Archived
1
0
This repo is archived. You can view files and clone it, but cannot push or open issues or pull requests.
kll/kll.py

588 lines
19 KiB
Python
Raw Normal View History

#!/usr/bin/env python3
# KLL Compiler
# Keyboard Layout Langauge
#
# Copyright (C) 2014 by Jacob Alexander
#
# This file is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This file is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this file. If not, see <http://www.gnu.org/licenses/>.
### Imports ###
import argparse
import io
import os
import re
import sys
import token
import importlib
from tokenize import generate_tokens
from re import VERBOSE
from pprint import pformat
from kll_lib.hid_dict import *
from kll_lib.containers import *
from funcparserlib.lexer import make_tokenizer, Token, LexerError
from funcparserlib.parser import (some, a, many, oneplus, skip, finished, maybe, skip, forward_decl, NoParseError)
### Decorators ###
## Print Decorator Variables
ERROR = '\033[5;1;31mERROR\033[0m:'
## Python Text Formatting Fixer...
## Because the creators of Python are averse to proper capitalization.
textFormatter_lookup = {
"usage: " : "Usage: ",
"optional arguments" : "Optional Arguments",
}
def textFormatter_gettext( s ):
return textFormatter_lookup.get( s, s )
argparse._ = textFormatter_gettext
### Argument Parsing ###
def checkFileExists( filename ):
if not os.path.isfile( filename ):
print ( "{0} {1} does not exist...".format( ERROR, filename ) )
sys.exit( 1 )
def processCommandLineArgs():
# Setup argument processor
pArgs = argparse.ArgumentParser(
usage="%(prog)s [options] <file1>...",
description="Generates .h file state tables and pointer indices from KLL .kll files.",
epilog="Example: {0} mykeyboard.kll -d colemak.kll -p hhkbpro2.kll -p symbols.kll".format( os.path.basename( sys.argv[0] ) ),
formatter_class=argparse.RawTextHelpFormatter,
add_help=False,
)
# Positional Arguments
pArgs.add_argument( 'files', type=str, nargs='+',
help=argparse.SUPPRESS ) # Suppressed help output, because Python output is verbosely ugly
# Optional Arguments
pArgs.add_argument( '-b', '--backend', type=str, default="kiibohd",
help="Specify target backend for the KLL compiler.\n"
"Default: kiibohd" )
pArgs.add_argument( '-d', '--default', type=str, nargs='+',
help="Specify .kll files to layer on top of the default map to create a combined map." )
pArgs.add_argument( '-p', '--partial', type=str, nargs='+', action='append',
help="Specify .kll files to generate partial map, multiple files per flag.\n"
"Each -p defines another partial map.\n"
"Base .kll files (that define the scan code maps) must be defined for each partial map." )
pArgs.add_argument( '-t', '--template', type=str, default="templates/kiibohdKeymap.h",
help="Specify template used to generate the keymap.\n"
"Default: templates/kiibohdKeymap.h" )
pArgs.add_argument( '-o', '--output', type=str, default="templateKeymap.h",
help="Specify output file. Writes to current working directory by default.\n"
"Default: generatedKeymap.h" )
pArgs.add_argument( '-h', '--help', action="help",
help="This message." )
# Process Arguments
args = pArgs.parse_args()
# Parameters
baseFiles = args.files
defaultFiles = args.default
partialFileSets = args.partial
if defaultFiles is None:
defaultFiles = []
if partialFileSets is None:
partialFileSets = [[]]
# Check file existance
for filename in baseFiles:
checkFileExists( filename )
for filename in defaultFiles:
checkFileExists( filename )
for partial in partialFileSets:
for filename in partial:
checkFileExists( filename )
return (baseFiles, defaultFiles, partialFileSets, args.backend, args.template, args.output)
