koreader/frontend/luxl.lua

--[[
    This code was derived from the pico_xml project
    which can be found here:

    http://kd7yhr.org/bushbo/pico_xml.md

    The original C code, written by: Brian O. Bush

    Pure Lua Version written by: William A Adams
    Dramatic Speed Improvements by: Robert G Jakabosky
    https://github.com/Wiladams/LAPHLibs/blob/master/laphlibs/luxl.lua

    References

    http://www.faqs.org/rfcs/rfc3076.html
    http://www.w3.org/TR/REC-xml/

]]

local ffi = require "ffi"
local bit = require "bit"
local band = bit.band


--[[
 Types of characters;
     0 is not valid
     1 is letters,
     2 are digits (including '.')
     3 whitespace
--]]

local char_type = ffi.new("const int[256]", {
    0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 0, 0, 3, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 0,
    2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0,
    0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0,
    0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
});



-- Internal states that the parser can be in at any given time.
local ST_START = 0;            -- starting base state; default state
local ST_TEXT =1;              -- text state
local ST_START_TAG = 2;        -- start tag state
local ST_START_TAGNAME =3;     -- start tagname state
local ST_START_TAGNAME_END =4; -- start tagname ending state
local ST_END_TAG =5;           -- end tag state
local ST_END_TAGNAME=6;        -- end tag tagname state
local ST_END_TAGNAME_END=7;    -- end tag tagname ending
local ST_EMPTY_TAG=8;          -- empty tag state
local ST_SPACE=9;              -- linear whitespace state
local ST_ATTR_NAME=10;         -- attribute name state
local ST_ATTR_NAME_END=11;     -- attribute name ending state
local ST_ATTR_VAL=12;          -- attribute value starting state
local ST_ATTR_VAL2=13;         -- attribute value state
local ST_ERROR=14;             -- error state

-- character classes that we will match against; This could be expanded if
--   need be, however, we are aiming for simple.
local CCLASS_NONE = 0;         -- matches nothing; a base state
local CCLASS_LEFT_ANGLE=1;     -- matches start tag '<'
local CCLASS_SLASH=2;          -- matches forward slash
local CCLASS_RIGHT_ANGLE=3;    -- matches end tag '>'
local CCLASS_EQUALS=4;         -- matches equals sign
local CCLASS_QUOTE=5;          -- matches double-quotes
local CCLASS_LETTERS=6;        -- matches a-zA-Z letters and digits 0-9
local CCLASS_SPACE=7;          -- matches whitespace
local CCLASS_ANY=8;            -- matches any ASCII character; will match all above classes

-- Types of events: start element, end element, text, attr name, attr
-- val and start/end document. Other events can be ignored!
local EVENT_START = 0;        -- Start tag
local EVENT_END = 1;          -- End tag
local EVENT_TEXT = 2;         -- Text
local EVENT_ATTR_NAME = 3;    -- Attribute name
local EVENT_ATTR_VAL = 4;     -- Attribute value
local EVENT_END_DOC = 5;      -- End of document
local EVENT_MARK = 6;         -- Internal only; notes position in buffer
local EVENT_NONE = 7;         -- Internal only; should never see this event

local entity_refs =  {
    ["&lt;"] = '<',
    ["&gt;"] = '>',
    ["&amp;"] = '&',
    ["&apos;"] = '\'',
    ["&quot;"] = '"',
}

-- Map constant values to constant names.
local STATE_NAMES = {
    [ST_START] = "ST_START",
    [ST_TEXT] = "ST_TEXT",
    [ST_START_TAG] = "ST_START_TAG",
    [ST_START_TAGNAME] = "ST_START_TAGNAME",
    [ST_START_TAGNAME_END] = "ST_START_TAGNAME_END",
    [ST_END_TAG] = "ST_END_TAG",
    [ST_END_TAGNAME] = "ST_END_TAGNAME",
    [ST_END_TAGNAME_END] = "ST_END_TAGNAME_END",
    [ST_EMPTY_TAG] = "ST_EMPTY_TAG",
    [ST_SPACE] = "ST_SPACE",
    [ST_ATTR_NAME] = "ST_ATTR_NAME",
    [ST_ATTR_NAME_END] = "ST_ATTR_NAME_END",
    [ST_ATTR_VAL] = "ST_ATTR_VAL",
    [ST_ATTR_VAL2] = "ST_ATTR_VAL2",
    [ST_ERROR] = "ST_ERROR",
}

--[[
 State transition table element; contains:
 (1) current state,
 (2) clazz that must match,
 (3) next state if we match, and
 (4) event that is emitted upon match.
--]]

