I wish to understand how does a parser work. I learnt about the LL, LR(0), LR(1) parts, how to build, NFA, DFA, parse tables, etc.
Now the problem is, i know that a lexer should extract tokens only on the parser demand in some situation, when it's not possible to extract all the tokens in one separated pass. I don't exactly understand this kind of situation, so i'm open to any explanation about this.
The question now is, how should a lexer does its job ? should it base its recognition on the current "contexts", the current non-terminals supposed to be parsed ? is it something totally different ?
What about the GLR parsing : is it another case where a lexer could try different terminals, or is it only a syntactic business ?
I would also want to understand what it's related to, for example is it related to the kind of parsing technique (LL, LR, etc) or only the grammar ?
Thanks a lot
The simple answer is that lexeme extraction has to be done in context. What one might consider be lexemes in the language may vary considerably in different parts of the language. For example, in COBOL, the data declaration section has 'PIC' strings and location-sensitive level numbers 01-99 that do not appear in the procedure section.
The lexer thus to somehow know what part of the language is being processed, to know what lexemes to collect. This is often handled by having lexing states which each process some subset of the entire language set of lexemes (often with considerable overlap in the subset; e.g., identifiers tend to be pretty similar in my experience). These states form a high level finite state machine, with transitions between them when phase changing lexemes are encountered, e.g., the keywords that indicate entry into the data declaration or procedure section of the COBOL program. Modern languages like Java and C# minimize the need for this but most other languages I've encountered really need this kind of help in the lexer.
So-called "scannerless" parsers (you are thinking "GLR") work by getting rid of the lexer entirely; now there's no need for the lexer to produce lexemes, and no need to track lexical states :-} Such parsers work by simply writing the grammar down the level of individual characters; typically you find grammar rules that are the exact equivalent of what you'd write for a lexeme description. The question is then, why doesn't such a parser get confused as to which "lexeme" to produce? This is where the GLR part is useful. GLR parsers are happy to process many possible interpretations of the input ("locally ambiguous parses") as long as the choice gets eventually resolved. So what really happens in the case of "ambiguous tokens" is the the grammar rules for both "tokens" produce nonterminals for their respectives "lexemes", and the GLR parser continues to parse until one of the parsing paths dies out or the parser terminates with an ambiguous parse.
My company builds lots of parsers for languages. We use GLR parsers because they are very nice for handling complex languages; write the context-free grammar and you have a parser. We use lexical-state based lexeme extractors with the usual regular-expression specification of lexemes and lexical-state-transitions triggered by certain lexemes. We could arguably build scannerless GLR parsers (by making our lexers produce single characters as tokens :) but we find the efficiency of the state-based lexers to be worth the extra trouble.
As practical extensions, our lexers actually use push-down-stack automata for the high level state machine rather than mere finite state machines. This helps when one has high level FSA whose substates are identical, and where it is helpful for the lexer to manage nested structures (e.g, match parentheses) to manage a mode switch (e.g., when the parentheses all been matched).
A unique feature of our lexers: we also do a little tiny bit of what scannerless parsers do: sometimes when a keyword is recognized, our lexers will inject both a keyword and an identifier into the parser (simulates a scannerless parser with a grammar rule for each). The parser will of course only accept what it wants "in context" and simply throw away the wrong alternative. This gives us an easy to handle "keywords in context otherwise interpreted as identifiers", which occurs in many, many languages.
Ideally, the tokens themselves should be unambiguous; you should always be able to tokenise an input stream without the parser doing any additional work.
This isn't always so simple, so you have some tools to help you out:
Start conditions
A lexer action can change the scanner's start condition, meaning it can activate different sets of rules.
A typical example of this is string literal lexing; when you parse a string literal, the rules for tokenising usually become completely different to the language containing them. This is an example of an exclusive start condition.
You can separate ambiguous lexings if you can identify two separate start conditions for them and ensure the lexer enters them appropriately, given some preceding context.