### Tokenizer ###
def tokenize( string ):
"""str -> Sequence(Token)"""
# Basic Tokens Spec
specs = [
( 'Comment', ( r' *#.*', ) ),
( 'Space', ( r'[ \t\r\n]+', ) ),
( 'USBCode', ( r'U(("[^"]+")|(0x[0-9a-fA-F]+)|([0-9]+))', ) ),
( 'USBCodeStart', ( r'U\[', ) ),
( 'ScanCode', ( r'S((0x[0-9a-fA-F]+)|([0-9]+))', ) ),
( 'ScanCodeStart', ( r'S\[', ) ),
( 'CodeEnd', ( r'\]', ) ),
( 'String', ( r'"[^"]*"', VERBOSE ) ),
( 'SequenceString', ( r"'[^']*'", ) ),
( 'Operator', ( r'=>|:\+|:-|:|=', ) ),
( 'Comma', ( r',', ) ),
( 'Dash', ( r'-', ) ),
( 'Plus', ( r'\+', ) ),
( 'Parenthesis', ( r'\(|\)', ) ),
( 'Number', ( r'-?(0x[0-9a-fA-F]+)|(0|([1-9][0-9]*))', VERBOSE ) ),
( 'Name', ( r'[A-Za-z_][A-Za-z_0-9]*', ) ),
( 'VariableContents', ( r'''[^"' ;:=>()]+''', ) ),
( 'EndOfLine', ( r';', ) ),
]
# Tokens to filter out of the token stream
useless = ['Space', 'Comment']
tokens = make_tokenizer( specs )
return [x for x in tokens( string ) if x.type not in useless]
### Parsing ###
## Map Arrays
macros_map = Macros()
variable_dict = dict()
capabilities_dict = Capabilities()
## Parsing Functions
def make_scanCode( token ):
scanCode = int( token[1:], 0 )
# Check size, to make sure it's valid
if scanCode > 0xFF:
print ( "{0} ScanCode value {1} is larger than 255".format( ERROR, scanCode ) )
raise
return scanCode
def make_usbCode( token ):
# If first character is a U, strip
if token[0] == "U":
token = token[1:]
# If using string representation of USB Code, do lookup, case-insensitive
if '"' in token:
try:
usbCode = kll_hid_lookup_dictionary[ token[1:-1].upper() ]
except LookupError as err:
print ( "{0} {1} is an invalid USB Code Lookup...".format( ERROR, err ) )
raise
else:
usbCode = int( token, 0 )
# Check size, to make sure it's valid
if usbCode > 0xFF:
print ( "{0} USBCode value {1} is larger than 255".format( ERROR, usbCode ) )
raise
return usbCode
def make_seqString( token ):
# Shifted Characters, and amount to move by to get non-shifted version
# US ANSI
shiftCharacters = (
( "ABCDEFGHIJKLMNOPQRSTUVWXYZ", 0x20 ),
( "+", 0x12 ),
( "&(", 0x11 ),
( "!#$%<>", 0x10 ),
( "*", 0x0E ),
( ")", 0x07 ),
( '"', 0x05 ),
( ":", 0x01 ),
( "^", -0x10 ),
( "_", -0x18 ),
( "{}|", -0x1E ),
( "~", -0x20 ),
( "@", -0x32 ),
( "?", -0x38 ),
)
listOfLists = []
shiftKey = kll_hid_lookup_dictionary["SHIFT"]
# Creates a list of USB codes from the string: sequence (list) of combos (lists)
for char in token[1:-1]:
processedChar = char
# Whether or not to create a combo for this sequence with a shift
shiftCombo = False
# Depending on the ASCII character, convert to single character or Shift + character
for pair in shiftCharacters:
if char in pair[0]:
shiftCombo = True
processedChar = chr( ord( char ) + pair[1] )
break
# Do KLL HID Lookup on non-shifted character
# NOTE: Case-insensitive, which is why the shift must be pre-computed
usbCode = kll_hid_lookup_dictionary[ processedChar.upper() ]
# Create Combo for this character, add shift key if shifted
charCombo = []
if shiftCombo:
charCombo = [ [ shiftKey ] ]
charCombo.append( [ usbCode ] )
# Add to list of lists
listOfLists.append( charCombo )
return listOfLists
def make_string( token ):
return token[1:-1]
def make_number( token ):
return int( token, 0 )
# Range can go from high to low or low to high
def make_scanCode_range( rangeVals ):
start = rangeVals[0]
end = rangeVals[1]
# Swap start, end if start is greater than end
if start > end:
start, end = end, start
# Iterate from start to end, and generate the range
return list( range( start, end + 1 ) )
# Range can go from high to low or low to high
# Warn on 0-9 (as this does not do what one would expect) TODO
# Lookup USB HID tags and convert to a number
def make_usbCode_range( rangeVals ):
