ebnf.py

# ebnf.py - EBNF -> Python-Parser compilation for DHParser
#
# Copyright 2016  by Eckhart Arnold (arnold@badw.de)
#                 Bavarian Academy of Sciences an Humanities (badw.de)
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
# implied.  See the License for the specific language governing
# permissions and limitations under the License.


"""
Module ``ebnf`` provides an EBNF-parser-generator that compiles an
EBNF-Grammar into avPython-code that can be executed to parse source text
conforming to this grammar into concrete syntax trees.

Specifying Grammers with EBNF
-----------------------------

With DHParser, Grammars can be specified either directly in Python-code
(see :py:mod:`DHParser.parse`) or in one of several EBNF-dialects. (Yes,
DHParser supports several different variants of EBNF! This makes it easy
to crate a parser directly from Grammars found in external sources.)
"EBNF" stands for the "Extended-Backus-Naur-Form" which is a common
formalism for specifying Grammars for context-free-languages.
(see https://en.wikipedia.org/wiki/Extended_Backus%E2%80%93Naur_form)

The recommended way of compiling grammars with DHParser is to either
write the EBNF-specification for that Grammar into a text-file and then
compile EBNF-source to an executable as well as importable Python-module
with the help of the "dhparser"-skript. Or, for bigger projects, to
create a new domain-specific-language-project with the DHParser-skript
as described in the step-by-step-guide.

However, here we will show how to compile an EBNF-specified grammar
from within Python-code and how to execute the parser that was
generated by compiling the grammar.

As an example, we will realize a json-parser (https://www.json.org/).
Let's start with creating some test-data::

    >>> testobj = {'array': [1, 2.0, "a string"], 'number': -1.3e+25, 'bool': False}
    >>> import json
    >>> testdata = json.dumps(testobj)
    >>> testdata
    '{"array": [1, 2.0, "a string"], "number": -1.3e+25, "bool": false}'

We define the json-Grammar (see https://www.json.org/) in
top-down manner in EBNF. We'll use a regular-expression look-alike
syntax. EBNF, as you may recall, consists of a sequence of symbol
definitions. The definiens of those definitions either is a string
literal or regular expression or other symbols or a combination
of these with four different operators: 1. sequences
2. alternatives 3. options and 4. repetitions. Here is how these
elements are denoted in classical and regex-like EBNF-syntax:

========================  ==================  ================
element                   classical EBNF      regex-like
========================  ==================  ================
insignificant whitespace  ~                   ~
string literal            "..." or \\`...\\`    "..." or \\`...\\`
regular expr.             /.../               /.../
sequences                 A B C               A B C
alternatives              A | B | C           A | B | C
options                   [ ... ]             ...?
repetions                 { ... }             ...*
one or more                                   ...+
grouping                  (...)               (...)
========================  ==================  ================

"insignificant whitespace" is a speciality of DHParser. Denoting
insignificant whitespace with a particular sign `~` allows to eliminate
it already during the parsing process without burdening later
syntax-tree-processing stages with this common task. DHParser offers
several more facilities to restrain the verbosity of the concrete
syntax tree, so that the outcome of the parsing stage comes close (or
at least closer) to the intended abstract-syntax-tree, already.

JSON consists of two complex data types, 1) associative arrays,
called "object" and sequences of heterogeneous data, called array; and
of four simple data types, 1) string, 2) number, 3) bool and 4) null.
The structure of a JSON file can easily be described in EBNF::

    >>> grammar = '''
    ...     json     = ~ _element _EOF
    ...     _EOF     = /$/
    ...     _element = object | array | string | number | bool | null
    ...     object   = "{" ~ member ( "," ~ §member )* "}" ~
    ...     member   = string ":" ~ _element
    ...     array    = "[" ~ ( _element ( "," ~ _element )* )? "]" ~
    ...     string   = `"` _CHARS `"` ~
    ...     _CHARS   = /[^"\\\\\\]+/ | /\\\\\\[\\\\/bnrt\\\\\\]/
    ...     number   = _INT _FRAC? _EXP? ~
    ...     _INT     = `-`? ( /[1-9][0-9]+/ | /[0-9]/ )
    ...     _FRAC    = `.` /[0-9]+/
    ...     _EXP     = (`E`|`e`) [`+`|`-`] /[0-9]+/
    ...     bool     = "true" ~ | "false" ~
    ...     null     = "null" ~                                   '''

This is a rather common EBNF-grammar. A few peculiarities are noteworthy, though:
First of all you might notice that some components of the grammar
(or "prduction rules" as they are commonly called) have names with a leading
underscore "_". It is a convention to mark those elements, in which we are on
interested on their own account, with an underscore "_". When moving from the
concrete syntax-tree to a more abstract syntax-tree, these elements could be
substituted by their content, to simplify the tree.

Secondly, some production rules carry a name written in captial letters. This is also
a convention to mark those symbols which with other parser-generators would
represent tokens delivered by a lexical scanner. DHParser is a "scanner-less"
parser, which means that the breaking down of the string into meaningful tokens
is done in place with regular expressions (like in the definition of "_EOF")
or simple combinations of regular expressions (see the definition of "_INT" above).
Their is no sharp distinction between tokens and other symbols in DHParser,
but we keep it as a loose convention. Regular expressions are enclosed in forward
slashes and follow the standard syntax of Perl-style regular expression that is
also used by the "re"-module of the Python standard library. (Don't worry about
the number of backslashes in the line defining "_CHARS" for now!)

Finally, it is another helpful conention to indent the defintions of symbols
that have only been introduced to simplify an otherwise uneccessarily
complicated definition (e.g. the definition of "number", above) or to make
it more understandable by giving names to its componentns (like "_EOF").

Let's try this grammar on our test-string.  In order to compile
this grammar into executable Python-code, we use the high-level-function
:py:func:`create_parser()` from :py:mod:`DHParser.dsl`-module.

    >>> from DHParser.dsl import create_parser
    >>> # from DHParser.dsl import compileEBNF
    >>> # print(compileEBNF(grammar))
    >>> parser = create_parser(grammar, branding="JSON")
    >>> syntax_tree = parser(testdata)
    >>> syntax_tree.content
    '{"array": [1, 2.0, "a string"], "number": -1.3e+25, "bool": false}'

As expected serializing the content of the resulting syntax-tree yields exactly
the input-string of the parsing process. What we cannot see here, is that the
parser has structured the string into the individual elements described in the
grammar. Since the concrete syntax-tree that the parser vields is rather
verbose, it would not make sense to print it out. We'll just look at a small
part of it, to see what it looks like. Let's just pick the sub-tree that
captures the first json-array within the syntax-tree::

    >>> print(syntax_tree.pick('array').as_sxpr(compact=True))
    (array
      (:Text "[")
      (_element
        (number
          (_INT "1")))
      (:Text ",")
      (:Whitespace " ")
      (_element
        (number
          (_INT "2")
          (_FRAC
            (:Text ".")
            (:RegExp "0"))))
      (:Text ",")
      (:Whitespace " ")
      (_element
        (string
          (:Text '"')
          (_CHARS "a string")
          (:Text '"')))
      (:Text "]"))

The nodes of the syntax-tree carry the names of the production rules
by which they have been generated. Nodes, that have been created by
components of a prduction receive the name of of the parser-type
that has created the node (see :py:mod:`DHParser.parse`) prefixed
with a colon ":". In DHParser, these nodes are called "anonymous",
because they lack the name of a proper grammatical component.