Lexical tie-ins
This is a fancy name for carrying state in the lexer, and modifying it in the parser. If a certain action in your parser gets executed, it modifies some state in the lexer, which results in lexer actions returning different tokens. This should be avoided when necessary, because it makes your lexer and parser both more difficult to reason about, and makes some things (like GLR parsers) impossible.
The upside is that you can do things that would require significant grammar changes with relatively minor impact on the code; you can use information from the parse to influence the behaviour of the lexer, which in turn can come some way to solving your problem of what you see as an "ambiguous" grammar.
Logic, reasoning
It's probable that it is possible to lex it in one parse, and the above tools should come second to thinking about how you should be tokenising the input and trying to convert that into the language of lexical analysis. :)
The fact is, your input is comprised of tokens—whether you like it or not!—and all you need to do is find a way to make a program understand the rules you already know.
Related
Does there exist a formal definition of the purpose, or at a clear best practice of usage, of lexical analysis (lexer) during/before parsing?
I know that the purpose of a lexer is to transform a stream of characters to a stream of tokens, but can't it happen that in some (context-free) languages the intended notion of a "token" could nonetheless depend on the context and "tokens" could be hard to identify without complete parsing?
There seems to be nothing obviously wrong with having a lexer that transforms every input character into a token and lets the parser do the rest. But would it be acceptable to have a lexer that differentiates, for example, between a "unary minus" and a usual binary minus, instead of leaving this to the parser?
Are there any precise rules to follow when deciding what shall be done by the lexer and what shall be left to the parser?
Does there exist a formal definition of the purpose [of a lexical analyzer]?
No. Lexical analyzers are part of the world of practical programming, for which formal models are useful but not definitive. A program which purports to do something should do that thing, of course, but "lexically analyze my programming language" is not a sufficiently precise requirements statement.
… or a clear best practice of usage
As above, the lexical analyzer should do what it purports to do. It should also not attempt to do anything else. Code duplication should be avoided. Ideally, the code should be verifiable.
These best practices motivate the use of a mature and well-document scanner framework whose input language doubles as a description of the lexical grammar being analyzed. However, practical considerations based on the idiosyncracies of particular programming languages normally result in deviations from this ideal.
There seems to be nothing obviously wrong with having a lexer that transforms every input character into a token…
In that case, the lexical analyzer would be redundant; the parser could simply use the input stream as is. This is called "scannerless parsing", and it has its advocates. I'm not one of them, so I won't enter into a discussion of pros and cons. If you're interested, you could start with the Wikipedia article and follow its links. If this style fits your problem domain, go for it.
can't it happen that in some (context-free) languages the intended notion of a "token" could nonetheless depend on the context?
Sure. A classic example is found in EcmaScript regular expression "literals", which need to be lexically analyzed with a completely different scanner. EcmaScript 6 also defines string template literals, which require a separate scanning environment. This could motivate scannerless processing, but it can also be implemented with an LR(1) parser with lexical feedback, in which the reduce action of particular marker non-terminals causes a switch to a different scanner.
But would it be acceptable to have a lexer that differentiates, for example, between a "unary minus" and a usual binary minus, instead of leaving this to the parser?
Anything is acceptable if it works, but that particular example strikes me as not particular useful. LR (and even LL) expression parsers do not require any aid from the lexical scanner to show the context of a minus sign. (Naïve operator precedence grammars do require such assistance, but a more carefully thought out op-prec architecture wouldn't. However, the existence of LALR parser generators more or less obviates the need for op-prec parsers.)
Generally speaking, for the lexer to be able to identify syntactic context, it needs to duplicate the analysis being done by the parser, thus violating one of the basic best practices of code development ("don't duplicate functionality"). Nonetheless, it can occasionally be useful, so I wouldn't go so far as to advocate an absolute ban. For example, many parsers for yacc/bison-like production rules compensate for the fact that a naïve grammar is LALR(2) by specially marking ID tokens which are immediately followed by a colon.