# Check if already integers
if isinstance( rangeVals[0], int ):
start = rangeVals[0]
else:
start = make_usbCode( rangeVals[0] )
if isinstance( rangeVals[1], int ):
end = rangeVals[1]
else:
end = make_usbCode( rangeVals[1] )
# Swap start, end if start is greater than end
if start > end:
start, end = end, start
# Iterate from start to end, and generate the range
return list( range( start, end + 1 ) )
pass
## Base Rules
const = lambda x: lambda _: x
unarg = lambda f: lambda x: f(*x)
flatten = lambda list: sum( list, [] )
tokenValue = lambda x: x.value
tokenType = lambda t: some( lambda x: x.type == t ) >> tokenValue
operator = lambda s: a( Token( 'Operator', s ) ) >> tokenValue
parenthesis = lambda s: a( Token( 'Parenthesis', s ) ) >> tokenValue
eol = a( Token( 'EndOfLine', ';' ) )
def listElem( item ):
return [ item ]
def listToTuple( items ):
return tuple( items )
# Flatten only the top layer (list of lists of ...)
def oneLayerFlatten( items ):
mainList = []
for sublist in items:
for item in sublist:
mainList.append( item )
return mainList
# Capability arguments may need to be expanded (e.g. 1 16 bit argument needs to be 2 8 bit arguments for the state machine)
def capArgExpander( items ):
newArgs = []
# For each defined argument in the capability definition
for arg in range( 0, len( capabilities_dict[ items[0] ][1] ) ):
argLen = capabilities_dict[ items[0] ][1][ arg ][1]
num = items[1][ arg ]
byteForm = num.to_bytes( argLen, byteorder='little' ) # XXX Yes, little endian from how the uC structs work
# For each sub-argument, split into byte-sized chunks
for byte in range( 0, argLen ):
newArgs.append( byteForm[ byte ] )
return tuple( [ items[0], tuple( newArgs ) ] )
# Expand ranges of values in the 3rd dimension of the list, to a list of 2nd lists
# i.e. [ sequence, [ combo, [ range ] ] ] --> [ [ sequence, [ combo ] ], <option 2>, <option 3> ]
def optionExpansion( sequences ):
expandedSequences = []
# Total number of combinations of the sequence of combos that needs to be generated
totalCombinations = 1
# List of leaf lists, with number of leaves
maxLeafList = []
# Traverse to the leaf nodes, and count the items in each leaf list
for sequence in sequences:
for combo in sequence:
rangeLen = len( combo )
totalCombinations *= rangeLen
maxLeafList.append( rangeLen )
# Counter list to keep track of which combination is being generated
curLeafList = [0] * len( maxLeafList )
# Generate a list of permuations of the sequence of combos
for count in range( 0, totalCombinations ):
expandedSequences.append( [] ) # Prepare list for adding the new combination
position = 0
# Traverse sequence of combos to generate permuation
for sequence in sequences:
expandedSequences[ -1 ].append( [] )
for combo in sequence:
expandedSequences[ -1 ][ -1 ].append( combo[ curLeafList[ position ] ] )
position += 1
# Increment combination tracker
for leaf in range( 0, len( curLeafList ) ):
curLeafList[ leaf ] += 1
# Reset this position, increment next position (if it exists), then stop
if curLeafList[ leaf ] >= maxLeafList[ leaf ]:
curLeafList[ leaf ] = 0
if leaf + 1 < len( curLeafList ):
curLeafList[ leaf + 1 ] += 1
break
return expandedSequences
# Converts USB Codes into Capabilities
def usbCodeToCapability( items ):
# Items already converted to variants using optionExpansion
for variant in range( 0, len( items ) ):
# Sequence of Combos
for sequence in range( 0, len( items[ variant ] ) ):
for combo in range( 0, len( items[ variant ][ sequence ] ) ):
# Only convert if an integer, otherwise USB Code doesn't need converting
if isinstance( items[ variant ][ sequence ][ combo ], int ):
