding a dependency on another parser generator (or moving other bits of the project from ANTLR to this new generator) I've hacked my way around the problem like this:
MONTH: 'mo' { _input.La(1) != 's' && _input.La(1) != 'l' && _input.La(1) != '_' }? ;
// (note: this is a C# project; java would use _input.LA instead)
but this isn't really a very extensible or maintainable solution, and like as not will have introduced other subtle issues I've not come across yet.
Your problem does not require smaller tokens to be preferred (In this case MONTH would never be matched). You need a backtracking behaviour dependent on the text being matched or not. Right?
ANTLR separates tokenization and parsing strictly. Consequently every solution to your problem will seem like a hack.
However other parser generators are specialized on problems like yours. Packrat Parsers (PEG) are backtracking and allow tokenization on the fly. Try out parboiled for this purpose.
Appears that the question is not being framed correctly.
I'm currently attempting to write a UCUM parser using ANTLR4. My current approach has involved defining every valid unit and prefix as a token.
But, according to the UCUM:
The expression syntax of The Unified Code for Units of Measure generates an infinite number of codes with the consequence that it is impossible to compile a table of all valid units.
The most to expect from the lexer is an unambiguous identification of the measurement string without regard to its semantic value. Similarly, a parser alone will be unable to validly select between unit sequences like MONTH LITRE and MOLE - both could reasonably apply to a leak rate - unless the problem space is statically constrained in the parser definition.
A heuristic, structural (explicitly identifying the problem space) or contextual (considering the relative nature of other units in the problem space), is most likely required to select the correct unit interpretation.
The best tool to use is the one that puts you in the best position to implement the heuristics necessary to disambiguate the unit strings. Antlr could do it using parse-tree walkers. Whether that is the appropriate approach requires further analysis.
This has got to be one of those well-known examples that's somewhere on the internet, but I can't seem to find it.
I'm trying to learn XText and I figured a calculator expression parser would be a good start. But I'm getting syntax errors in my grammar:
Expression:
Term (('+'|'-') Term)*;
Term:
Factor (('*'|'/') Factor)*;
Factor:
number=Number | variable=ID | ('(' expression=Expression ')');
I get this error in the Expression and Term lines:
Multiple markers at this line
- Cannot change type twice within a rule
- An unassigned rule call is not allowed, when the 'current'
was already created.
What gives? How can I fix this? And when do I have instanceName=Rule vs. Rule entries in a grammar?
I downloaded xtext integrated with eclipse and it comes with a calculator example which does approximately what you wish called arithmetics. From what I can gather you will need to assign an associativity to your tokens. This grammar runs fine for me:
Expression:
Term (({Plus.left=current}'+'|{Minus.left=current}'-') right=Term)*;
Term:
Factor (({Multiply.left=current} '*'| {Division.left=current}'/') right=Factor)*;
Factor:
number=NUMBER | variable=ID | ('(' expression=Expression ')');
The example grammar they have for arithmetics can be viewed here. It includes a bit more than your, like function calls, but the basics are the same.