Simplifying Syntax-Trees while Parsing
--------------------------------------

Usually, anonymous nodes are what you want to get rid of in the course
of transforming the concrete syntax-tree into an abstract syntax-tree.
(See :py:mod:`DHParser.transform`). DHParser already eliminates per
default all anonymous nodes that are not leaf-nodes by replacing them
with their children during parsing. Anonymous leaf-nodes will be
replaced by their content, if they are a single child of some parent,
and otherwise be left in place. Without this optimization, each
construct of the EBNF-grammar would leave a node in the syntax-tree::

    >>> from DHParser.parse import CombinedParser, TreeReduction
    >>> _ = TreeReduction(parser.json, CombinedParser.NO_TREE_REDUCTION)
    >>> syntax_tree = parser(testdata)
    >>> print(syntax_tree.pick('array').as_sxpr(compact=True))
    (array
      (:Text "[")
      (:Option
        (:Series
          (_element
            (number
              (_INT
                (:Alternative
                  (:RegExp "1")))))
          (:ZeroOrMore
            (:Series
              (:Text ",")
              (:Whitespace " ")
              (_element
                (number
                  (_INT
                    (:Alternative
                      (:RegExp "2")))
                  (:Option
                    (_FRAC
                      (:Text ".")
                      (:RegExp "0"))))))
            (:Series
              (:Text ",")
              (:Whitespace " ")
              (_element
                (string
                  (:Text '"')
                  (_CHARS
                    (:RegExp "a string"))
                  (:Text '"')))))))
      (:Text "]"))

This can be helpful for understanding how parsing that is directed by
an EBNF-grammar works (next to looking at the logs of the complete
parsing-process, see :py:mod:`DHParser.trace`), but other than that it
is advisable to streamline the syntax-tree as early on as possible,
because the processing time of all subsequent tree-processing stages
increases with the number of nodes in the tree.

Because of this, DHParser offers further means of simplifying
syntax-trees during the parsing stage, already. These are not turned
on by default, because they allow to drop content or to remove named
nodes from the tree; but they must be turned on by "directives" that
are listed at the top of an EBNF-grammar and that guide the
parser-generation process. DHParser-directives always start with an
`@`-sign. For example, the `@drop`-directive advises the parser to
drop certain nodes entirely, including their content. In the following
example, the parser is directed to drop all insignificant whitespace::

    >>> drop_insignificant_wsp = '@drop = whitespace \\n'

Directives look similar to productions, only that on the right hand
side of the equal sign follows a list of parameters. In the case
of the drop-directive these can be either names of non-anomymous
nodes that shall be dropped or one of four particular classes of
anonymous nodes (`strings`, `backticked`, `regexp`, `whitespace`) that
will be dropped.

Another useful directive advises the parser to treat named nodes as
anynouse nodes and to eliminate them accordingly during parsing. This
is usefule, if we have introduced certain names in our grammar
only as placeholders to render the definition of the grammar a bit
more readable, not because we are intested in the text that is
captured by the production associated with them in their own right::

    >>> disposable_symbols = '@disposable = /_\w+/ \\n'

Instead of passing a comma-separated list of symbols to the directive,
which would also have been possible, we have leveraged our convention
to prefix unimportant symbols with an underscore "_" by specifying the
symbols that shall by anonymized with a regular expression.

Now, let's examine the effect of these two directives::

    >>> refined_grammar = drop_insignificant_wsp + disposable_symbols + grammar
    >>> parser = create_parser(refined_grammar, 'JSON')
    >>> syntax_tree = parser(testdata)
    >>> syntax_tree.content
    '{"array":[1,2.0,"a string"],"number":-1.3e+25,"bool":false}'

You might have notived that all insigificant whitespaces adjacent to
the delimiters have been removed this time (but, of course not the
significant whitespace between "a" and "string" in "a string"). And
the difference, the use of these two directives makes, is even more
obvious, if we look at (a section of) the syntax-tree::

    >>> print(syntax_tree.pick('array').as_sxpr(compact=True))
    (array
      (:Text "[")
      (number "1")
      (:Text ",")
      (number
        (:RegExp "2")
        (:Text ".")
        (:RegExp "0"))
      (:Text ",")
      (string
        (:Text '"')
        (:RegExp "a string")
        (:Text '"'))
      (:Text "]"))

This tree looks more streamlined. But it still contains more structure
than we might like to see in an abstract syntax tree. In particular, it
still contains als the delimiters ("[", ",", '"', ...) next to the data. But
other than in the UTF-8 representation of our json data, the delimiters are
not needed any more, because the structural information is now retained
in the tree-structure.

So how can we get rid of those delimiters? The rather coarse-grained tools
that DHParser offers in the parsing stage require some care to do this
properly.

The @drop-directive allows to drop all unnamed strings (i.e. strings
that are not directly assigned to a symbol) and backticked strings (for
the difference between strings and backticked strings, see below) and
regular expressions. However, using `@drop = whitespace, strings, backticked`
would also drop those parts captured as string that contain data::

    >>> refined_grammar = '@drop = whitespace, strings, backticked \\n' \
                          + disposable_symbols + grammar
    >>> parser = create_parser(refined_grammar, 'JSON')
    >>> syntax_tree = parser(testdata)
    >>> print(syntax_tree.pick('array').as_sxpr(compact=True))
    (array
      (number "1")
      (number
        (:RegExp "2")
        (:RegExp "0"))
      (string "a string"))

Here, suddenly, the number "2.0" has been turned into "20"! There
are three ways to get around this problem:

1. Assigning all non-delmiter strings to symbols. In this case
   we would have to rewrite the definition of "number" as such::

      number     = _INT _FRAC? _EXP? ~
        _INT     = _MINUS? ( /[1-9][0-9]+/ | /[0-9]/ )
        _FRAC    = _DOT /[0-9]+/
        _EXP     = (_Ecap|_Esmall) [_PLUS|MINUS] /[0-9]+/
        _MINUS   = `-`
        _PLUS    = `+`
        _DOT     = `.`
        _Ecap    = `E`
        _Esmall  = `e`

   A simpler alternative of this technique would be to make use of
   the fact that the document-parts captured by regular expresseion
   are not dropped (although regular expressions can also be listed
   in the @drop-directive, if needed) and that at the same time
   delimiters are almost always simple strings containing keywords
   or punctuation characters. Thus, one only needs to rewrite those
   string-expressions that capture data as regular expressions::

      number     = _INT _FRAC? _EXP? ~
        _INT     = /[-]/ ( /[1-9][0-9]+/ | /[0-9]/ )
        _FRAC    = /[.]/ /[0-9]+/
        _EXP     = (/E/|/e/) [/[-+]/] /[0-9]+/

2. Assigning all delimiter strings to symbols and drop the nodes
   and content captured by these symbols. This means doing exactly
   the opposite of the first solution. Here is an excerpt of what
   a JSON-grammar emploing this technique would look like::

      @disposable = /_\w+/
      @drop = whitespace, _BEGIN_ARRAY, _END_ARRAY, _KOMMA, _BEGIN_OBJECT, ...
      ...
      array = _BEGIN_ARRAY ~ ( _element ( _KOMMA ~ _element )* )? §_END_ARRAY ~
      ...

   It is important that all symbols listed for dropping are also made
   disposable, either by listing them in the disposable-directive as well
   or using names that the regular-expressions for disposables matches.
   Otherwise, DHParser does not allow to drop the content of named nodes,
   because the default assumption is that symbols in the grammar are
   defined to capture meaningful parts of the document that contain
   relevant data.

3. Bailing out and leaving the further simplification of the syntax-tree
   to the next tree-processing stage which, if you follow DHParser's suggested
   usage pattern, is the abstract-syntax-tree-transformation proper
   and which allows for a much more fine-grained specification of
   transformation rules. See :py:mod:`DHParser.transformation`

To round this section up, we present the full grammar for a streamlined
JSON-Parser according to the first solution-strategy. Observe, that the
values of "bool" and "null" are now defined with regular expressions
instead of string-literals, because the latter would be dropped because
of the `@drop = ... strings, ...`-directive, leaving an empty named node
without a value, wheneever a bool value or null occurs in the input::

    >>> json_gr = '''
    ...      @disposable = /_\\\\w+/
    ...     @drop      = whitespace, strings, backticked, _EOF
    ...     json       = ~ _element _EOF
    ...       _EOF     = /$/
    ...     _element   = object | array | string | number | bool | null
    ...     object     = "{" ~ member ( "," ~ §member )* "}" ~
    ...     member     = string ":" ~ _element
    ...     array      = "[" ~ ( _element ( "," ~ _element )* )? "]" ~
    ...     string     = `"` _CHARS `"` ~
    ...       _CHARS   = /[^"\\\\\\]+/ | /\\\\\\[\\\\/bnrt\\\\\\]/
    ...     number     = _INT _FRAC? _EXP? ~
    ...       _INT     = /[-]/? ( /[1-9][0-9]+/ | /[0-9]/ )
    ...       _FRAC    = /[.]/ /[0-9]+/
    ...       _EXP     = /[Ee]/ [/[-+]/] /[0-9]+/
    ...     bool       = /true/ ~ | /false/ ~
    ...     null       = /null/ ~                                  '''
    >>> json_parser = create_parser(json_gr, 'JSON')
    >>> syntax_tree = json_parser(testdata)
    >>> print(syntax_tree.pick('array').as_sxpr(compact=True))
    (array
      (number "1")
      (number
        (:RegExp "2")
        (:RegExp ".")
        (:RegExp "0"))
      (string "a string"))

