2024-02-12 23:23:38 +01:00
|
|
|
import textwrap, sys, re, string
|
2024-02-11 20:46:36 +01:00
|
|
|
|
2024-02-12 23:23:38 +01:00
|
|
|
# Names not defined in Compose.pre
|
|
|
|
xkb_char_extra_names = {
|
|
|
|
"space": " ",
|
|
|
|
"minus": "-",
|
|
|
|
"asterisk": "*",
|
|
|
|
"colon": ":",
|
|
|
|
"equal": "=",
|
|
|
|
"exclam": "!",
|
|
|
|
"grave": "`",
|
|
|
|
"parenleft": "(",
|
|
|
|
"parenright": ")",
|
|
|
|
"percent": "%",
|
|
|
|
"period": ".",
|
|
|
|
"plus": "+",
|
|
|
|
"question": "?",
|
|
|
|
"semicolon": ";",
|
|
|
|
"underscore": "_",
|
|
|
|
}
|
|
|
|
|
|
|
|
dropped_sequences = 0
|
|
|
|
|
|
|
|
# Parse XKB's Compose.pre files
|
|
|
|
def parse_sequences_file_xkb(fname):
|
|
|
|
# Parse a line of the form:
|
|
|
|
# <Multi_key> <minus> <space> : "~" asciitilde # TILDE
|
|
|
|
# Sequences not starting with <Multi_key> are ignored.
|
|
|
|
line_re = re.compile(r'^((?:\s*<[^>]+>)+)\s*:\s*"((?:[^"\\]+|\\.)+)"\s*(\S+)?\s*(?:#.+)?$')
|
|
|
|
char_re = re.compile(r'\s*<(?:U([a-fA-F0-9]{4,6})|([^>]+))>')
|
|
|
|
def parse_seq_line(line):
|
|
|
|
global dropped_sequences
|
|
|
|
prefix = "<Multi_key>"
|
|
|
|
if not line.startswith(prefix):
|
|
|
|
return None
|
|
|
|
m = re.match(line_re, line[len(prefix):])
|
|
|
|
if m == None:
|
|
|
|
return None
|
|
|
|
def_ = m.group(1)
|
|
|
|
try:
|
|
|
|
def_ = parse_seq_chars(def_)
|
|
|
|
result = parse_seq_result(m.group(2))
|
|
|
|
except Exception as e:
|
|
|
|
# print(str(e) + ". Sequence dropped: " + line.strip(), file=sys.stderr)
|
|
|
|
dropped_sequences += 1
|
|
|
|
return None
|
|
|
|
return def_, result
|
|
|
|
char_names = { **xkb_char_extra_names }
|
|
|
|
# Interpret character names of the form "U0000" or using [char_names].
|
|
|
|
def parse_seq_char(c):
|
|
|
|
uchar, named_char = c
|
|
|
|
if uchar != "":
|
|
|
|
return chr(int(uchar, 16))
|
|
|
|
# else is a named char
|
|
|
|
if len(named_char) == 1:
|
|
|
|
return named_char
|
|
|
|
if not named_char in char_names:
|
|
|
|
raise Exception("Unknown char: " + named_char)
|
|
|
|
return char_names[named_char]
|
|
|
|
# Interpret the left hand side of a sequence.
|
|
|
|
def parse_seq_chars(def_):
|
|
|
|
return list(map(parse_seq_char, re.findall(char_re, def_)))
|
|
|
|
# Interpret the result of a sequence, as outputed by [line_re].
|
|
|
|
def parse_seq_result(r):
|
|
|
|
if len(r) == 2 and r[0] == '\\':
|
|
|
|
return r[1]
|
|
|
|
# The state machine can't represent characters that do not fit in a
|
|
|
|
# 16-bit char. This breaks some sequences that output letters with
|
|
|
|
# combined diacritics or emojis.
|
|
|
|
if len(r) > 1 or ord(r[0]) > 65535:
|
|
|
|
raise Exception("Char out of range: " + r)
|
|
|
|
return r
|
|
|
|
# Populate [char_names] with the information present in the file.
|
|
|
|
with open(fname, "r") as inp:
|
|
|
|
for line in inp:
|
|
|
|
m = re.match(line_re, line)
|
|
|
|
if m == None or m.group(3) == None:
|
|
|
|
continue
|
|
|
|
try:
|
|
|
|
char_names[m.group(3)] = parse_seq_result(m.group(2))
|
|
|
|
except Exception:
|
|
|
|
pass
|
|
|
|
# Parse the sequences
|
2024-02-11 20:46:36 +01:00
|
|
|
with open(fname, "r") as inp:
|
2024-02-12 23:23:38 +01:00
|
|
|
seqs = []
|
|
|
|
for line in inp:
|
|
|
|
s = parse_seq_line(line)
|
|
|
|
if s != None:
|
|
|
|
seqs.append(s)
|
|
|
|
return seqs
|
|
|
|
|
|
|
|
# Format of the sequences file is determined by its extension
|
|
|
|
def parse_sequences_file(fname):
|
|
|
|
if fname.endswith(".pre"):
|
|
|
|
return parse_sequences_file_xkb(fname)
|
|
|
|
raise Exception(fname + ": Unsupported format")
