It's a revolution baby.HEADmaster

1 files changed, 595 insertions, 0 deletions

diff --git a/doc/0/1-parse.md b/doc/0/1-parse.md
new file mode 100644
index 0000000..101a3b6
--- /dev/null
+++ b/doc/0/1-parse.md
@@ -0,0 +1,595 @@
+# Parser for Code & Data
+
+<!--TOC-->
+
+Zisp s-expressions represent an extremely minimal set of data types; only that
+which is necessary to strategically construct more complex values:
+
+    +---------+--------+----------+------+
+    | String  | Rune   | List     | Nil  |
+    +---------+--------+----------+------+
+    | foobar  | #name  | (X ...)  | ()   |
+    +---------+--------+----------+------+
+
+The parser recognizes various *syntax sugar* which abbreviates verbose syntax,
+and may result in special data structures (typically, a list with a rune in its
+first position) which another Zisp component called the *decoder* can transform
+into a rich set of value types.
+
+More details about syntax sugar, and the decoder, are explained later.
+
+
+## Character Encoding
+
+The parser does not consume Unicode characters; it consumes bytes.  Grammar is
+generally constructed by bytes corresponding to ASCII characters.
+
+Some elements of the grammar, such as comments and quoted strings, may contain
+arbitrary byte sequences, until terminated.  These sequences may happen to be
+valid UTF-8 text.  This way, quoted strings and comments may contain Unicode
+text encoded in UTF-8, but the parser does not check these for validity.
+
+Since comments and quoted strings may contain arbitrary byte sequences, a text
+editor or other program displaying Zisp s-expressions may need to use a special
+visual representation for bytes that don't represent valid text.
+
+The parser working on bytes rather than Unicode characters is not a limitation,
+but rather a feature: It allows Zisp s-expressions to be used as a structured
+data exchange format, which may contain binary data elements, without the need
+to encode these in Base64 or other such text representations of binary data.
+Consider the example:
+
+    ((image.webp "<BINARY>")
+     (video.webm "<BINARY>"))
+
+All that needs to be done for this to work, is that any incidental occurrences
+of the double-quote sign, and the backslash sign, are escaped with a backslash
+within the `<BINARY>` data; all other bytes can appear verbatim in the strings.
+
+
+## Stream Parsing
+
+The parser can be repeatedly invoked on a byte stream to consume the next datum
+within.  This does not require "unreading" or back-seeking within the stream;
+the parser always reads a full datum, and stops after some byte which cleanly
+terminates the currently parsed datum.
+
+This means Zisp s-expressions can be safely intermixed with other data within
+the same byte stream.  So long as the other data is consumed by some parser
+which similarly stops reading at a clear boundary, the Zisp parser can then
+continue operating on the same stream.  Consider the example:
+
+    ("image.webp" 8273)
+
+    << 8273 bytes >>
+
+    ("video.webm" 736)
+
+    << 736 bytes >>
+
+The "header" for each file in this stream is a Zisp s-expression containing
+information about how many bytes should be read after the header, before the
+next file header appears.  (The header data need to be terminated with a blank
+ASCII character such as a newline; the closing parenthesis does not act as a
+terminator unto itself due to the "join" syntax sugar.)
+
+To enable this stream parsing strategy, the parser does not use any automatic
+buffering.  If it did, it might inadvertently consume some bytes beyond the
+currently parsed datum, leaving the stream inconsistent.
+
+If the parser is meant to be used on an input stream associated with expensive
+system calls, such as a file handle or network socket, it's best to wrap that
+stream in some intermediate object which asks the system for large chunks of
+data at once, and stores the data in a buffer.
+
+
+## Comments
+
+Two types of comment are supported: datum comments and line comments.
+
+* A semicolon followed by a tilde instructs the parser to consume one datum and
+  discard it.  Whitespace may appear between the tilde and the datum to discard.
+
+* A semicolon, followed by a non-tilde byte, instructs the parser to consume and
+  discard bytes until a newline (ASCII Line Feed) is encountered.