This time the data is not distorted, any more. One oddity reamins, however: We
are most probably not interested in the fact that the number 2.0 consists of
three components, each of which hast been captured by a regular expression.
Luckiliy, there exists yet another directive that allows to reduce the tree
further by merging adjacent anonymous leaf-nodes::

    >>> json_gr = '@reduction = merge \\n' + json_gr
    >>> json_parser = create_parser(json_gr, 'JSON')
    >>> syntax_tree = json_parser(testdata)
    >>> print(syntax_tree.as_sxpr(compact=True))
    (json
      (object
        (member
          (string "array")
          (array
            (number "1")
            (number "2.0")
            (string "a string")))
        (member
          (string "number")
          (number "-1.3e+25"))
        (member
          (string "bool")
          (bool "false"))))

Merging adjacent anonymous leaf-nodes takes place after the @drop-directive
comes into effect. It should be observed that merging only produces the desired
result, if any delimiters have been dropped previously, because otherwise
delimiters would be merged with content. Therefore, the `@reduction = merge`-
directive should at best only be applied in conjunction with the `@drop` and
`@dispose`-directives.

Applying any of the here described tree-reduction (or "simplification" for
that matter) requires a bit of careful planning concerning which nodes
will be named and which nodes will be dropped. This, however, pays off in
terms of speed and considerably simplified abtract-syntax-tree generation
stage, because most of the unnecessary structure of concrete-syntax-trees
has already been eliminated at the parsing stage.

Comments and Whitespace
-----------------------

Why whitespace isn't trivial
^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Handling whitespace in text-documents is not all trivial, because
whitespace can serve several different purposes and there can be
different kinds of whitespace: Whitespace can serve a syntactic function
as delimiter. But whitespace can also be purely asthetic to render
a document more readable.

Depending on the data model, whitespace can be considered as
significant and be included in the data or as
insignificant and be excluded from the data and only be re-inserted
when displaying the data in a human-readable-form. (For example, one
can model a sentence as a seuqence of words and spaces or just as
a sequence of words.) Note, that "significance" does not correlate
with the syntatic or asthetic function, but only depends on whether
you'd like to keep the whitespace in you data or not.

There can be different kinds of whitespace with different meaning
(and differing significance). For example, one can make a difference
between horozontal whitespace (spaces and tabs) and vertical
whitespace (including linefeeds). And there can be different sizes
of whitespace with different meaning. For example in LaTeX, a single
linefeed still counts as plain whitespace while an empty line (i.e.
whitespace including two or more not linefeeds) signals a new
paragraph.

Finally, even the position of whitespace can make a difference.
A certain number of whitespaces at the beginning of a line can
have the meaning of "indentation" (as in Python code) while at
the end of the line or between brackets it is just plain
insignificant whitespace. (This is actually something, where
the boundaries of the EBNF-formalism become visible and you'd
probably use a preprocessor or some kind of "semantic actions"
to handle such cases. There is some support for either of these
in DHParser.)

Coding significant Whitespace in EBNF-Grammars
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

A reasonable approach to coding whitespace is to use one
particular symbol for each kind of whitespace. Those kinds of
whitespace that are insignficant, i.e. that do not need to
appear in the data, should be dropped from the syntax-tree.
With DHParser this can be done already while parsing, using
the `@disposable` and `@drop`-directives described earlier.

But let's first look at an example which only includes significant
whitespace. The following parser parses sequences of paragraphs which
consist of sequences of sentences which consist of sequences
of main clauses and subordinate clauses which consist of sequences
of words::

    >>> text_gr = '''
    ...     @ disposable = /_\\\\w+/
    ...     document       = PBR* S? paragraph (PBR paragraph)* PBR* S? _EOF
    ...       _EOF         = /$/
    ...     paragraph      = sentence (S sentence)*
    ...     sentence       = (clause _c_delimiter S)* clause _s_delimiter
    ...       _c_delimiter = KOMMA | COLON | SEMICOLON
    ...       _s_delimiter = DOT | QUESTION_MARK | EXCLAMATION_MARK
    ...     clause         = word (S word)*
    ...     word           = /(?:[A-Z]|[a-z])[a-z']*/
    ...     DOT            = `.`
    ...     QUESTION_MARK  = `?`
    ...     EXCLAMATION_MARK = `!`
    ...     KOMMA          = `,`
    ...     COLON          = `:`
    ...     SEMICOLON      = `;`
    ...     PBR            = /[ \\\\t]*\\\\n[ \\\\t]*\\\\n[ \\\\t]*/
    ...     S              = /(?=[ \\\\n\\\\t])[ \\\\t]*(?:\\\\n[ \\\\t]*)?(?!\\\\n)/ '''

Here, we have two types of significant whitespace `PBR` ("paragraph-break") and `S`
("space"). Both types allow for a certain amount of flexibility, so that two
whitespaces of the same type do not need to have exactly the same content, but
we could always normalize these whitespaces in a subsequent transformation step.

Two typical design patterns for significant whitespace are noteworthy, here:

1. Both whitespaces match only if there was at least one whitespace character.
   We may allow whitespace to be optional (as at the beginning and end of the
   document), but if the option has not been taken, we don't to see an empty
   whitespace-tag in the document, later on.
   (For insignificant whitespace, the opposite convention can be more convenient,
   because, typically, insignificant whitespace is dropped anyway, whether it's
   got content or not.)

2. The grammar is construed in such a way that the whitespace always appears
   *between* different elements at the same level, but not after the last or
   before the first element. The whitespace after the last word of a sentence
   or before the first word of a sentence is really whitespace between
   two sentences. If we pick out a sentence or a clause, we will have no
   dangling whitespace at its beginning or end.
   (Again, for soon to be dropped insignificant whitespace, another convention
   can be more advisable.)

Let's just try our grammar on an example::

    >>> text_example = '''
    ... I want to say, in all seriousness, that a great deal of harm is being
    ... done in the modern world by belief in the virtuousness of work, and that
    ... the road to happiness and prosperity lies in an organized diminution of
    ... work.
    ...
    ... First of all: what is work? Work is of two kinds: first, altering the
    ... position of matter at or near the earth's surface relatively to other
    ... such matter; second, telling other people to do so. The first kind is
    ... unpleasant and ill paid; the second is pleasant and highly paid.'''
    >>> text_parser = create_parser(text_gr, 'Text')
    >>> text_as_data = text_parser(text_example)
    >>> sentence = text_as_data.pick(\
            lambda nd: nd.tag_name == "sentence" and nd.content.startswith('First'))
    >>> print(sentence.as_sxpr(compact=True))
    (sentence
      (clause
        (word "First")
        (S " ")
        (word "of")
        (S " ")
        (word "all"))
      (COLON ":")
      (S " ")
      (clause
        (word "what")
        (S " ")
        (word "is")
        (S " ")
        (word "work"))
      (QUESTION_MARK "?"))

Again, it is a question of design, whether we leave whitespace in the data or
not. Leaving it has the advantage, that serialization become as simple as
printing the content of the data-tree::

    >>> print(sentence)
    First of all: what is work?

Otherwise one would have to programm a dedicated serialization routine. Especially,
if you receive data from a different source, you'll appreciate not having to
do this - and so will other people, receiving your data. Think about it! However,
dropping the whitespace will yield more consice data.

Coding Comments
^^^^^^^^^^^^^^^

Allowing comments in a domain-specific language almost always makes sense,
because it allows users to annotate the source texts while working on them
and to share those comments with collaborators. From a technical point of
view, adding comments to a DSL raises two questions:

1. At what places shall we allow to insert comments in the source code?
   Common answers are: a) at the end of a line, b) almost everywhere, or
   c) both.

2. How do we avoid pollution of the EBNF-grammar with comment markers?
   It's already curtails the readability that we have to put whitespace
   symbols in so many places. And speaking of comments at the end of
   the line: If linefeeds aren't important for us - as in our toy-grammar
   for prose-text, above - we probably wouldn't want to reframe our
   grammar jsut to allow for at the end of the line comments.