-- Note: States must be grouped in match order AND grouped together!
local LEXER_STATES = {
  -- [0-2] starting state, which also serves as the default state in case of error
  { state = ST_START,         cclass = CCLASS_SPACE,        next_state = ST_SPACE,             event = EVENT_NONE },
  { state = ST_START,         cclass = CCLASS_LEFT_ANGLE,   next_state = ST_START_TAG,         event = EVENT_NONE },
  { state = ST_START,         cclass = CCLASS_ANY,          next_state = ST_TEXT,              event = EVENT_MARK },

  -- [3-5] space state handles linear white space
  { state = ST_SPACE,         cclass = CCLASS_SPACE,        next_state = ST_SPACE,             event = EVENT_NONE },
  { state = ST_SPACE,         cclass = CCLASS_LEFT_ANGLE,   next_state = ST_START_TAG,         event = EVENT_TEXT },
  { state = ST_SPACE,         cclass = CCLASS_ANY,          next_state = ST_TEXT,              event = EVENT_MARK },

  -- [6-8] handle start tag
  { state = ST_START_TAG,     cclass = CCLASS_LETTERS,      next_state = ST_START_TAGNAME,     event = EVENT_MARK },
  { state = ST_START_TAG,     cclass = CCLASS_SLASH,        next_state = ST_END_TAG,           event = EVENT_MARK },
  { state = ST_START_TAG,     cclass = CCLASS_SPACE,        next_state = ST_START_TAG,         event = EVENT_NONE },         -- < tag >

  -- [9-12] handle start tag name
  { state = ST_START_TAGNAME, cclass = CCLASS_LETTERS,      next_state = ST_START_TAGNAME,     event = EVENT_NONE },
  { state = ST_START_TAGNAME, cclass = CCLASS_SPACE,        next_state = ST_START_TAGNAME_END, event = EVENT_START },
  { state = ST_START_TAGNAME, cclass = CCLASS_SLASH,        next_state = ST_EMPTY_TAG,         event = EVENT_END },
  { state = ST_START_TAGNAME, cclass = CCLASS_RIGHT_ANGLE,  next_state = ST_START,             event = EVENT_START },

  -- [13-16] handle start tag name end
  { state = ST_START_TAGNAME_END,  cclass = CCLASS_LETTERS, next_state = ST_ATTR_NAME,         event = EVENT_MARK },
  { state = ST_START_TAGNAME_END,  cclass = CCLASS_SPACE,   next_state = ST_START_TAGNAME_END, event = EVENT_NONE },
  { state = ST_START_TAGNAME_END,  cclass = CCLASS_RIGHT_ANGLE, next_state = ST_START,         event = EVENT_START },
  { state = ST_START_TAGNAME_END,  cclass = CCLASS_SLASH,   next_state = ST_EMPTY_TAG,         event = EVENT_MARK },         -- Empty tag <br />

  -- [17] handle empty tags, e.g., <br />
  { state = ST_EMPTY_TAG,     cclass = CCLASS_RIGHT_ANGLE,  next_state = ST_START,             event = EVENT_END },          -- Empty tag <br />

  -- [18] handle end tag, e.g., <tag />
  { state = ST_END_TAG,       cclass = CCLASS_LETTERS,      next_state = ST_END_TAGNAME,       event = EVENT_NONE },

  -- [19-21] handle end tag name
  { state = ST_END_TAGNAME,   cclass = CCLASS_LETTERS,      next_state = ST_END_TAGNAME,       event = EVENT_NONE },
  { state = ST_END_TAGNAME,   cclass = CCLASS_RIGHT_ANGLE,  next_state = ST_START,             event = EVENT_END },
  { state = ST_END_TAGNAME,   cclass = CCLASS_SPACE,        next_state = ST_END_TAGNAME_END,   event = EVENT_END },          -- space after end tag name </br >

  -- [22-23] handle ending of end tag name
  { state = ST_END_TAGNAME_END, cclass = CCLASS_SPACE,      next_state = ST_END_TAGNAME_END,   event = EVENT_NONE },
  { state = ST_END_TAGNAME_END, cclass = CCLASS_RIGHT_ANGLE,next_state = ST_START,             event = EVENT_NONE },

  -- [24-26] handle text
  { state = ST_TEXT,          cclass = CCLASS_SPACE,        next_state = ST_SPACE,             event = EVENT_NONE },
  { state = ST_TEXT,          cclass = CCLASS_LEFT_ANGLE,   next_state = ST_START_TAG,         event = EVENT_TEXT },
  { state = ST_TEXT,          cclass = CCLASS_ANY,          next_state = ST_TEXT,              event = EVENT_NONE },

  -- [27-29] handle attribute names
  { state = ST_ATTR_NAME,     cclass = CCLASS_LETTERS,      next_state = ST_ATTR_NAME,         event = EVENT_MARK },
  { state = ST_ATTR_NAME,     cclass = CCLASS_SPACE,        next_state = ST_ATTR_NAME_END,     event = EVENT_ATTR_NAME },    -- space before '=' sign
  { state = ST_ATTR_NAME,     cclass = CCLASS_EQUALS,       next_state = ST_ATTR_VAL,          event = EVENT_ATTR_NAME },    -- <tag attr ="2">