+
+
+## Value vs. Datum
+
+A Zisp *value* that has an *external representation* in the form of a sequence
+of bytes is called a *datum*.  Every datum is a value, but not every value is a
+datum.  In other words, a datum is a value that can be printed out as a byte
+sequence which the parser can turn back into an equivalent datum.
+
+A value that is not a datum may nevertheless be *encoded* into one, allowing it
+to have an external representation.  After parsing, it needs to be *decoded* to
+actually become the expected value.
+
+One may speak of an *external representation of a value* where the value is not
+itself a datum, but has an encoding as one.  The more strictly correct term for
+this is: "The external representation of the datum encoding the value."
+
+### Syntax sugar
+
+The parser recognizes various *syntax sugar* to abbreviate an equivalent datum
+construction, or express a datum that encodes a more complex value.
+
+As an example, the expression `#(x y z)` is an abbreviation for the equivalent
+`(#HASH x y z)`.  These are two external representations for the same datum;
+after parsing, both will yield values that are indistinguishable in all but
+their memory address.
+
+An example of syntax sugar that is not a mere abbreviation is a quoted string
+which contains bytes that could not appear in a *bare* string:
+
+    "foo bar"  ->  (#DQUOTE <STRING>)
+
+In this example, the visual token `<STRING>` represents the actual string value
+in program memory, which has no direct external representation in bytes because
+it contains a space character.
+
+Those familiar with Lisp and Scheme may expect bare strings to be parsed into a
+separate type called *symbol* while quoted strings are parsed directly into a
+string type, but this is not the case in Zisp.
+
+### Decoder
+
+The *decoder* transforms Zisp data into values of more complex types, including
+values that are not of a datum type.
+
+Combined with syntax sugar, this allows Zisp to offer familiar syntax elements.
+For example, the expression `#(x y z)` which parses into `(#HASH x y z)` can be
+decoded into an array, so the result is similar to the vector syntax of Scheme.
+
+Decoding also resolves datum labels, goes over bare strings to find ones that
+represent a number literal, and takes care of a number of other transforms.
+This offloads complexity, allowing the parser to remain extremely simple.
+
+See the dedicated documentation of the [decoder](2-decode.html) for more.
+
+
+## Data types
+
+Following is a more in-depth explanation of each data type constructed by the
+Zisp s-expression parser.
+
+These are in fact value types, though the term "data type" is often used due to
+familiarity.  A Zisp value that is a member of one of the following value types
+is only a *datum* if it adheres to additional constraints as explained below.
+
+### String
+
+Strings can appear *bare* or be quoted in various ways.  A quoted string is in
+fact parsed into a list value with a rune in the first position to identify the
+quotation variant that was parsed, and the string value in the second position;
+or, in case of at-quoted strings, a special construct we will look at later.
+
+    +-----------+-------------------------------+
+    | Syntax    | Parse output                  |
+    +-----------+-------------------------------+
+    | |bytes|   | (#PQSTR <STRING>)             |
+    +-----------+-------------------------------+
+    | "bytes"   | (#DQSTR <STRING>)             |
+    +-----------+-------------------------------+
+    | @_bytes_  | (#ATSTR <SENTINEL> <STRING>)  |
+    +-----------+-------------------------------+
+
+The visual token `<STRING>` denotes the actual string, as a Zisp value, in the
+second position of the list.  The visual token `<SENTINEL>` stands for a Zisp
+integer value between 0 and 254.
+
+These external representations of strings will be explained in more detail
+further below, including backslash escape sequences allowed within, and how
+exactly at-quoted strings work.
+
+Strings have a fixed length, counted in bytes.  Each byte can have any value,
+including zero (ASCII NUL).  The parser reads bytes, not Unicode characters; a
+string may contain UTF-8 byte sequences, but these are not tested for validity.
+
+A string that is up to 255 bytes long is automatically *interned*, meaning any
+occurrence of the same string -- equal in length and containing the same byte
+values -- ends up being represented by the same bit-pattern; either a memory
+address, or an immediate representation within a CPU word for short strings.