Luckily, there exists a simple and highly intuitive solution that takes
care of both of these concerns: We admitt comments, whereever whitespace
is allowed. And we code this by defining a symbol that means: "whitespace
and, optionally, a comment".

Let's try this with our prose-text-grammar. In order to do so, we have
to define a symbols for comments, a symbol for pure whitespace, and,
finally, a symbol for whitespace with optional comment. Since, in
our grammar, we actually have two kinds of whitespace, `S` and `PBR`,
we'll have to redefine both of them. As delimiters for comments, we
use curly braces::

    >>> wsp_gr = '''
    ...     PBR      = pure_PBR COMMENT (pure_PBR | pure_S)?
    ...              | (pure_S? COMMENT)? pure_PBR
    ...     S        = pure_S COMMENT pure_S? | COMMENT? pure_S
    ...     COMMENT  = /\{[^}]*\}/
    ...     pure_PBR = /[ \\\\t]*\\\\n[ \\\\t]*\\\\n[ \\\\t]*/
    ...     pure_S   = /(?=[ \\\\n\\\\t])[ \\\\t]*(?:\\\\n[ \\\\t]*)?(?!\\\\n)/'''

As can be seen, the concrete re-definition of the whitespace tokens
requires a bit of careful consideration, because we want to allow
additional whitespace next to comments, but at the same time avoid
ending up with two whitespaces in sequence in our data. Let's see, if
we have succeeded::

    >>> extended_text_gr = text_gr[:text_gr.rfind(' PBR')] + wsp_gr
    >>> extended_parser = create_parser(extended_text_gr, 'Text')
    >>> syntax_tree = extended_parser('What {check this again!} is work?')
    >>> print(' '.join(nd.tag_name for nd in syntax_tree.pick('clause').children))
    word S word S word
    >>> print(syntax_tree.pick('clause').as_sxpr(compact=True))
    (clause
      (word "What")
      (S
        (pure_S " ")
        (COMMENT "{check this again!}")
        (pure_S " "))
      (word "is")
      (S
        (pure_S " "))
      (word "work"))

We will not worry about the more sub-structure of the S-nodes right now. If
we are not interested in the comments, we could use the `@disposable`,
`@drop` and `@reduction = merge`-directives to simplify these at the
parsing stage. Or, we could extract the comments and normalize the whitespace
at a later tree-processing stage. For now, let's just check wehter our
comments work as expected::

    >>> syntax_tree = extended_parser('What{check this again!} is work?')
    >>> print(' '.join(nd.tag_name for nd in syntax_tree.pick('clause').children))
    word S word S word
    >>> syntax_tree = extended_parser('What {check this again!}is work?')
    >>> print(' '.join(nd.tag_name for nd in syntax_tree.pick('clause').children))
    word S word S word
    >>> syntax_tree = extended_parser('What{check this again!}is work?')
    >>> print(syntax_tree.errors[0])
    1:1: Error (1040): Parser "document = {PBR} [S] paragraph {PBR paragraph} {PBR} [S] _EOF" did not match!

The last error was to be expected, because we did not allow comments
to serve a substitutes for whitespace. The error message might not be
as clear about the actual error as we might wish, though, but this is a topic
for later. Let's check whether putting comments near paragraph breaks
works as well::

    >>> test_text = '''Happiness lies in the diminuniation of work.
    ...
    ... { Here comes the comment }
    ...
    ... What is work?'''
    >>> syntax_tree = extended_parser(test_text)
    >>> print(' '.join(nd.tag_name for nd in syntax_tree.children))
    paragraph PBR paragraph
    >>> test_text = '''Happiness lies in the diminuniation of work.
    ... { Here comes the comment }
    ... What is work?'''
    >>> syntax_tree = extended_parser(test_text)
    >>> print(' '.join(nd.tag_name for nd in syntax_tree.children))
    paragraph

The last result might look surprising at first, but since a paragraph
break requires at least one empty line as a separator, the input text
is correctly understood by the parser as a sinlge paragrpah with
two sentence interspersed by a sinlge whitespace which, incidently,
contains a comment::

    >>> print(' '.join(nd.tag_name for nd in syntax_tree.pick('paragraph').children))
    sentence S sentence
    >>> print(syntax_tree.pick('paragraph')['S'].as_sxpr(compact=True))
    (S
      (pure_S
        ""
        "")
      (COMMENT "{ Here comes the comment }")
      (pure_S
        ""
        ""))

A common problem with whitespace is that it tends to pollute
the Grammar, because whereever you'd like to allow whitespace,
you'd have to insert a symbol for whitespace. The same problem
existis when it comes to allowing comments, because you'd
probably allow to insert comments in as many places as possible.

DHParser's support for insignificant whitespace and comments
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Coding insignificant whitespace and comments is exactly the
same as coding siginificant whitespace and comments and does not
need to be repeated, here. (The combination of insignificant
whitespace and significant comments, is slightly more complicated,
and probably best outsourced to some degree to the post-parsing
processing stages. It will not be discussed here.) However,
DHParser offers some special support for insignificant
whitesapce and comments, which can make working with these
easier in some cases.

First of all, DHParser has a special dedicated token for
insignificant whitespace which is the tilde `~`-character.
We have seen this earlier in the definition of the json-Grammar.

The `~` whitespace marker differs from the usual pattern for
defining whitespace in that it is implicitly optional, or what
amounts to the same, it matches the empty string. Normally,
it is to be considered bad practice to define a symbol as
optional. Rahter, a symbol should always match something and
only at the places where it is used, it should be marked as
optional. If this rule is obeyed, it is always easy to tell,
wether some element is optional or not at a specific place
in the Grammar. Otherwise, it can become quite confusing
indeed. However, since the tilde character is usually used
very often, it is more convenient not to mark it with a
question-mark or, if you use classical EBNF-syntax, to enclose
it with square brackets.

The default regular expression for the tilde-whitespace captures
arbitraily many spaces and tabs and at most one linefeed, but
not an empty line (`[ \t]*(?:\n[ \t]*)?(?!\n)`), as this is
the most convenient way to define whitespace for text-data.
However, the tilde whitespace can also be definied with any
other regular expression with the `@whitespace`-directive.

Let's go back to our JSON-grammar and define the optional
insignificant whitespace marked by the tilde-character in such a
way that it matches any amount of horizontal or vertical
whitespace, which makes much more sense in the context of json
than the default tilde-whitespace that is restricted vertically
to at most a single linefeed::

    >>> testdata = '{"array": [1, 2.0, "a string"], \\n\\n\\n "number": -1.3e+25, "bool": false}'
    >>> syntax_tree = json_parser(testdata)
    >>> print(syntax_tree.errors[0])
    1:32: Error (1010): member expected by parser 'object', » \\n \\n \\n  "numb...« found!
    >>> json_gr = '@whitespace = /\\\\s*/ \\n' + json_gr
    >>> json_parser = create_parser(json_gr, "JSON")
    >>> syntax_tree = json_parser(testdata)
    >>> print(syntax_tree.errors)
    []

When redefining the tilde-whitespace, make sure that your regular expression
also matches the empty string! There is no need to worry that the syntax tree
get's cluttered by empty whitespace-nodes, because tilde-whitespace always
yeidls anonymous nodes and DHParser drops empty anonymous nodes right away.

Comments can be defined using the `@comment`-directive. DHParser automatically
intermingles comments and whitespace so that where-ever tilde-whitespace is
allowed, a comment defined by the `@comment`-directive is also allowed:

    >>> json_gr = '@comment = /#[^\\\\n]*(?:\\\\n|$)/ \\n' + json_gr
    >>> json_parser = create_parser(json_gr, "JSON")
    >>> testdata = '''{"array": [1, 2.0, "a string"], # a string
    ...                "number": -1.3e+25,  # a number
    ...                "bool": false}  # a bool'''
    >>> syntax_tree = json_parser(testdata)
    >>> print(syntax_tree.as_sxpr(compact = True))
    (json
      (object
        (member
          (string "array")
          (array
            (number "1")
            (number "2.0")
            (string "a string")))
        (member
          (string "number")
          (number "-1.3e+25"))
        (member
          (string "bool")
          (bool "false"))))

Since the json-grammar still contains the `@drop = whitespace, ...`-
directive from earlier on (next to other tree-reductions), the comments
have been nicely dropped along with the tilde-whitespace.

