package fmlib_parse

A parser is a consumer of tokens. At the end of consumption there is a result which is either a successfully parsed structure or a syntax or semantic error.

'a t Type of a parse combinator returning an 'a.

p >>= f

Parse first the input according to the combinator p. In case of success, feed the returned value of p into the function f to get the combinator to parse next.

let* x = p in f x is equivalent to p >>= f

The let* combinator let us express parsing sequences conveniently. Example:

let* x = p in       (* parse [p], result [x] in case of success. *)
let* y = q x in     (* parse [q x], result [y] ... *)
let* z = r x y in   (* ... *)
...
return f x y z ...

The wildcard let* _ = ... can be used to ignore results of intermediate parsing steps.

map f p

Try combinator p. In case of success, map the returned value x to f x. In case of failure, do nothing.

map f p is equivalent to let* x = p in return (f x).

map_and_update f p

Try combinator p. In case of success, map the returned state state and value a to f state a. In case of failure, do nothing.

succeed a

Succeed immediately without consuming token. Return object a as result.

return a is equivalent to succeed a.

unexpected expect triggers a syntax error signalling the expectation expect.

• deprecated

  Don't use this function. It will be removed in future versions.

clear_last_expectation p Clear last failed expectation.

This is useful e.g. after stripping whitespace. Since stripping whitespace means skip_one_or_more ws or skip_zero_or_more ws, after skipping whitespace the parser can still expect more whitespace. Therefore there is a failed expectation *whitespace* on the stack. However you rarely want this expectation to be reported.

• deprecated

  Use no_expectations

fail error triggers a semantic error.

p </> q

Try first combinator p. In case of success or failure with consumed token, p </> q is equivalent to p.

If p fails without consuming token, then p </> q is equivalent to q.

choices p [q r t ...] is equivalent to p </> q </> r </> t </> ....

p <?> expect

Try combinator p. In case of success or failure with consumed token, p <?> expect is equivalent to p.

If p fails without consuming token, then the failed expectations are replaced with the failed expectation expect.

Usually p is a combinator implementing a choice between various alternatives of a grammar construct. The <?> combinator allows to replace the set of failed grammar alternatives with a higher abstraction of the failed expectation. E.g. instead of getting the failed expectations identifier, '(', -, ... we can get the failed expectation expression.

no_expectations p

Parse the combinator p.

• p fails: no_expectations p fails with the same error.

• p succeeds without consuming tokens: no_expectations p succeeds without any added expectations.

• p succeeds and consumes some token: no_expectations p succeeds without any expectations.

Many combinators can succeed with expectations. E.g. the combinator optional p expects a p and succeeds if it does not encounter a construct described by p. All repetitive combinators like one_or_more try to consume as many items as possible. At the end they are still expecting an item.

This combinator allows to clear such unneeded expectations. It is particularly useful when removing whitespace. The expectation of whitespace is not a meaningful error message to the user.

Get the current user state.

Set the user state.

update f Update the user state using f.

get_and_update f Get the current user state and then update the user state. The returned value is the old state.

state_around before p after

If s0 is the initial state, then execute p with the start state before s0 and set the update the final state s1 by after s0 a s1 where a is the returned value in case of success and s1 is the final state after executing p.

optional p

Try combinator p.

• Success: Return Some a where a is the returned value.
• Failure without consuming token: Return None
• Failure with consuming token: Remain in the error state.

zero_or_more_fold_left start f p

Try the combinator p as often as possible. Accumulate the results to the start value start using the folding function f.

one_or_more_fold_left first f p

Try the combinator p at least once and then as often as possible. Put the first value returned by p into the function first returning a result and accumulate the subsequent values as often as possible and accumulate the results to the start value returned by first using the folding function f.

zero_or_more p Parse zero or more occurrences of p and return the collected result in a list.

zero_or_more p Parse one or more occurrences of p and return the collected results as a pair of the first value and a list of the remaining values.

skip_zero_or_more p Parse zero or more occurrences of p, ignore the result and return the number of occurrences.

skip_one_or_more p Parse one or more occurrences of p, ignore the result and return the number of occurrences.

one_or_more_separated first next p sep

Parse one or more occurrences of p separated by sep. Use first to convert the first occurrence of p into the result and use next to accumulate the results.

parenthesized make lpar p rpar

Parse an expression recognized by the combinator p enclosed within parentheses. lpar recognizes the left parenthesis and rpar recognizes the right parenthesis. The value returned by lpar is given to rpar. With that mechanism it is possible to recognize matching parentheses of different kinds.

After successful parsing the function make is called with the final value (and the parentheses).

The combinator p is entered as a thunk in order to be able to call it recursively. In the combinator parenthesized the combinator p is called only guardedly. Therefore the combinator p can contain nested parenthesized expressions.

Precondition: The combinator lpar has to consume at least one token in case of success.

operator_expression
    primary         (* Parse a primary expression *)
    unary_operator  (* Parse a unary operator *)
    binary_operator (* Parse a binary operator *)
    is_left         (* Is the left operator binding stronger? *)
    make_unary      (* Make a unary expression from the operator and
                       its operand *)
    make_binary     (* Make a binary expression from the operator
                       and its operands *)

Parse an operator expression by using the following combinators:

• is_left o1 o2 decides, if the operator o1 on the left has more binding power than the operator o2. I.e. if the unary operator u has more binding power than the binary operator o, then u a o b is parsed as (u a) o b. If the binary operator o1 has more binding power than the binary operator o2, then a o1 b o2 b is parsed as (a o1 b) o2 c.

• make_unary u a makes the unary expression (u a).

• make_binary a o b makes the binary expression (a o b).

• primary parses a primary expression.

• unary_operator parses a unary operator.

• binary_operator parses a binary operator.