+The quotation method is inconsequential to this process; for example, while
+`|foo bar|` and `"foo bar"` will parse into different list values, the actual
+string they hold a reference to will be the same one in program memory.  This
+behavior is however configurable and can be disabled entirely for cases where
+large numbers of arbitrary binary strings are being parsed.
+
+Strings of length greater than 255 bytes are stored separately in memory, even
+if they are equal in length and content.
+
+### Rune
+
+A rune is represented by an ASCII character sequence of 1 to 6 bytes, that must
+begin with a letter, and may only contain letters and digits.  This character
+sequence of letters and digits is called the *name* of the rune.  A rune that
+follows this constraint is valid as a datum.
+
+Zisp code may explicitly construct values of the rune type that violate the
+above constraints.  Such runes are not valid data and cannot be printed or
+parsed.
+
+Runes are case-sensitive, and the parser always emits runes using upper-case
+letters when expressing syntax sugar.  Uppercase rune names are reserved for
+Zisp's internal use and standard library; users can use lowercase runes with
+custom meaning without worrying about clashes, with the exception of a small
+number of lowercase runes such as `#true` and `#false` that are part of the
+default decoder settings and documented explicitly as such.
+
+Runes are always stored directly in a CPU word; never by memory address.
+
+### List
+
+A list is a contiguous array of one or more values in memory, whose length may
+be encoded directly within the pointer to the head of the array, or else the
+array is terminated with a special sentinel bit-pattern that is not otherwise
+valid as a Zisp value.
+
+The parser allocates a unique array in program memory for every list, and the
+list as a value is then represented by the memory address of that array, with
+either an exact length tag or a tag indicating that it's sentinel-terminated.
+
+Lists are valid data if one of the following holds true:
+
+* The list encodes a quoted string, datum label, or shebang line.
+
+* All values in the list are a valid datum.
+
+Further, a structure of nested list values may not contain cyclic references
+back up in the structure (which would make the above definition diverge into
+infinity).  Such cycles must be broken up with datum labels, or else the list
+cannot be considered a datum, since it cannot be printed or parsed.
+
+### Nil
+
+The Zisp nil value is a singleton and a datum.  There is exactly one nil value,
+used in lieu of a list of zero length; it has the external representation `()`.
+
+
+## Quoted strings
+
+Three quoted string types exist: Pipe-quoted, double-quoted, and at-quoted.
+This section goes into the details of each variant.
+
+### Pipe-quoted
+
+Strings can be quoted with pipes, like symbols in R7RS Scheme, which triggers
+the parser to generate a list with the structure:
+
+    (#PQSTR <STRING>)   ;; <STRING> is visual aid, not syntax
+
+The decoder, using default settings, would emit this string verbatim as a value.
+Then, during code evaluation, this would be seen as an identifier.  In this way,
+pipe-quoted strings are equivalent to bare strings in functionality.
+
+It is important to understand that the decoder sits between the parser and the
+[evaluator](3-eval.html), and in opposition to Lisp and Scheme tradition, it is
+common for the evaluator to receive values that are not valid as a datum; here,
+a string unto itself that may not be a valid datum.  Yet, it is valid as an
+identifier for the purposes of the evaluator.
+
+### Double-quoted
+
+Strings wrapped in the double-quote symbol parse into:
+
+    (#DQSTR <STRING>)   ;; <STRING> is visual aid, not syntax
+
+Under default settings, the decoder would transform this into a value which,
+when evaluated as code, simply yields the contained string as a value.
+
+### At-quoted
+
+This is a special type of syntax for "raw" strings, meaning that no backslash
+escapes nor any other kind of escape sequence are recognized within them.
+
+The syntax begins with an at sign, followed by any byte.  That byte becomes a
+termination marker, and the string cannot contain an occurrence of it, since
+there are no escape sequences.  The byte value 255 has a special meaning; see
+further below.