There is one caveat: When using comments alongside with whitespace that
captures at most one linefeed, the comments should be defined in such
a way that the last charcter of a comment is never a linefeed.

Also a few limitations of the tilde-whitespace and directive-defined
comments should be kept in mind: 1. Only one kind of insignificant
whitespace can be defined this way. If there are more kinds of
insignificant whitespace, all but one need to be defined conventionally
as part of the grammar. 2. Both directive-defined comments and
tilde-whitespace can only be defined by a regular expresseion. In
particular, nested comments are impossible to define with regular
expressions, only.

However, using tilde-whitespace has yet one more benefit: With the
tilde-whitespace, cluttering of the grammar with whitespace-markers
can be avoid, by adding implicit whitespace adjacent to string-literals.
Remember the definition of the JSON-Grammar earlier. If you look at
a definition like: `object = "{" ~ member ( "," ~ §member )* "}" ~`,
you'll notice that there are three whitespace markers, one next to
each delimiter. Naturally so, because one usually wants to allow users
of a domain specific language to put whitespace around delimiters.

You may wonder, why the tilde appears only on the right hand side
of the literals, although you'd probably like to allow whitespace
on both side of a literal like "{". But if you look at the grammar
closely, you'll find that almost every symbol definition ends
either with a tilde sign or a symbol the definition of which ends
with a tilde sign, which means that they allow whitespace on the
right hand side. Now, if all elements of the grammar allow
whitespace on the right hand side, this means that automatically
also have whitespace on the left-hand side, too, which is, of
course the whitespace on the right hand side of the previous
element.

In order to reduce cluttering the grammar with tile-signs, DHParser
allows to turn on implicit tilde-whitespace adjacent to any
string literal with the diretive `@ literalws = right` or
`@ literalws = left`. As the argument of the directive suggests,
whitespace is either "eaten" at the right hand side or the left
hand side of the literal. String literals can either be
enclose in double quotes "..." or single quotes '...'. Both
kinds of literals will have implicit whitespace, if the
`@literalws`-directive is used.

(Don't confuse implicit whitespace
with insignificant whitespace: Insignificnat whitespace is whitespace
you do not need any more after parsing. Implicit whitespace is
whitespace you do not denote explicitly in the grammar. It's
a speciality of DHParser and DHParser allows onl the insignificant
whitespace denoted by the tilde-character to be declared as
"implicit".)

If left-adjacent whitespace is declared as implicit with the
`@literalws`-directive, the expression::

    object     = "{" ~ member ( "," ~ §member )* "}" ~

can be written as::

    object     = "{" member ( "," §member )* "}"

which is easier to read.

For situations where implicit whitespace is not desired, DHParser
has a special kind of string literal, written with backticks, which
never carries any implicit whitespace. This is important, when
literals are used for signs that enclose content, like the quotation
marks for the string literals in our JSON-Grammar::

    string     = `"` _CHARS '"'  # mind the difference between `"` and '"'!

Regular expressions, also, never carry implicit whitespace.
So, if you are using regular expressions as delimiters, you
still have to add the tilde character, if adjacent insignificant
whitespace is to be allowed::

    bool       = /true/~ | /false/~

The compliete json-grammar now looks like this::

    >>> json_gr = '''
    ...     @disposable = /_\\\\w+/
    ...     @drop      = whitespace, strings, backticked, _EOF
    ...     @reduction = merge
    ...     @whitespace= /\\\\s*/
    ...     @comment   = /#[^\\\\n]*(?:\\\\n|$)/
    ...     @literalws = right
    ...     json       = ~ _element _EOF
    ...       _EOF     = /$/
    ...     _element   = object | array | string | number | bool | null
    ...     object     = "{" member ( "," §member )* "}"
    ...     member     = string ":" _element
    ...     array      = "[" ( _element ( "," _element )* )? "]"
    ...     string     = `"` _CHARS '"'
    ...       _CHARS   = /[^"\\\\\\]+/ | /\\\\\\[\\\\/bnrt\\\\\\]/
    ...     number     = _INT _FRAC? _EXP? ~
    ...       _INT     = /[-]/? ( /[1-9][0-9]+/ | /[0-9]/ )
    ...       _FRAC    = /[.]/ /[0-9]+/
    ...       _EXP     = /[Ee]/ [/[-+]/] /[0-9]+/
    ...     bool       = /true/~ | /false/~
    ...     null       = /null/~                                    '''
    >>> json_parser = create_parser(json_gr, "JSON")
    >>> syntax_tree_ = json_parser(testdata)
    >>> assert syntax_tree_.equals(syntax_tree)

The whitespace defined by the `@whitespace`-directive can be access from
within the grammar via the name `WHITESPACE__`. Other than the tilde-sign
this name refers to the pure whitespace that is not intermingles with
comments. Similarly, comments defined by the `@comment`-directive can
be accessed via the symbol `COMMENT__`.

Lookahead and Lookbehind
------------------------

Lookahead and lookbehind operators are a convenient way to resolve or rather
avoid ambiguities while at the same time keeping the DSL lean. Assume for
example a simple DSL for writing definitions like::

    >>> definitions = '''
    ...     dog   := carnivorous quadrupel that barks
    ...     human := featherless biped'''

Now, let's try to draw up a grammar for definitions::

    >>> def_DSL_first_try = ''' # WARNING: This grammar doesn't work, yet!
    ...     @literalws = right
    ...     definitions = ~ definition { definition } EOF
    ...     definition  = definiendum ":=" definiens
    ...     definiendum = word
    ...     definiens   = word { word }
    ...     word        = /[A-Z]?[a-z]*/
    ...     EOF         = /$/ '''
    >>> def_parser = create_parser(def_DSL_first_try, "defDSL")



Fail-tolerant Parsing
---------------------



Semantic Actions and Storing Variables
--------------------------------------


"""


from collections import OrderedDict
from functools import partial
import keyword
import os
from typing import Callable, Dict, List, Set, FrozenSet, Tuple, Sequence, Union, Optional

from DHParser.compile import CompilerError, Compiler, ResultTuple, compile_source, visitor_name
from DHParser.configuration import access_thread_locals, get_config_value, \
    NEVER_MATCH_PATTERN, ALLOWED_PRESET_VALUES
from DHParser.error import Error, AMBIGUOUS_ERROR_HANDLING, WARNING, REDECLARED_TOKEN_WARNING,\
    REDEFINED_DIRECTIVE, UNUSED_ERROR_HANDLING_WARNING, INAPPROPRIATE_SYMBOL_FOR_DIRECTIVE, \
    DIRECTIVE_FOR_NONEXISTANT_SYMBOL, UNDEFINED_SYMBOL_IN_TRANSTABLE_WARNING, \
    UNCONNECTED_SYMBOL_WARNING, REORDERING_OF_ALTERNATIVES_REQUIRED, BAD_ORDER_OF_ALTERNATIVES, \
    EMPTY_GRAMMAR_ERROR, MALFORMED_REGULAR_EXPRESSION
from DHParser.parse import Parser, Grammar, mixin_comment, mixin_nonempty, Forward, RegExp, \
    Drop, Lookahead, NegativeLookahead, Alternative, Series, Option, ZeroOrMore, OneOrMore, \
    Text, Capture, Retrieve, Pop, optional_last_value, GrammarError, Whitespace, Always, Never, \
    INFINITE, matching_bracket, ParseFunc, update_scanner, CombinedParser
from DHParser.preprocess import nil_preprocessor, PreprocessorFunc
from DHParser.syntaxtree import Node, RootNode, WHITESPACE_PTYPE, TOKEN_PTYPE
from DHParser.toolkit import load_if_file, escape_re, escape_ctrl_chars, md5, \
    sane_parser_name, re, expand_table, unrepr, compile_python_object, DHPARSER_PARENTDIR, \
    cython
from DHParser.transform import TransformationFunc, traverse, remove_brackets, \
    reduce_single_child, replace_by_single_child, is_empty, remove_children, \
    remove_tokens, flatten, forbid, assert_content, remove_children_if, all_of, not_one_of, \
    BLOCK_LEAVES
from DHParser.versionnumber import __version__