  -- [30-32] attribute name end
  { state = ST_ATTR_NAME_END, cclass = CCLASS_SPACE,        next_state = ST_ATTR_NAME_END,     event = EVENT_NONE },
  { state = ST_ATTR_NAME_END, cclass = CCLASS_LETTERS,      next_state = ST_ATTR_NAME,         event = EVENT_MARK },
  { state = ST_ATTR_NAME_END, cclass = CCLASS_EQUALS,       next_state = ST_ATTR_VAL,          event = EVENT_NONE },

  -- [33-34] handle attribute values, initial quote and spaces
  { state = ST_ATTR_VAL,      cclass = CCLASS_QUOTE,        next_state = ST_ATTR_VAL2,         event = EVENT_NONE },
  { state = ST_ATTR_VAL,      cclass = CCLASS_SPACE,        next_state = ST_ATTR_VAL,          event = EVENT_NONE },        -- initial spaces before quoted attribute value

  -- [35-37] handle actual attribute values
  { state = ST_ATTR_VAL2,     cclass = CCLASS_QUOTE,        next_state = ST_START_TAGNAME_END, event = EVENT_ATTR_VAL },
  { state = ST_ATTR_VAL2,     cclass = CCLASS_LETTERS,      next_state = ST_ATTR_VAL2,         event = EVENT_MARK },
  { state = ST_ATTR_VAL2,     cclass = CCLASS_SLASH,        next_state = ST_ATTR_VAL2,         event = EVENT_MARK },

  -- [38] End of table marker
  { state = ST_ERROR,         cclass = CCLASS_NONE,         next_state = ST_ERROR,             event = EVENT_NONE }
};

ffi.cdef[[
struct parse_state {
    const char* buf;      /* reference to buffer */
    int bufsz;            /* size of buf */
    int mark;
    int i;
    int ix;               /* index into buffer */
};

]]

local cclass_match = {
    [CCLASS_LETTERS] = "(ctype == 1 or ctype == 2)",
    [CCLASS_LEFT_ANGLE] = "(c == T_LT)",
    [CCLASS_SLASH] = "(c == T_SLASH)",
    [CCLASS_RIGHT_ANGLE] = "(c == T_GT)",
    [CCLASS_EQUALS] = "(c == T_EQ)",
    [CCLASS_QUOTE] = "(c == T_QUOTE)",
    [CCLASS_SPACE] = "(ctype == 3)",
    [CCLASS_ANY] = "true",
}

local STATES = {}
local STATE_FUNCS = {}

local function next_char(ps, state, verbose)
    local i = ps.ix
    if i >= ps.bufsz then
        return EVENT_END_DOC, 0, i, state
    end
    ps.i = i
    local c = band(ps.buf[i], 0xff);
    ps.ix = i + 1
    if verbose then
        verbose(i, STATE_FUNCS[state], c)
    end

    -- run state function to find next state.
    return state(ps, c, verbose)
end


local fsm_code = ''
local function code(...)
    for i=1,select("#", ...) do
        fsm_code = fsm_code .. tostring(select(i, ...))
    end
end
code[[
local STATES, next_char, char_type = ...

local T_LT = string.byte('<')
local T_SLASH = string.byte('/')
local T_GT = string.byte('>')
local T_EQ = string.byte('=')
local T_QUOTE = string.byte('"')

]]
-- pre-define locals for state functions.
for i=0,#STATE_NAMES do
    local name = STATE_NAMES[i]
    code('local ', name, '_f\n')
end
-- group LEXER states.
for i=1,#LEXER_STATES do
    local p_state = LEXER_STATES[i]
    local state = STATES[p_state.state]
    local cclasses
    if not state then
        cclasses = {}
        state = { state = p_state.state, cclasses = cclasses }
        STATES[p_state.state] = state
    else
        cclasses = state.cclasses
    end
    cclasses[#cclasses + 1] = p_state
end
local function gen_cclass_code(prefix, cclass)
    local next_state = STATE_NAMES[cclass.next_state]
    if cclass.event == EVENT_MARK then
        code(prefix, "if(ps.mark == 0) then ps.mark = ps.i end -- mark the position\n")
    elseif cclass.event ~= EVENT_NONE then
        code(prefix, "if(ps.mark > 0) then\n")
        code(prefix,'  return ', cclass.event,', ',next_state,'_f\n')
        code(prefix, "end\n")
    end
    code(prefix,'return next_char(ps, ', next_state,'_f, verbose)\n')
end
-- generate state functions.
for i=0,#STATE_NAMES do
    local name = STATE_NAMES[i]
    local state = STATES[i]
    local cclasses = state.cclasses
    code('function ', name, '_f(ps, c, verbose)\n')
    code('  local ctype = char_type[c]\n')
    local has_any
    for i=1,#cclasses do
        local cclass = cclasses[i]
        local id = cclass.cclass
        local condition = cclass_match[id]
        if id == CCLASS_ANY then
            has_any = cclass
        elseif i == 1 then
            code('  if ', condition, ' then\n')
            gen_cclass_code('    ', cclass)
        else
            code('  elseif ', condition, ' then\n')
            gen_cclass_code('    ', cclass)
        end
    end
    code('  end\n')
    -- catch-all for cclass_any or goto error state.
    if has_any then
        gen_cclass_code('  ', has_any)
    else
        -- exit from FSM on error.
        code('  return nil, ',name,'_f, c\n')
    end
    code('end\n')
    -- map state id to state function
    code('STATES[', i, '] = ', name, '_f\n')
    -- reverse map state function to state id
    code('STATES[', name, '_f] = ', i, '\n')
end
-- Compile FSM code
local state_funcs = assert(loadstring(fsm_code, "luxl FSM code"))
state_funcs(STATE_FUNCS, next_char, char_type)
fsm_code = nil