Precondition: primary, unary_operator and binary_operator have to consume at least one token in case of success. Otherwise infinite recursion can happen.

backtrack p expect

Try the combinator p. In case of failure with consuming token, push the consumed token back to the lookahead and let it fail without consuming token. Use expect to record the failed expectation.

Backtracking reduces the performance, because the token pushed back to the lookahead have to be parsed again. Try to avoid backtracking whenever possible.

followed_by p expect

Parses p and backtracks (i.e. all tokens of p will be pushed back to the lookahead). In case p succeeds, the followed_by parser succeeds without consuming token. Otherwise it fails without consuming tokens.

not_followed_by p expect

Parses p and backtracks (i.e. all tokens of p will be pushed back to the lookahead). In case p succeeds, the not_followed_by parser fails without consuming token. Otherwise it succeeds without consuming tokens.

located p Parse p and return its result with its start and end position.

Note: If p removes whitespace at the end, the returned end position is at the end of the whitespace. This is not what you usually want. Therefore first parse the essential part located and then remove the whitespace.

The current position in the file.

indent i p Indent p by i columns relative to its parent.

Precondition: 0 <= i

The indentation of p is defined by the indentation of its first token. The first token has to be indented at least i columns relative to the parent of p. After the first token of p has been parsed successfully, all subsequent tokens must have at least the same indentation.

Note: Indentation of p relative to its parent only makes sense, if the first token of p is not the first token of its parent! I.e. the parent of p should have consumed at least one token before the parsing of p starts.

align p

Use the start position of the first token of p to align it with other constructs. If p does not consume any token, then align p has no effect.

Alignment makes sense if there are at least two combinators which are aligned and indented. E.g. suppose there are two combinators p and q. Then we can form

indent 1 (
        let* a = align p in
        let* b = align q in
        return (a,b)
)

This combinator parses p whose first token has to be indented at least one column relative to its parent. And then it parses q whose first token must be aligned with the first token of p.

The indentation decouples the alignment of p and q with other aligned siblings or parents. indent 0 ... can be used to make the indentation optional.

left_align p

Align a construct described by p at its leftmost possible column. If a whole block of constructs have to be vertically left aligned, then it is important that at least the first construct is left aligned. The subsequent constructs will be aligned exactly vertically. For the subsequent constructs left_align has the same effect as align.

detach p Parse p without any indentation and alignment restrictions.

Detachment is needed to parse whitespace. The whitespace at the beginning of a line never satisfies any nontrivial indentation or aligment requirements.

expect_end error a Expect the end of token stream.

In case of success return a.

In case of failure return the syntax error with the error string error.

CAUTION: There is usually no need to use this combinator! This combinator is needed only for partial parsers which are not based on a combinator constructed with lexer. A combinator constructed by lexer handles the end of input internally.

Never ever backtrack over this combinator.

charp p expect Parse a character which satisfies the predicate p.

In case of failure, report the failed expectation expect.

range c1 c2 Parses a character in the range between c1 and c2, i.e. a character c which satisfies c1 <= c && c <= c2.

char c Parse the character c.

one_of_chars str expect

Parse one of the characters in the string str. In case of failure, report the failed expectation expect.

string str Parse the string str.

Parse an uppercase letter.

Parse a lowercase letter.

Parse a letter.

Parse a digit 0..9 and return it as character.

Parse a digit and return it as number.

word first inner error

Parse a word which starts with a character satisfying the predicate first followed by zero or more characters satisfying the predicate inner. In case of failure add the expectation error.

Equivalent to range 'A' 'F' and then converted to the corresponding number between 10 and 15.

Equivalent to range 'a' 'f' and then converted to the corresponding number between 10 and 15.

Parse a hexadecimal digit and return the corresponding number between 0 and 15.

base64 start next Parse a base64 encoding into an object of type 'r.

A base64 encoding is a sequence of zero or more base64 characters (A-Za-z0-9+/) grouped into sequences of 4 characters and optionally padded with the character =. Each group of 2-4 base64 characters are decoded into a string of 1-3 bytes.

start gets the first 1-3 bytes and next gets all subsequent 1-3 bytes until the end of the encoding is reached.

Parse a base64 encoding and decode it into a string.

lexer whitespace end_token tok

A lexer combinator.

• The whitespace combinator recognizes a possibly empty sequence of whitespace (usually blanks, tabs, newlines, comments, ...).

• end_token is a token which the lexer returns when it has successfully consumed the end of input.

• tok is a combinator recognizing tokens (usually tok1 </> tok2 </> ... </> tokn).

The lexer combinator recognizes tokens in an input stream of the form

           WS Token WS Token .... WS EOF

Note: If a combinator fails to recognize a token and having consumed some input, then the subsequent combinators are not used anymore as alternatives. Therefore if there are tokens which can begin with the same prefix, then it is necessary to make the recognition of the common prefixes backtrackable in all but the last combinator recognizing a token with the same prefix. The same applies to whitespace if part of the whitespace can begin like a token.

Examples:

• comment: "// ...."
• division operator: "/"

In this case the recognition at least of the first slash of the comment has to be backtrackable.

make state c

Make a parser which starts in state state and parses a construct defined by the combinator c. The token stream must be ended by put_end, otherwise the parse won't succeed.

CAUTION: c must not be a combinator containing expect_end. Moreover it must not have been constructed by lexer.

make_partial state c

Make parser which analyzes a part of the input stream. The parser starts in state state and parses a construct defined by the combinator c. The parser can succeed even if no end token has been pushed into the parser.

restart_partial c p

Restart the partial parser p by using the combinator c to recognize the next part of the input stream. The restarted parser starts with the state and the file position of p.

Preconditions:

• has_succeeded p
• not (has_consumed_end p)