+
+    @"foo \ bar"  ->  (#ATSTR <SENTINEL> <STRING>)
+
+The visual tokens `<SENTINEL>` and `<STRING>` represent an integer and string
+value, respectively.  Here, the integer would be 34, which is the ASCII value
+for a double-quote sign.  The string contains a literal backslash, since there
+is no backslash escape parsing.
+
+This style of quoting can be useful, for instance, when representing regular
+expressions as strings in code:
+
+    ;; Matches e.g. foo\bar.["blah"]
+
+    @/^foo\\(bar|baz)\.\[".*"\]$/
+
+Were it not for this syntax, this regular expression would only be possible to
+represent through a quoted string such as the following:
+
+    ;; Same as above, but so many backslashes
+
+    "^foo\\\\(bar|baz)\\t\\[\".*\"\\]$"
+
+The byte that follows the at sign need not be a printable character or even a
+valid ASCII byte; it can be absolutely any byte value, even NUL.  This can be
+useful to easily encode binary data which is known to not contain a specific
+byte; an example would be C strings which cannot contain NUL.
+
+If however the byte value is 255, then it does not stand for a sentinel, but
+rather indicates that 6 more bytes follow, interpreted as a big-endian 48-bit
+integer, which is the count of bytes making up the contents of the string.
+
+Example sequence of bytes, represented as a mixture of ASCII and raw integers:
+
+    '@' 255 0 0 0 0 2 100 <612 bytes>  ->  (#ATSTR <STRING>)
+
+One may ask why the length is not included in the list.  This is unnecessary,
+since strings in Zisp already carry length information in their own metadata
+structure.
+    
+
+### Backslash escapes
+
+In pipe-quoted and double-quoted strings, the following ASCII characters may
+follow a backslash to insert a certain character.
+
+    +-------+----------------------------+
+    | Char  | Meaning                    |
+    +-------+----------------------------+
+    | \     | Literal backslash          |
+    +-------+----------------------------+
+    | |     | Literal pipe symbol        |
+    +-------+----------------------------+
+    | "     | Literal double-quote       |
+    +-------+----------------------------+
+    | 0     | ASCII NUL                  |
+    +-------+----------------------------+
+    | a     | ASCII Alert                |
+    +-------+----------------------------+
+    | b     | ASCII Backspace            |
+    +-------+----------------------------+
+    | t     | ASCII Tab (Horizontal)     |
+    +-------+----------------------------+
+    | n     | ASCII Newline (Line Feed)  |
+    +-------+----------------------------+
+    | v     | ASCII Vertical Tab         |
+    +-------+----------------------------+
+    | f     | ASCII Form Feed            |
+    +-------+----------------------------+
+    | r     | ASCII Carriage Return      |
+    +-------+----------------------------+
+    | e     | ASCII Escape               |
+    +-------+----------------------------+
+
+In words:
+
+* A backslash, followed by a backslash, pipe, or double-quote character, is
+  substituted with a literal occurrence of that character.
+
+* The characters 0, a, b, t, n, v, f, r, and e have the same meanings as in the
+  C programming language, representing common ASCII control characters.
+
+Further, the following Regular Expression patterns following a backslash have
+special meaning.
+
+    +---------------------+-----------------------+
+    | Regular Expression  | Meaning               |
+    +---------------------+-----------------------+
+    | [\t ]*\n[\t ]*      | Discarded             |
+    +---------------------+-----------------------+
+    | x([0-9a-fA-F]{2})*; | Arbitrary bytes       |
+    +---------------------+-----------------------+
+    | u[0-9a-fA-F]+;      | Unicode Scalar Value  |
+    +---------------------+-----------------------+
+
+Explanations:
+
+* A backslash followed by any number of blanks (space or tab), a newline, and
+  again any number of blanks, is substituted with nothing.  This is to allow
+  splitting a string into multiple lines for human readability.
+
+      (define p "This paragraph has been visually split into multiple \
+                 lines, but the newline is escaped, so it's one line.")