local luxl = {
    EVENT_START = EVENT_START;         -- Start tag
    EVENT_END = EVENT_END;             -- End tag
    EVENT_TEXT = EVENT_TEXT;           -- Text
    EVENT_ATTR_NAME = EVENT_ATTR_NAME; -- Attribute name
    EVENT_ATTR_VAL = EVENT_ATTR_VAL;   -- Attribute value
    EVENT_END_DOC = EVENT_END_DOC;     -- End of document
    EVENT_MARK = EVENT_MARK;           -- Internal only; notes position in buffer
    EVENT_NONE = EVENT_NONE;           -- Internal only; should never see this event
}
local luxl_mt = { __index = luxl }

function luxl.new(buffer, bufflen)
    local newone = {
        buf = ffi.cast("const uint8_t *", buffer); -- pointer to "uint8_t *" buffer (0 based)
        bufsz = bufflen;       -- size of input buffer
        state = ST_START;      -- current state
        event = EVENT_NONE;    -- current event
        err = 0;               -- number of errors thus far
        markix = 0;            -- offset of current item of interest
        marksz = 0;            -- size of current item of interest
        MsgHandler = nil;      -- Routine to handle messages
        ErrHandler = nil;      -- Routine to call when there's an error
        EventHandler = nil;
        ps = ffi.new('struct parse_state', {
            buf = buffer,
            bufsz = bufflen,
            mark = 0,
            i = 0,
            ix = 0,
        }),
    }
    setmetatable(newone, luxl_mt);

    return newone;
end

function luxl:Reset(buffer, bufflen)
    self.buf = buffer          -- pointer to "uint8_t *" buffer (0 based)
    self.bufsz = bufflen       -- size of input buffer
    self.state = ST_START      -- current state
    self.event = EVENT_NONE    -- current event
    self.err = 0               -- number of errors thus far
    self.markix = 0            -- offset of current item of interest
    self.marksz = 0            -- size of current item of interest
    local ps = self.ps
    ps.buf = buffer
    ps.bufsz = bufflen
    ps.mark = 0
    ps.i = 0
    ps.ix = 0
end

function luxl:SetMessageHandler(handler)
    self.MsgHandler = handler;
end


--[[
    GetNext is responsible for moving through the stream
    of characters.  At the moment, it's fairly naive in
    terms of character encodings.

    In a more robust implementation, luxl will read from a
    stream, which knows about the specific encoding, and
    will hand out code points based on that particular encoding.

    So, only straight ASCII for the moment.

    Returns event type, starting offset, size
--]]

function luxl:GetNext()
    local event, state_f, c
    local ps = self.ps
    ps.mark = 0
    state_f = STATE_FUNCS[self.state]
    repeat
        event, state_f, c = next_char(ps, state_f, self.MsgHandler)
        -- update state id.
        self.state = STATE_FUNCS[state_f]
        if not event then
            -- handle error
            -- default to start state
            self.err = self.err + 1;
            if self.ErrHandler then
                self.ErrHandler(ps.i, self.state, c);
            end
        end
    until event
    -- basically we are guaranteed never to have an event of
    --   type EVENT_MARK or EVENT_NONE here.
    self.event = event
    local markix = ps.mark
    local marksz = ps.i-ps.mark
    self.markix = markix
    self.marksz = marksz
    if(self.EventHandler) then
        self.EventHandler(event, markix, marksz)
    end
    return event, markix, marksz
end

function luxl:Lexemes()
    return function()
        local event, offset, size = self:GetNext();
        if(event == EVENT_END_DOC) then
            return nil;
        else
            return event, offset, size;
        end
    end
end

return luxl