|
2024-02-11 20:46:36 +01:00
|
|
|
|
|
|
|
# Turn a list of sequences into a trie.
|
|
|
|
def add_sequences_to_trie(seqs, trie):
|
|
|
|
for seq, result in seqs:
|
|
|
|
t_ = trie
|
|
|
|
i = 0
|
|
|
|
while i < len(seq) - 1:
|
|
|
|
c = seq[i]
|
|
|
|
if c not in t_:
|
|
|
|
t_[c] = {}
|
|
|
|
t_ = t_[c]
|
|
|
|
i += 1
|
|
|
|
c = seq[i]
|
|
|
|
t_[c] = result
|
|
|
|
|
|
|
|
# Compile the trie into a state machine.
|
|
|
|
def make_automata(tree_root):
|
|
|
|
states = []
|
|
|
|
def add_tree(t):
|
|
|
|
# Index and size of the new node
|
|
|
|
i = len(states)
|
|
|
|
s = len(t.keys())
|
|
|
|
# Add node header
|
2024-02-12 23:23:38 +01:00
|
|
|
states.append(("\0", s + 1))
|
2024-02-11 20:46:36 +01:00
|
|
|
i += 1
|
|
|
|
# Reserve space for the current node in both arrays
|
|
|
|
for c in range(s):
|
|
|
|
states.append((None, None))
|
|
|
|
# Add nested nodes and fill the current node
|
|
|
|
for c in sorted(t.keys()):
|
|
|
|
node_i = len(states)
|
|
|
|
add_node(t[c])
|
|
|
|
states[i] = (c, node_i)
|
|
|
|
i += 1
|
|
|
|
def add_leaf(c):
|
|
|
|
states.append((c, 1))
|
|
|
|
def add_node(n):
|
|
|
|
if type(n) == str:
|
|
|
|
add_leaf(n)
|
|
|
|
else:
|
|
|
|
add_tree(n)
|
|
|
|
add_tree(tree_root)
|
|
|
|
return states
|
|
|
|
|
2024-02-12 23:23:38 +01:00
|
|
|
def batched(ar, n):
|
|
|
|
i = 0
|
|
|
|
while i + n < len(ar):
|
|
|
|
yield ar[i:i+n]
|
|
|
|
i += n
|
|
|
|
if i < len(ar):
|
|
|
|
yield ar[i:]
|
|
|
|
|
2024-02-11 20:46:36 +01:00
|
|
|
# Print the state machine compiled by make_automata into java code that can be
|
|
|
|
# used by [ComposeKeyData.java].
|
|
|
|
def gen_java(machine):
|
2024-02-12 23:23:38 +01:00
|
|
|
chars_map = {
|
|
|
|
# These characters cannot be used in unicode form as Java's parser
|
|
|
|
# unescape unicode sequences before parsing.
|
|
|
|
"\"": "\\\"",
|
|
|
|
"\\": "\\\\",
|
|
|
|
"\n": "\\n",
|
|
|
|
ord("\""): "\\\"",
|
|
|
|
ord("\\"): "\\\\",
|
|
|
|
ord("\n"): "\\n",
|
|
|
|
}
|
|
|
|
def char_repr(c):
|
|
|
|
if c in chars_map:
|
|
|
|
return chars_map[c]
|
|
|
|
if type(c) == int: # The edges array contains ints
|
|
|
|
return "\\u%04x" % c
|
|
|
|
if c in string.printable:
|
|
|
|
return c
|
|
|
|
return "\\u%04x" % ord(c)
|
|
|
|
def gen_array(array):
|
|
|
|
chars = list(map(char_repr, array))
|
|
|
|
return "\" +\n \"".join(map(lambda b: "".join(b), batched(chars, 72)))
|
2024-02-11 20:46:36 +01:00
|
|
|
print("""package juloo.keyboard2;
|
|
|
|
|
|
|
|
/** This file is generated, see [srcs/compose/compile.py]. */
|
|
|
|
|
|
|
|
public final class ComposeKeyData
|
|
|
|
{
|
2024-02-12 23:23:38 +01:00
|
|
|
public static final char[] states =
|
|
|
|
("%s").toCharArray();
|
2024-02-11 20:46:36 +01:00
|
|
|
|
2024-02-12 23:23:38 +01:00
|
|
|
public static final char[] edges =
|
|
|
|
("%s").toCharArray();
|
2024-02-11 20:46:36 +01:00
|
|
|
}""" % (
|
2024-02-12 23:23:38 +01:00
|
|
|
# Break the edges array every few characters using string concatenation.
|
|
|
|
gen_array(map(lambda s: s[0], machine)),
|
|
|
|
gen_array(map(lambda s: s[1], machine)),
|
2024-02-11 20:46:36 +01:00
|
|
|
))
|
|
|
|
|
|
|
|
total_sequences = 0
|
|
|
|
trie = {}
|
|
|
|
for fname in sys.argv[1:]:
|
|
|
|
sequences = parse_sequences_file(fname)
|
|
|
|
add_sequences_to_trie(sequences, trie)
|
|
|
|
total_sequences += len(sequences)
|
2024-02-12 23:23:38 +01:00
|
|
|
automata = make_automata(trie)
|
|
|
|
gen_java(automata)
|
|
|
|
print("Compiled %d sequences into %d states. Dropped %d sequences." % (total_sequences, len(automata), dropped_sequences), file=sys.stderr)
|