+
+* An x, followed by pairs of hexadecimal digits (case insensitive), terminated
+  by a semicolon, is substituted with the sequence of bytes represented by the
+  corresponding pairs of hexadecimal digits.  E.g.: `"foo\xDEADBEEF;bar"`
+
+* A u, followed by a hexadecimal digit sequence (case insensitive), terminated
+  by a semicolon, is substituted with the canonical UTF-8 byte sequence for the
+  Unicode Scalar Value represented by that hexadecimal number.  The number must
+  be in the range `0` to `10FFFF`.  E.g.: `"foo\u00A0;bar"`
+
+### Newlines in strings
+
+Normally, a newline in a string has no special meaning and simply becomes part
+of the string.  However, newlines can be backslash-escaped, which simple erases
+them; the escaped newline can also be preceded or followed by any number of tab
+and space characters, which are all stripped as well.  (Note: It's not blanks
+preceding the backslash that are stripped, but blanks following the backslash
+and preceding the newline; i.e., blanks at the end of the line.)
+
+Following are some examples of how multi-line strings can appear in source code
+with different intentions and meanings:
+
+    (define paragraph "This paragraph has been visually split into multiple \
+                       lines, but the newlines are escaped, so it's one line.")
+
+    (define json-object '|   ;; use '|| so double-quotes need no escaping
+      {
+        "key": "value"
+      }
+    |)
+
+The second example is actually slightly problematic.  It begins with a newline,
+which may be undesirable, but escaping that newline would cause the first line
+to have no indentation, thus the opening `{` would not line up with the closing
+`}` when this string is printed out.  Further, if the entire block of code is
+indented, then the string contents may be more indented than intended.  (No pun
+or rhyme intended.)  Consider:
+
+    (let ((foo one))
+      (let ((bar two))
+        (let ((json-object '|
+                 {
+                   "key": "value"
+                 }
+               |))
+          (do-whatever))))
+
+The string bound to `json-object` has redundant indentation.  Should the parser
+attempt to solve this issue?
+
+Thankfully, we have the decoder to handle such complexities.  Under the default
+settings, the rune `#HASH` is bound to a decoder rule which detects a payload
+value that is a string literal, and implements the same algorithm as seen in
+Java 15 Text Blocks: [JEP 378: Text Blocks](https://openjdk.org/jeps/378)
+
+Thus, we can do the following:
+
+    (let ((foo one))
+      (let ((bar two))
+        (let ((json-object #|
+    ...........  {
+    ...........    "key": "value"
+    ...........  }
+    ...........|))
+          (do-whatever))))
+
+(Dots represent whitespace that is deleted.  The initial newline is, as well.)
+
+The only feature Zisp does not offer is a way to fence off multi-line strings
+with a longer token such as `"""` as seen in Python and Java, or an arbitrary
+word as seen in Bourne shell and PHP "here doc" syntax.
+
+However, if a programmer truly wanted to have arbitrary text blocks in code,
+without needing to escape anything in them, it's possible to abuse at-quoted
+string syntax, using it with an ASCII control character which is displayed
+visibly by a text editor.  In the following, the characters `^\` are meant to
+represent a literal ASCII File Separator character in the source code:
+
+    (define json-object #@^\
+      {
+        "key": "value"
+      }
+      ^\)
+
+It works fine in Emacs, so why not?  Use `C-q C-\` to insert the `^\`.
+
+This is indeed quite an eldritch syntax, but hopefully most programs would not
+need to use it.
+
+
+## Other syntax
+
+The following table summarizes commonly useful syntax abbreviations:
+
+    [...]   -> (#SQUARE ...)        #datum       -> (#HASH datum)
+
+    {...}   -> (#BRACE ...)         #rune(...)   -> (#rune ...)
+
+    'datum  -> (#QUOTE datum)       dat1dat2     -> (#JOIN dat1 dat2)
+
+    `datum  -> (#GRAVE datum)       dat1.dat2    -> (#DOT dat1 dat2)
+
+    ,datum  -> (#COMMA datum)       dat1:dat2    -> (#COLON dat1 dat2)
+
+Notes:
+
+* The terms datum, dat1, and dat2 each refer to an arbitrary datum; ellipsis
+  means zero or more data.