# Use backend capability name and a single argument
items[ variant ][ sequence ][ combo ] = tuple( [ backend.usbCodeCapability(), tuple( [ items[ variant ][ sequence ][ combo ] ] ) ] )
return items
## Evaluation Rules
def eval_scanCode( triggers, operator, results ):
# Convert to lists of lists of lists to tuples of tuples of tuples
# Tuples are non-mutable, and can be used has index items
triggers = tuple( tuple( tuple( sequence ) for sequence in variant ) for variant in triggers )
results = tuple( tuple( tuple( sequence ) for sequence in variant ) for variant in results )
# Iterate over all combinations of triggers and results
for trigger in triggers:
for result in results:
# Append Case
if operator == ":+":
macros_map.appendScanCode( trigger, result )
# Remove Case
elif operator == ":-":
macros_map.removeScanCode( trigger, result )
# Replace Case
elif operator == ":":
macros_map.replaceScanCode( trigger, result )
def eval_usbCode( triggers, operator, results ):
# Convert to lists of lists of lists to tuples of tuples of tuples
# Tuples are non-mutable, and can be used has index items
triggers = tuple( tuple( tuple( sequence ) for sequence in variant ) for variant in triggers )
results = tuple( tuple( tuple( sequence ) for sequence in variant ) for variant in results )
# Iterate over all combinations of triggers and results
for trigger in triggers:
scanCodes = macros_map.lookupUSBCodes( trigger )
for scanCode in scanCodes:
for result in results:
# Cache assignment until file finishes processing
macros_map.cacheAssignment( operator, scanCode, result )
def eval_variable( name, content ):
# Content might be a concatenation of multiple data types, convert everything into a single string
assigned_content = ""
for item in content:
assigned_content += str( item )
variable_dict[ name ] = assigned_content
def eval_capability( name, function, args ):
capabilities_dict[ name ] = [ function, args ]
map_scanCode = unarg( eval_scanCode )
map_usbCode = unarg( eval_usbCode )
set_variable = unarg( eval_variable )
set_capability = unarg( eval_capability )
## Sub Rules
usbCode = tokenType('USBCode') >> make_usbCode
scanCode = tokenType('ScanCode') >> make_scanCode
name = tokenType('Name')
number = tokenType('Number') >> make_number
comma = tokenType('Comma')
dash = tokenType('Dash')
plus = tokenType('Plus')
content = tokenType('VariableContents')
string = tokenType('String') >> make_string
unString = tokenType('String') # When the double quotes are still needed for internal processing
seqString = tokenType('SequenceString') >> make_seqString
# Code variants
code_end = tokenType('CodeEnd')
# Scan Codes
scanCode_start = tokenType('ScanCodeStart')
scanCode_range = number + skip( dash ) + number >> make_scanCode_range
scanCode_listElem = number >> listElem
scanCode_innerList = oneplus( ( scanCode_range | scanCode_listElem ) + skip( maybe( comma ) ) ) >> flatten
scanCode_expanded = skip( scanCode_start ) + scanCode_innerList + skip( code_end )
scanCode_elem = scanCode >> listElem
scanCode_combo = oneplus( ( scanCode_expanded | scanCode_elem ) + skip( maybe( plus ) ) )
scanCode_sequence = oneplus( scanCode_combo + skip( maybe( comma ) ) )
# USB Codes
usbCode_start = tokenType('USBCodeStart')
usbCode_range = ( number | unString ) + skip( dash ) + ( number | unString ) >> make_usbCode_range
usbCode_listElemTag = unString >> make_usbCode
usbCode_listElem = ( number | usbCode_listElemTag ) >> listElem
usbCode_innerList = oneplus( ( usbCode_range | usbCode_listElem ) + skip( maybe( comma ) ) ) >> flatten
usbCode_expanded = skip( usbCode_start ) + usbCode_innerList + skip( code_end )
usbCode_elem = usbCode >> listElem