Another example, again drawn from EcmaScript, is efficient handling of automatic semicolon insertion (ASI), which can be done using a lookup table whose keys are 2-tuples of consecutive tokens. Similarly, Python's whitespace-aware syntax is conveniently handled by assistance from the lexical scanner, which must be able to understand when indentation is relevant (not inside parentheses or braces, for example).
I am writing a parser in Bison for a language which has the following constructs, among others:
self-dispatch: [identifier arguments]
dispatch: [expression . identifier arguments]
string slicing: expression[expression,expression] - similar to Python.
arguments is a comma-separated list of expressions, which can be empty too. All of the above are expressions on their own, too.
My problem is that I am not sure how to parse both [method [other_method]] and [someString[idx1, idx2].toInt] or if it is possible to do this at all with an LALR(1) parser.
To be more precise, let's take the following example: [a[b]] (call method a with the result of method b). When it reaches the state [a . [b]] (the lookahead is the second [), it won't know whether to reduce a (which has already been reduced to identifier) to expression because something like a[b,c] might follow (which could itself be reduced to expression and continue with the second construct from above) or to keep it identifier (and shift it) because a list of arguments will follow (such as [b] in this case).
Is this shift/reduce conflict due to the way I expressed this grammar or is it not possible to parse all of these constructs with an LALR(1) parser?
And, a more general question, how can one prove that a language is/is not parsable by a particular type of parser?
Assuming your grammar is unambiguous (which the part you describe appears to be) then your best bet is to specify a %glr-parser. Since in most cases, the correct parse will be forced after only a few tokens, the overhead should not be noticeable, and the advantage is that you do not need to complicate either the grammar or the construction of the AST.
The one downside is that bison cannot verify that the grammar is unambiguous -- in general, this is not possible -- and it is not easy to prove. If it turns out that some input is ambiguous, the GLR parser will generate an error, so a good test suite is important.
Proving that the language is not LR(1) would be tricky, and I suspect that it would be impossible because the language probably is recognizable with an LALR(1) parser. (Impossible to tell without seeing the entire grammar, though.) But parsing (outside of CS theory) needs to create a correct parse tree in order to be useful, and the sort of modifications required to produce an LR grammar will also modify the AST, requiring a post-parse fixup. The difficultly in creating a correct AST spring from the difference in precedence between
a[b[c],d]
and
[a[b[c],d]]
In the first (subset) case, b binds to its argument list [c] and the comma has lower precedence; in the end, b[c] and d are sibling children of the slice. In the second case (method invocation), the comma is part of the argument list and binds more tightly than the method application; b, [c] and d are siblings in a method application. But you cannot decide the shape of the parse tree until an arbitrarily long input (since d could be any expression).
That's all a bit hand-wavey since "precedence" is not formally definable, and there are hacks which could make it possible to adjust the tree. Since the LR property is not really composable, it is really possible to provide a more rigorous analysis. But regardless, the GLR parser is likely to be the simplest and most robust solution.
One small point for future reference: CFGs are not just a programming tool; they also serve the purpose of clearly communicating the grammar in question. Nirmally, if you want to describe your language, you are better off using a clear CFG than trying to describe informally. Of course, meaningful non-terminal names will help, and a few examples never hurt, but the essence of the grammar is in the formal description and omitting that makes it harder for others to "be helpful".
If I understand correctly, parsing turns a sequence of symbols into a tree. My question is, is it possible to use some standard procedure (LR, LL, PEG, ..?) to parse the following two examples or is it necessary to write a specialized parser by hand?
Python source code, i.e. the whitespace-indented blocks
I think I read somewhere that the parser keeps track of the number of leading spaces, and pretends to replace them with curly brackets to delimitate the blocks. Is it fundamentally required because the standard parsing techniques are not powerful enough or is it for performance reasons?
PNG image format, where a block starts with a header and block size, after which there is the content of the block
The content could contain bytes which resemble some header so it is necessary to "know" that the next x bytes are not to be "parsed", i.e. they should be skipped. How to express this, say, with PEG? In other words, the "closing bracket" is represented by the length of the content.