__all__ = ('DHPARSER_IMPORTS',
           'get_ebnf_preprocessor',
           'get_ebnf_grammar',
           'get_ebnf_transformer',
           'get_ebnf_compiler',
           'parse_ebnf',
           'transform_ebnf',
           'compile_ebnf_ast',
           'EBNFGrammar',
           'EBNFTransform',
           'EBNFCompilerError',
           'EBNFDirectives',
           'EBNFCompiler',
           'grammar_changed',
           'compile_ebnf',
           'PreprocessorFactoryFunc',
           'ParserFactoryFunc',
           'TransformerFactoryFunc',
           'CompilerFactoryFunc')


########################################################################
#
# source code support
#
########################################################################


DHPARSER_IMPORTS = '''
import collections
from functools import partial
import os
import sys
from typing import Tuple, List, Union, Any, Optional, Callable

try:
    scriptpath = os.path.dirname(__file__)
except NameError:
    scriptpath = ''
dhparser_parentdir = os.path.abspath(os.path.join(scriptpath, r'{dhparser_parentdir}'))
if scriptpath not in sys.path:
    sys.path.append(scriptpath)
if dhparser_parentdir not in sys.path:
    sys.path.append(dhparser_parentdir)

try:
    import regex as re
except ImportError:
    import re
from DHParser import start_logging, suspend_logging, resume_logging, is_filename, load_if_file, \\
    Grammar, Compiler, nil_preprocessor, PreprocessorToken, Whitespace, Drop, AnyChar, \\
    Lookbehind, Lookahead, Alternative, Pop, Text, Synonym, Counted, Interleave, INFINITE, \\
    Option, NegativeLookbehind, OneOrMore, RegExp, Retrieve, Series, Capture, TreeReduction, \\
    ZeroOrMore, Forward, NegativeLookahead, Required, CombinedParser, mixin_comment, \\
    compile_source, grammar_changed, last_value, matching_bracket, PreprocessorFunc, is_empty, \\
    remove_if, Node, TransformationFunc, TransformationDict, transformation_factory, traverse, \\
    remove_children_if, move_adjacent, normalize_whitespace, is_anonymous, matches_re, \\
    reduce_single_child, replace_by_single_child, replace_or_reduce, remove_whitespace, \\
    replace_by_children, remove_empty, remove_tokens, flatten, all_of, any_of, \\
    merge_adjacent, collapse, collapse_children_if, transform_content, WHITESPACE_PTYPE, \\
    TOKEN_PTYPE, remove_children, remove_content, remove_brackets, change_tag_name, \\
    remove_anonymous_tokens, keep_children, is_one_of, not_one_of, has_content, apply_if, peek, \\
    remove_anonymous_empty, keep_nodes, traverse_locally, strip, lstrip, rstrip, \\
    transform_content, replace_content_with, forbid, assert_content, remove_infix_operator, \\
    add_error, error_on, recompile_grammar, left_associative, lean_left, set_config_value, \\
    get_config_value, node_maker, access_thread_locals, access_presets, PreprocessorResult, \\
    finalize_presets, ErrorCode, RX_NEVER_MATCH, set_tracer, resume_notices_on, \\
    trace_history, has_descendant, neg, has_ancestor, optional_last_value, insert, \\
    positions_of, replace_tag_names, add_attributes, delimit_children, merge_connected, \\
    has_attr, has_parent, ThreadLocalSingletonFactory, Error, canonical_error_strings, \\
    has_errors, ERROR, FATAL, set_preset_value, get_preset_value, NEVER_MATCH_PATTERN, \\
    gen_find_include_func, preprocess_includes, make_preprocessor, chain_preprocessors
'''


########################################################################
#
# EBNF scanning
#
########################################################################


def get_ebnf_preprocessor() -> PreprocessorFunc:
    """
    Returns the preprocessor function for the EBNF compiler.
    As of now, no preprocessing is needed for EBNF-sources. Therefore,
    just a dummy function is returned.
    """
    return nil_preprocessor


########################################################################
#
# EBNF parsing
#
########################################################################


class EBNFGrammar(Grammar):
    r"""Parser for a FlexibleEBNF source file.

    This grammar is tuned for flexibility, that is, it supports as many
    different flavors of EBNF as possible. However, this flexibility
    comes at the cost of some ambiguities. In particular:

       1. the alternative OR-operator / could be mistaken for the start
          of a regular expression and vice versa, and
       2. character ranges [a-z] can be mistaken for optional blocks
          and vice versa

    A strategy to avoid these ambiguities is to do all of the following:

        - replace the free_char-parser by a never matching parser
        - if this is done, it is safe to replace the char_range_heuristics-
          parser by an always matching parser
        - replace the regex_heuristics by an always matching parser

    Ambiguities can also be avoided by NOT using all the syntactic variants
    made possible by this EBNF-grammar within one and the same EBNF-document.

    EBNF-definition of the Grammar::

        @ comment    = /(?!#x[A-Fa-f0-9])#.*(?:\n|$)|\/\*(?:.|\n)*?\*\/|\(\*(?:.|\n)*?\*\)/
        # comments can be either C-Style: /* ... */
        # or pascal/modula/oberon-style: (* ... *)
        # or python-style: # ... \n,
        # excluding, however, character markers: #x20

        @ whitespace = /\s*/                            # whitespace includes linefeeds
        @ literalws  = right                            # trailing whitespace of literals will be ignored tacitly
        @ disposable = pure_elem, countable, FOLLOW_UP, SYM_REGEX, ANY_SUFFIX, EOF
        @ drop       = whitespace, EOF                  # do not include these even in the concrete syntax tree
        @ RNG_BRACE_filter = matching_bracket()         # filter or transform content of RNG_BRACE on retrieve

        # re-entry-rules for resuming after parsing-error

        @ definition_resume = /\n\s*(?=@|\w+\w*\s*=)/
        @ directive_resume  = /\n\s*(?=@|\w+\w*\s*=)/

        # specialized error messages for certain cases

        @ definition_error  = /,/, 'Delimiter "," not expected in definition!\nEither this was meant to '
                                   'be a directive and the directive symbol @ is missing\nor the error is '
                                   'due to inconsistent use of the comma as a delimiter\nfor the elements '
                                   'of a sequence.'


        #: top-level

        syntax     = ~ { definition | directive } EOF
        definition = symbol §:DEF~ [ :OR~ ] expression :ENDL~ & FOLLOW_UP  # [:OR~] to support v. Rossum's syntax

        directive  = "@" §symbol "=" (regexp | literals | procedure | symbol !DEF)
                     { "," (regexp | literals | procedure | symbol !DEF) } & FOLLOW_UP
        literals   = { literal }+                       # string chaining, only allowed in directives!
        procedure  = SYM_REGEX "()"                     # procedure name, only allowed in directives!

        FOLLOW_UP  = `@` | symbol | EOF


        #: components

        expression = sequence { :OR~ sequence }
        sequence   = ["§"] ( interleave | lookaround )  # "§" means all following terms mandatory
                     { :AND~ ["§"] ( interleave | lookaround ) }
        interleave = difference { "°" ["§"] difference }
        lookaround = flowmarker § (oneormore | pure_elem)
        difference = term ["-" § (oneormore | pure_elem)]
        term       = oneormore | counted | repetition | option | pure_elem


        #: elements

        countable  = option | oneormore | element
        pure_elem  = element § !ANY_SUFFIX              # element strictly without a suffix
        element    = [retrieveop] symbol !:DEF          # negative lookahead to be sure it's not a definition
                   | literal
                   | plaintext
                   | regexp
                   | char_range
                   | character ~
                   | any_char
                   | whitespace
                   | group


        ANY_SUFFIX = /[?*+]/


        #: flow-operators

        flowmarker = "!"  | "&"                         # '!' negative lookahead, '&' positive lookahead
                   | "<-!" | "<-&"                      # '<-' negative lookbehind, '<-&' positive lookbehind
        retrieveop = "::" | ":?" | ":"                  # '::' pop, ':?' optional pop, ':' retrieve


        #: groups

        group      = "(" no_range §expression ")"
        oneormore  = "{" no_range expression "}+" | element "+"
        repetition = "{" no_range §expression "}" | element "*" no_range
        option     = !char_range "[" §expression "]" | element "?"
        counted    = countable range | countable :TIMES~ multiplier | multiplier :TIMES~ §countable

        range      = RNG_BRACE~ multiplier [ :RNG_DELIM~ multiplier ] ::RNG_BRACE~
        no_range   = !multiplier | &multiplier :TIMES
        multiplier = /[1-9]\d*/~