usbCode_combo = oneplus( ( usbCode_expanded | usbCode_elem ) + skip( maybe( plus ) ) ) >> listElem
usbCode_sequence = oneplus( ( usbCode_combo | seqString ) + skip( maybe( comma ) ) ) >> oneLayerFlatten
# Capabilities
capFunc_arguments = many( number + skip( maybe( comma ) ) ) >> listToTuple
capFunc_elem = name + skip( parenthesis('(') ) + capFunc_arguments + skip( parenthesis(')') ) >> capArgExpander >> listElem
capFunc_combo = oneplus( ( usbCode_expanded | usbCode_elem | capFunc_elem ) + skip( maybe( plus ) ) ) >> listElem
capFunc_sequence = oneplus( ( capFunc_combo | seqString ) + skip( maybe( comma ) ) ) >> oneLayerFlatten
# Trigger / Result Codes
triggerCode_outerList = scanCode_sequence >> optionExpansion
triggerUSBCode_outerList = usbCode_sequence >> optionExpansion >> usbCodeToCapability
resultCode_outerList = capFunc_sequence >> optionExpansion >> usbCodeToCapability
## Main Rules
#| <variable> = <variable contents>;
variable_contents = name | content | string | number | comma | dash
variable_expression = name + skip( operator('=') ) + oneplus( variable_contents ) + skip( eol ) >> set_variable
#| <capability name> => <c function>;
capability_arguments = name + skip( operator(':') ) + number + skip( maybe( comma ) )
capability_expression = name + skip( operator('=>') ) + name + skip( parenthesis('(') ) + many( capability_arguments ) + skip( parenthesis(')') ) + skip( eol ) >> set_capability
#| <trigger> : <result>;
operatorTriggerResult = operator(':') | operator(':+') | operator(':-')
scanCode_expression = triggerCode_outerList + operatorTriggerResult + resultCode_outerList + skip( eol ) >> map_scanCode
usbCode_expression = triggerUSBCode_outerList + operatorTriggerResult + resultCode_outerList + skip( eol ) >> map_usbCode
def parse( tokenSequence ):
"""Sequence(Token) -> object"""
# Top-level Parser
expression = scanCode_expression | usbCode_expression | variable_expression | capability_expression
kll_text = many( expression )
kll_file = maybe( kll_text ) + skip( finished )
return kll_file.parse( tokenSequence )
def processKLLFile( filename ):
with open( filename ) as file:
data = file.read()
tokenSequence = tokenize( data )
#print ( pformat( tokenSequence ) ) # Display tokenization
tree = parse( tokenSequence )
### Main Entry Point ###
if __name__ == '__main__':
(baseFiles, defaultFiles, partialFileSets, backend_name, template, output) = processCommandLineArgs()
# Load backend module
global backend
backend_import = importlib.import_module( "backends.{0}".format( backend_name ) )
backend = backend_import.Backend( template )
# Process base layout files
for filename in baseFiles:
processKLLFile( filename )
macros_map.completeBaseLayout() # Indicates to macros_map that the base layout is complete
# Default combined layer
for filename in defaultFiles:
processKLLFile( filename )
# Apply assignment cache, see 5.1.2 USB Codes for why this is necessary
macros_map.replayCachedAssignments()
# Iterate through additional layers
for partial in partialFileSets:
# Increment layer for each -p option
macros_map.addLayer()
# Iterate and process each of the file in the layer
for filename in partial:
processKLLFile( filename )
# Apply assignment cache, see 5.1.2 USB Codes for why this is necessary
macros_map.replayCachedAssignments()
# Remove un-marked keys to complete the partial layer
macros_map.removeUnmarked()
# Do macro correlation and transformation
macros_map.generate()
# Process needed templating variables using backend
backend.process( capabilities_dict, macros_map )
# Generate output file using template and backend
backend.generate( output )
# Successful Execution
sys.exit( 0 )