Neither of the examples in the question are context-free, so strictly speaking they cannot be parsed with context-free grammars. But in practical terms, they are both pretty easy to parse.
The python algorithm is well-described in the Python reference manual (although you need to read that in context.) What's described there is a pre-processing step in which whitespace at the beginning of a line is systematically replaced with INDENT and DEDENT tokens.
To clarify: It's not really a preprocessing step, and it's important to observe that it happens after implicit and explicit line joining. (There are previous sections in the reference manual which describe these procedures.) In particular, lines are implicitly joined inside parentheses, braces and brackets, so the process is intertwined with parsing.
In practical terms, both the line-joining and indentation algorithms can be accomplished programmatically; typically, these would be done inside a custom scanner (tokenizer) which maintains both a stack of parentheses and indent levels. The token stream can then be parsed with normal context-free algorithms, but the tokenizer -- although it might use regular expressions -- needs context-sensitive logic (counting spaces, for example). [Note 1]
Similarly, formats which contain explicit sizes (such as most serialization formats, including PNG files, Google protobufs, and HTTP chunked encoding) are not context-free, but are obviously easy to tokenize since the tokenizer simply has to read the length and then read that many bytes.
There are a variety of context-sensitive formalisms, and these definitely have their uses, but in practical parsing the most common strategy is to use a Turing-equivalent formalism (such as any programming language, possibly augmented with a scanner-generator like flex) for the tokenizer and a context-free formalism for the parser. [Note 2]
Notes:
It may not be immediately obvious that Python indenting is not context-free, since context-free grammars can accept some categories of agreement. For example, {ωω-1 | ω∈Σ*} (the language of all even-length palindromes) is context-free, as is {anbn}.
However, these examples can't be extended, because the only count-agreement possible in a context-free language is bracketing. So while palindromes are context-free (you can implement the check with a single stack), the apparently very similar {ωω | ω∈Σ*} is not, and neither is {anbncn}
One such formalism is back-references in "regular" expressions, which might be available in some PEG implementation. Back-references allow the expression of a variety of context-sensitive languages, but do not allow the expression of all context-free languages. Unfortunately, regular expressions with back-references really suck in practice, because the problem of determining whether a string matches a regex with back-references is NP complete. You might find this question on a sister SE site interesting. (And you might want to reformulate your question in a way that could be asked on that site, http://cs.stackexchange.com.)
As a practical matter, almost all parser construction requires some clever hacks around the edges to overcome the limitations of the parsing machinery.
Pure context free parsers can't do Python; all the parser technologies you have listed are weaker than pure-context free, so they can't do it either. A hack in the lexer to keep track of indentation, and generate INDENT/DEDENT tokens, turns the indenting problem into explicit "parentheses", which are easily handled by context-free parsers.
Most binary files can't be processed either, as they usually contain, somewhere, a list of length N, where N is provided before the list body is encountered (this is kind of the example you gave). Again, you can get around this, with a more complicated hack; something must keep a stack of nested list lengths, and the parser has to signal when it moves from one list element to the next. The top-most length counter gets decremented, and the parser gets back a signal "reduce" or "shift". Other more complex linked structures are generally pretty hard to parse this way. Getting the parser to cooperate this way isn't always easy.
Is a theoretical LR parser with infinite lookahead capable of parsing (unambiguous) languages which can be desribed by a context-free grammar?
Normally LR(k) parsers are restricted to deterministic context-free languages. I think this means that there always has to be exactly one grammar rule that can be applied currently. Meaning within the current lookahead context not more than one possible way of parsing is allowed to occur. The book "Language Implementation Patterns" states that a "... parser is nondeterministic - it cannot determine which alternative to choose." if the lookahead sets overlap. In contrast a non-deterministic parser just chooses one way if there are multiple alternatives and then goes back to the decision point and chooses the next alternative if it is impossible at a certain point to continue with the decision previously made.