        #: leaf-elements

        symbol     = SYM_REGEX ~                        # e.g. expression, term, parameter_list
        literal    = /"(?:(?<!\\)\\"|[^"])*?"/~         # e.g. "(", '+', 'while'
                   | /'(?:(?<!\\)\\'|[^'])*?'/~         # whitespace following literals will be ignored tacitly.
        plaintext  = /`(?:(?<!\\)\\`|[^`])*?`/~         # like literal but does not eat whitespace
                   | /´(?:(?<!\\)\\´|[^´])*?´/~
        regexp     = :RE_LEADIN RE_CORE :RE_LEADOUT ~   # e.g. /\w+/, ~/#.*(?:\n|$)/~
        # regexp     = /\/(?:(?<!\\)\\(?:\/)|[^\/])*?\//~     # e.g. /\w+/, ~/#.*(?:\n|$)/~
        char_range = `[` &char_range_heuristics
                         [`^`] (character | free_char) { [`-`] character | free_char } "]"
        character  = :CH_LEADIN HEXCODE
        free_char  = /[^\n\[\]\\]/ | /\\[nrt`´'"(){}\[\]\/\\]/
        any_char   = "."
        whitespace = /~/~                               # insignificant whitespace

        #: delimiters

        EOF = !/./ [:?DEF] [:?OR] [:?AND] [:?ENDL]      # [:?DEF], [:?OR], ... clear stack by eating stored value
                   [:?RNG_DELIM] [:?BRACE_SIGN] [:?CH_LEADIN] [:?TIMES] [:?RE_LEADIN] [:?RE_LEADOUT]

        DEF        = `=` | `:=` | `::=` | `<-` | /:\n/ | `: `  # if `: `, retrieve marker mustn't be followed by blank!
        OR         = `|` | `/` !regex_heuristics
        AND        = `,` | ``
        ENDL       = `;` | ``

        RNG_BRACE  = :BRACE_SIGN
        BRACE_SIGN = `{` | `(`
        RNG_DELIM  = `,`
        TIMES      = `*`

        RE_LEADIN  = `/` &regex_heuristics | `^/`
        RE_LEADOUT = `/`

        CH_LEADIN  = `0x` | `#x`

        #: heuristics

        char_range_heuristics  = ! ( /[\n\t ]/
                                   | ~ literal_heuristics
                                   | [`::`|`:?`|`:`] SYM_REGEX /\s*\]/ )
        literal_heuristics     = /~?\s*"(?:[\\]\]|[^\]]|[^\\]\[[^"]*)*"/
                               | /~?\s*'(?:[\\]\]|[^\]]|[^\\]\[[^']*)*'/
                               | /~?\s*`(?:[\\]\]|[^\]]|[^\\]\[[^`]*)*`/
                               | /~?\s*´(?:[\\]\]|[^\]]|[^\\]\[[^´]*)*´/
                               | /~?\s*\/(?:[\\]\]|[^\]]|[^\\]\[[^\/]*)*\//
        regex_heuristics       = /[^ ]/ | /[^\/\n*?+\\]*[*?+\\][^\/\n]\//