+
+* The `#datum` form only applies when the datum following the hash sign is
+  anything other than a bare string, since otherwise this would be ambiguous
+  with a rune literal.  A bare string can nevertheless follow the hash sign by
+  separating the two with a backslash:
+
+      #\string  ->  (#HASH string)
+
+* Though not represented in the table due to notational difficulty, the form
+  `#rune(...)` doesn't require a list in the second position; any datum that
+  works with the `#datum` syntax also works with `#rune<DATUM>`.
+
+      #rune1#rune2  -> (#rune1 #rune2)
+
+      #rune\string  -> (#rune string)
+
+      #rune'string  -> (#rune (#QUOTE string))
+
+      #rune"string" -> (#rune (#DQSTR |string|))
+
+  As a counter-example, following a rune immediately with a bare string isn't
+  possible without the delimiting backslash, since that would be ambiguous:
+
+      #abcdefgh  ;Could be (#abcdef gh) or (#abcde fgh) or ...
+
+* Syntax sugar can combine arbitrarily.  Some examples follow.  Any of these may
+  or may not actually have a meaning in code; some might simply end up producing
+  an error during decoding, or later evaluation of code.
+
+      #{...}            -> (#HASH (#BRACE ...))
+
+      #'foo             -> (#HASH (#QUOTE foo))
+
+      ##'[...]          -> (#HASH (#HASH (#QUOTE (#SQUARE ...))))
+
+      {x y}[i j]        -> (#JOIN (#BRACE x y) (#SQUARE i j))
+
+      foo.bar.baz{x y}  -> (#JOIN (#DOT (#DOT foo bar) baz) (#BRACE x y))
+
+* Those used to thinking in Lisp and Scheme may think that `(#QUOTE ...)` halts
+  further decoding of enclosed data.  This is not so, since quoting is related
+  to code evaluation, not decoding.
+
+### Datum labels
+
+Valid data cannot be cyclic, since that would mean it has infinite length in
+bytes.  To externally represent a value with cyclic structure, one uses datum
+labels in the data encoding of the value.
+
+A datum label either wraps another datum to assign a number to it, or contains
+just a reference to a previous assignment.
+
+    +------------------+----------------------------+
+    | Syntax           | Internal datum structure   |
+    +------------------+----------------------------+
+    | #%<HEX>=<DATUM>  | (#LABEL <NUMBER> <DATUM>)  |
+    +------------------+----------------------------+
+    | #%<HEX>%         | (#LABEL <NUMBER>)          |
+    +------------------+----------------------------+
+
+In this visual, the token `<HEX>` stands for a hexadecimal digit sequence, the
+token `<DATUM>` stands for any other datum, and `<NUMBER>` is a stand-in for a
+number value; that which is represented by `<HEX>`.
+
+For clarity, concrete examples follow:
+
+    +-------------------+------------------------------+
+    | Byte sequence     | Parse result                 |
+    +-------------------+------------------------------+
+    | #%1234abcd=(foo)  | (#LABEL <0x1234abcd> (foo))  |
+    +-------------------+------------------------------+
+    | #%1234abcd%       | (#LABEL <0x1234abcd>)        |
+    +-------------------+------------------------------+
+
+Here, the visual token `<0x1234abcd>` stands for a Zisp value of a numeric type
+with an integer value.  Note that the decoder may not accept a bare string here,
+meaning this syntax sugar is not merely an abbreviation.
+
+### Shebang
+
+Finally, the parser recognizes the Unix *shebang* syntax and outputs a datum to
+hold the string values found within:
+
+    #!interpreter          ->  (#SHBANG interpreter)
+
+    #!interpreter argline  ->  (#SHBANG interpreter argline)
+
+When executing a script file, Zisp simply stores this into a global value that
+may be inspected if desired.
+
+
+<!--
+;; Local Variables:
+;; fill-column: 80
+;; End:
+-->