Wherever I read definitions of LR(k) parsers (like on Wikipedia or in the Dragon Book) I always read something like: "k is the number of lookahead tokens" or cases when "k > 1" but never if k can be infinite. Wouldn't an infinite lookahead be the same as trying all alternatives until one succeeds?
Could it be that k is assumed to be finite in order to (implicitly) distinguish LR(k) parsers from non-deterministic parsers?
You are raising several issues here that are difficult to answer in a short form. Nevertheless I will try.
First of all, what is "infinite lookahead"? There is no book that describes such parser. If you have clear idea of what is this, you need to describe it first. Only after that we can discuss this topic. For now parsing theory discusses only LR(k) grammars, where the k is finite.
Normally LR(k) parsers are restricted to deterministic context-free
languages. I think this means that there always has to be exactly one
grammar rule that can be applied currently.
This is wrong. LR(k) grammars may have "grammar conflicts". Dragon book briefly mentions them without going into any details. "Grammars may have conflicts" means that some grammars do not have conflicts, while all other grammars have them. When grammar does not have conflicts, then there is ALWAYS only one rule and the situation is relatively simple.
When grammar has conflicts, this means that in certain cases more than one rule is applicable. Classic parsers cannot help here. What makes matters worse is that some input statement may have a set of correct parsings, not just one. From the grammar theory stand point all these parsings have the same value and importance.
The book "Language Implementation Patterns" states that a "... parser
is nondeterministic - it cannot determine which alternative to
choose."
I have impression that there is no definitive agreement on what "nondeterministic parser" means. I would tend to say that nondeterministic parser just picks up one of the alternatives randomly and goes on.
Practically only 2 strategies of resolving conflicts are used. The first one is conflict resolution in the callback handler. Callback handler is a regular code. Programmer, who writes it, checks whatever he wants in any way he wants. This code only gives back the result - what action to take. For the parser on top this callback handler is a black box. There is no theory here.
Second approach is called "backtracking". The idea behind is very simple. We do not know where to go. Ok, let's try all possible alternatives. In this case all variants are tried. There is nothing non deterministic here. There are several different flavors of backtracking.
If this is not enough I can write a little bit more.
nondeterminism means that in order to produce the correct result(s!), a finite state machine reads a token and then has N>1 next states. You can recognize a nondeterministic FSM if a node has more than one outgoing edge with the same label. Note that not every branch has to be valid, but the FSM can't pick just one. In practice you could fork here, resulting in N state machines or you could try a branch completely and then come back and try the next one until every outgoing statetransfer was tested.
So I'm doing a Parser, where I favor flexibility over speed, and I want it to be easy to write grammars for, e.g. no tricky workaround rules (fake rules to solve conflicts etc, like you have to do in yacc/bison etc.)
There's a hand-coded Lexer with a fixed set of tokens (e.g. PLUS, DECIMAL, STRING_LIT, NAME, and so on) right now there are three types of rules:
TokenRule: matches a particular token
SequenceRule: matches an ordered list of rules
GroupRule: matches any rule from a list
For example, let's say we have the TokenRule 'varAccess', which matches token NAME (roughly /[A-Za-z][A-Za-z0-9_]*/), and the SequenceRule 'assignment', which matches [expression, TokenRule(PLUS), expression].
Expression is a GroupRule matching either 'assignment' or 'varAccess' (the actual ruleset I'm testing with is a bit more complete, but that'll do for the example)
But now let's say I want to parse
var1 = var2
And let's say the Parser begins with rule Expression (the order in which they are defined shouldn't matter - priorities will be solved later). And let's say the GroupRule expression will first try 'assignment'. Then since 'expression' is the first rule to be matched in 'assignment', it will try to parse an expression again, and so on until the stack is filled up and the computer - as expected - simply gives up in a sparkly segfault.