        #: basic-regexes

        RE_CORE    = /(?:(?<!\\)\\(?:\/)|[^\/])*/       # core of a regular expression, i.e. the dots in /.../
        SYM_REGEX  = /(?!\d)\w+/                        # regular expression for symbols
        HEXCODE    = /[A-Fa-f0-9]{1,8}/
    """
    countable = Forward()
    element = Forward()
    expression = Forward()
    source_hash__ = "3bda01686407a47a9fd0a709bda53ae3"
    disposable__ = re.compile('component$|pure_elem$|countable$|FOLLOW_UP$|SYM_REGEX$|ANY_SUFFIX$|EOF$')
    static_analysis_pending__ = []  # type: List[bool]
    parser_initialization__ = ["upon instantiation"]
    error_messages__ = {'definition': [
        (re.compile(r','),
        'Delimiter "," not expected in definition!\\nEither this was meant to be a directive '
        'and the directive symbol @ is missing\\nor the error is due to inconsistent use of the '
        'comma as a delimiter\\nfor the elements of a sequence.')]}
    resume_rules__ = {'definition': [re.compile(r'\n\s*(?=@|\w+\w*\s*=)')],
                      'directive': [re.compile(r'\n\s*(?=@|\w+\w*\s*=)')]}
    COMMENT__ = r'(?!#x[A-Fa-f0-9])#.*(?:\n|$)|\/\*(?:.|\n)*?\*\/|\(\*(?:.|\n)*?\*\)'
    comment_rx__ = re.compile(COMMENT__)
    WHITESPACE__ = r'\s*'
    WSP_RE__ = mixin_comment(whitespace=WHITESPACE__, comment=COMMENT__)
    wsp__ = Whitespace(WSP_RE__)
    dwsp__ = Drop(Whitespace(WSP_RE__))
    HEXCODE = RegExp('[A-Fa-f0-9]{1,8}')
    SYM_REGEX = RegExp('(?!\\d)\\w+')
    RE_CORE = RegExp('(?:(?<!\\\\)\\\\(?:/)|[^/])*')
    regex_heuristics = Alternative(RegExp('[^ ]'), RegExp('[^/\\n*?+\\\\]*[*?+\\\\][^/\\n]/'))
    literal_heuristics = Alternative(RegExp('~?\\s*"(?:[\\\\]\\]|[^\\]]|[^\\\\]\\[[^"]*)*"'),
                                     RegExp("~?\\s*'(?:[\\\\]\\]|[^\\]]|[^\\\\]\\[[^']*)*'"),
                                     RegExp('~?\\s*`(?:[\\\\]\\]|[^\\]]|[^\\\\]\\[[^`]*)*`'),
                                     RegExp('~?\\s*´(?:[\\\\]\\]|[^\\]]|[^\\\\]\\[[^´]*)*´'),
                                     RegExp('~?\\s*/(?:[\\\\]\\]|[^\\]]|[^\\\\]\\[[^/]*)*/'))
    char_range_heuristics = NegativeLookahead(Alternative(
        RegExp('[\\n\\t ]'), Series(dwsp__, literal_heuristics),
        Series(Option(Alternative(Text("::"), Text(":?"), Text(":"))),
               SYM_REGEX, RegExp('\\s*\\]'))))
    CH_LEADIN = Capture(Alternative(Text("0x"), Text("#x")))
    RE_LEADOUT = Capture(Text("/"))
    RE_LEADIN = Capture(Alternative(Series(Text("/"), Lookahead(regex_heuristics)), Text("^/")))
    TIMES = Capture(Text("*"))
    RNG_DELIM = Capture(Text(","))
    BRACE_SIGN = Capture(Alternative(Text("{"), Text("(")))
    RNG_BRACE = Capture(Retrieve(BRACE_SIGN))
    ENDL = Capture(Alternative(Text(";"), Text("")))
    AND = Capture(Alternative(Text(","), Text("")))
    OR = Capture(Alternative(Text("|"), Series(Text("/"), NegativeLookahead(regex_heuristics))))
    DEF = Capture(Alternative(Text("="), Text(":="), Text("::="),
                              Text("<-"), RegExp(':\\n'), Text(": ")))
    EOF = Drop(Drop(Series(Drop(NegativeLookahead(RegExp('.'))),
                           Drop(Option(Drop(Pop(DEF, match_func=optional_last_value)))),
                           Drop(Option(Drop(Pop(OR, match_func=optional_last_value)))),
                           Drop(Option(Drop(Pop(AND, match_func=optional_last_value)))),
                           Drop(Option(Drop(Pop(ENDL, match_func=optional_last_value)))),
                           Drop(Option(Drop(Pop(RNG_DELIM, match_func=optional_last_value)))),
                           Drop(Option(Drop(Pop(BRACE_SIGN, match_func=optional_last_value)))),
                           Drop(Option(Drop(Pop(CH_LEADIN, match_func=optional_last_value)))),
                           Drop(Option(Drop(Pop(TIMES, match_func=optional_last_value)))),
                           Drop(Option(Drop(Pop(RE_LEADIN, match_func=optional_last_value)))),
                           Drop(Option(Drop(Pop(RE_LEADOUT, match_func=optional_last_value)))))))
    whitespace = Series(RegExp('~'), dwsp__)
    any_char = Series(Text("."), dwsp__)
    free_char = Alternative(RegExp('[^\\n\\[\\]\\\\]'), RegExp('\\\\[nrt`´\'"(){}\\[\\]/\\\\]'))
    character = Series(Retrieve(CH_LEADIN), HEXCODE)
    char_range = Series(Text("["), Lookahead(char_range_heuristics), Option(Text("^")),
                        Alternative(character, free_char),
                        ZeroOrMore(Alternative(Series(Option(Text("-")), character), free_char)),
                        Series(Text("]"), dwsp__))
    regexp = Series(Retrieve(RE_LEADIN), RE_CORE, Retrieve(RE_LEADOUT), dwsp__)
    plaintext = Alternative(Series(RegExp('`(?:(?<!\\\\)\\\\`|[^`])*?`'), dwsp__),
                            Series(RegExp('´(?:(?<!\\\\)\\\\´|[^´])*?´'), dwsp__))
    literal = Alternative(Series(RegExp('"(?:(?<!\\\\)\\\\"|[^"])*?"'), dwsp__),
                          Series(RegExp("'(?:(?<!\\\\)\\\\'|[^'])*?'"), dwsp__))
    symbol = Series(SYM_REGEX, dwsp__)
    multiplier = Series(RegExp('[1-9]\\d*'), dwsp__)
    no_range = Alternative(NegativeLookahead(multiplier),
                           Series(Lookahead(multiplier), Retrieve(TIMES)))
    range = Series(RNG_BRACE, dwsp__, multiplier,
                   Option(Series(Retrieve(RNG_DELIM), dwsp__, multiplier)),
                   Pop(RNG_BRACE, match_func=matching_bracket), dwsp__)
    counted = Alternative(Series(countable, range),
                          Series(countable, Retrieve(TIMES), dwsp__, multiplier),
                          Series(multiplier, Retrieve(TIMES), dwsp__, countable, mandatory=3))
    option = Alternative(Series(NegativeLookahead(char_range), Series(Text("["), dwsp__),
                                expression, Series(Text("]"), dwsp__), mandatory=2),
                         Series(element, Series(Text("?"), dwsp__)))
    repetition = Alternative(Series(Series(Text("{"), dwsp__), no_range,
                                    expression, Series(Text("}"), dwsp__), mandatory=2),
                             Series(element, Series(Text("*"), dwsp__), no_range))
    oneormore = Alternative(Series(Series(Text("{"), dwsp__), no_range, expression,
                                   Series(Text("}+"), dwsp__)),
                            Series(element, Series(Text("+"), dwsp__)))
    group = Series(Series(Text("("), dwsp__), no_range,
                   expression, Series(Text(")"), dwsp__), mandatory=2)
    retrieveop = Alternative(Series(Text("::"), dwsp__),
                             Series(Text(":?"), dwsp__),
                             Series(Text(":"), dwsp__))
    flowmarker = Alternative(Series(Text("!"), dwsp__), Series(Text("&"), dwsp__),
                             Series(Text("<-!"), dwsp__), Series(Text("<-&"), dwsp__))
    ANY_SUFFIX = RegExp('[?*+]')
    element.set(Alternative(Series(Option(retrieveop), symbol, NegativeLookahead(Retrieve(DEF))),
                            literal, plaintext, regexp, char_range, Series(character, dwsp__),
                            any_char, whitespace, group))
    pure_elem = Series(element, NegativeLookahead(ANY_SUFFIX), mandatory=1)
    countable.set(Alternative(option, oneormore, element))
    term = Alternative(oneormore, counted, repetition, option, pure_elem)
    difference = Series(term, Option(Series(Series(Text("-"), dwsp__),
                                            Alternative(oneormore, pure_elem), mandatory=1)))
    lookaround = Series(flowmarker, Alternative(oneormore, pure_elem), mandatory=1)
    interleave = Series(difference, ZeroOrMore(Series(Series(Text("°"), dwsp__),
                                                      Option(Series(Text("§"), dwsp__)),
                                                      difference)))
    sequence = Series(Option(Series(Text("§"), dwsp__)), Alternative(interleave, lookaround),
                      ZeroOrMore(Series(Retrieve(AND), dwsp__, Option(Series(Text("§"), dwsp__)),
                                        Alternative(interleave, lookaround))))
    expression.set(Series(sequence, ZeroOrMore(Series(Retrieve(OR), dwsp__, sequence))))
    FOLLOW_UP = Alternative(Text("@"), symbol, EOF)
    procedure = Series(SYM_REGEX, Series(Text("()"), dwsp__))
    literals = OneOrMore(literal)
    component = Alternative(regexp, literals, procedure, Series(symbol, NegativeLookahead(DEF)))
    directive = Series(
        Series(Text("@"), dwsp__), symbol, Series(Text("="), dwsp__),
        Alternative(Series(component, ZeroOrMore(Series(Series(Text(","), dwsp__), component))),
                    expression),
        Lookahead(FOLLOW_UP), mandatory=1)
    definition = Series(symbol, Retrieve(DEF), dwsp__, Option(Series(Retrieve(OR), dwsp__)),
                        expression, Retrieve(ENDL), dwsp__, Lookahead(FOLLOW_UP), mandatory=1)
    syntax = Series(dwsp__, ZeroOrMore(Alternative(definition, directive)), EOF)
    root__ = syntax

    def __init__(self, root: Parser = None, static_analysis: Optional[bool] = None) -> None:
        Grammar.__init__(self, root, static_analysis)
        self.free_char_parsefunc__ = self.free_char._parse
        self.char_range_heuristics_parsefunc__ = self.char_range_heuristics._parse
        self.regex_heuristics_parserfunc__ = self.regex_heuristics._parse

    @property
    def mode(self) -> str:
        def which(p: Parser) -> str:
            if p._parse_proxy.__qualname__ == 'Never._parse':
                return 'never'
            elif p._parse_proxy.__qualname__ == 'Always._parse':
                return 'always'
            else:
                return 'custom'
        signature = (
            which(self.free_char),
            which(self.regex_heuristics),
            which(self.char_range_heuristics)
        )
        if signature == ('custom', 'custom', 'custom'):
            return 'heuristic'
        elif signature == ('never', 'always', 'always'):
            return 'strict'  # or 'classic'
        elif signature == ('custom', 'never', 'always'):
            return 'peg-like'
        elif signature == ('custom', 'always', 'always'):
            return 'regex-like'
        else:
            return "undefined"

    @mode.setter
    def mode(self, mode: str):
        if mode == self.mode:
            return

        def set_parsefunc(p: Parser, f: ParseFunc):
            method = f.__get__(p, type(p))  # bind function f to parser p
            p._parse_proxy = method

        always = Always._parse
        never = Never._parse
        if mode == 'heuristic':
            set_parsefunc(self.free_char, self.free_char_parsefunc__)
            set_parsefunc(self.regex_heuristics, self.regex_heuristics_parserfunc__)
            set_parsefunc(self.char_range_heuristics, self.char_range_heuristics_parsefunc__)
        elif mode in ('strict', 'classic'):
            set_parsefunc(self.free_char, never)
            set_parsefunc(self.regex_heuristics, always)
            set_parsefunc(self.char_range_heuristics, always)
        elif mode == 'peg-like':
            set_parsefunc(self.free_char, self.free_char_parsefunc__)
            set_parsefunc(self.regex_heuristics, never)
            set_parsefunc(self.char_range_heuristics, always)
        elif mode == 'regex-like':
            set_parsefunc(self.free_char, self.free_char_parsefunc__)
            set_parsefunc(self.regex_heuristics, always)
            set_parsefunc(self.char_range_heuristics, always)
        else:
            raise ValueError('Mode must be one of: ' + ', '.join(
                ALLOWED_PRESET_VALUES['syntax_variant']))


class FixedEBNFGrammar(Grammar):
    r"""Faster version of EBNF, where delimiters are not determined on
    first use, but defined as constant Text-parsers. They can still be
    adjusted with function `parse.update_scanner()`.

    Different syntactical variants can be configured either by adjusting
    the definitions of DEF, OR, AND, ENDL, RNG_OPEN, RNG_CLOSE, RNG_DELIM,