So what I did is - SequenceRules add themselves as 'leafs' to their first rule, and become non-roôt rules. Root rules are rules that the parser will first try. When one of those is applied and matches, it tries to subapply each of its leafs, one by one, until one matches. Then it tries the leafs of the matching leaf, and so on, until nothing matches anymore.
So that it can parse expressions like
var1 = var2 = var3 = var4
Just right =) Now the interesting stuff. This code:
var1 = (var2 + var3)
Won't parse. What happens is, var1 get parsed (varAccess), assign is sub-applied, it looks for an expression, tries 'parenthesis', begins, looks for an expression after the '(', finds var2, and then chokes on the '+' because it was expecting a ')'.
Why doesn't it match the 'var2 + var3' ? (and yes, there's an 'add' SequenceRule, before you ask). Because 'add' isn't a root rule (to avoid infinite recursion with the parse-expresssion-beginning-with-expression-etc.) and that leafs aren't tested in SequenceRules otherwise it would parse things like
reader readLine() println()
as
reader (readLine() println())
(e.g. '1 = 3' is the expression expected by add, the leaf of varAccess a)
whereas we'd like it to be left-associative, e.g. parsing as
(reader readLine()) println()
So anyway, now we've got this problem that we should be able to parse expression such as '1 + 2' within SequenceRules. What to do? Add a special case that when SequenceRules begin with a TokenRule, then the GroupRules it contains are tested for leafs? Would that even make sense outside that particular example? Or should one be able to specify in each element of a SequenceRule if it should be tested for leafs or not? Tell me what you think (other than throw away the whole system - that'll probably happen in a few months anyway)
P.S: Please, pretty please, don't answer something like "go read this 400pages book or you don't even deserve our time" If you feel the need to - just refrain yourself and go bash on reddit. Okay? Thanks in advance.
LL(k) parsers (top down recursive, whether automated or written by hand) require refactoring of your grammar to avoid left recursion, and often require special specifications of lookahead (e.g. ANTLR) to be able to handle k-token lookahead. Since grammars are complex, you get to discover k by experimenting, which is exactly the thing you wish to avoid.
YACC/LALR(1) grammars aviod the problem of left recursion, which is a big step forward. The bad news is that there are no real programming langauges (other than Wirth's original PASCAL) that are LALR(1). Therefore you get to hack your grammar to change it from LR(k) to LALR(1), again forcing you to suffer the experiments that expose the strange cases, and hacking the grammar reduction logic to try to handle K-lookaheads when the parser generators (YACC, BISON, ... you name it) produce 1-lookahead parsers.
GLR parsers (http://en.wikipedia.org/wiki/GLR_parser) allow you to avoid almost all of this nonsense. If you can write a context free parser, under most practical circumstances, a GLR parser will parse it without further effort. That's an enormous relief when you try to write arbitrary grammars. And a really good GLR parser will directly produce a tree.
BISON has been enhanced to do GLR parsing, sort of. You still have to write complicated logic to produce your desired AST, and you have to worry about how to handle failed parsers and cleaning up/deleting their corresponding (failed) trees. The DMS Software Reengineering Tookit provides standard GLR parsers for any context free grammar, and automatically builds ASTs without any additional effort on your part; ambiguous trees are automatically constructed and can be cleaned up by post-parsing semantic analyis. We've used this to do define 30+ language grammars including C, including C++ (which is widely thought to be hard to parse [and it is almost impossible to parse with YACC] but is straightforward with real GLR); see C+++ front end parser and AST builder based on DMS.
Bottom line: if you want to write grammar rules in a straightforward way, and get a parser to process them, use GLR parsing technology. Bison almost works. DMs really works.
My favourite parsing technique is to create recursive-descent (RD) parser from a PEG grammar specification. They are usually very fast, simple, and flexible. One nice advantage is you don't have to worry about separate tokenization passes, and worrying about squeezing the grammar into some LALR form is non-existent. Some PEG libraries are listed [here][1].
Sorry, I know this falls into throw away the system, but you are barely out of the gate with your problem and switching to a PEG RD parser, would just eliminate your headaches now.