94 lines
3.2 KiB
Python
Executable File
94 lines
3.2 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
|
|
import code
|
|
import sys
|
|
import os
|
|
import glob
|
|
import re
|
|
|
|
#src_file = "src/duckstation-qt/qttranslations.cpp"
|
|
src_dir = os.path.join(os.path.dirname(__file__), "..", "src")
|
|
fa_file = os.path.join(os.path.dirname(__file__), "..", "dep", "imgui", "include", "IconsFontAwesome5.h")
|
|
pf_file = os.path.join(os.path.dirname(__file__), "..", "dep", "imgui", "include", "IconsPromptFont.h")
|
|
dst_file = os.path.join(os.path.dirname(__file__), "..", "src", "util", "imgui_manager.cpp")
|
|
|
|
all_source_files = glob.glob(os.path.join(src_dir, "**", "*.cpp"), recursive=True) + \
|
|
glob.glob(os.path.join(src_dir, "**", "*.h"), recursive=True) + \
|
|
glob.glob(os.path.join(src_dir, "**", "*.inl"), recursive=True)
|
|
|
|
tokens = set()
|
|
pf_tokens = set()
|
|
for filename in all_source_files:
|
|
data = None
|
|
with open(filename, "r") as f:
|
|
try:
|
|
data = f.read()
|
|
except:
|
|
continue
|
|
|
|
tokens = tokens.union(set(re.findall("(ICON_FA_[a-zA-Z0-9_]+)", data)))
|
|
pf_tokens = pf_tokens.union(set(re.findall("(ICON_PF_[a-zA-Z0-9_]+)", data)))
|
|
|
|
print("{}/{} tokens found.".format(len(tokens), len(pf_tokens)))
|
|
if len(tokens) == 0 and len(pf_tokens) == 0:
|
|
sys.exit(0)
|
|
|
|
u8_encodings = {}
|
|
with open(fa_file, "r") as f:
|
|
for line in f.readlines():
|
|
match = re.match("#define (ICON_FA_[^ ]+) \"([^\"]+)\"", line)
|
|
if match is None:
|
|
continue
|
|
u8_encodings[match[1]] = bytes.fromhex(match[2].replace("\\x", ""))
|
|
with open(pf_file, "r") as f:
|
|
for line in f.readlines():
|
|
match = re.match("#define (ICON_PF_[^ ]+) \"([^\"]+)\"", line)
|
|
if match is None:
|
|
continue
|
|
u8_encodings[match[1]] = bytes.fromhex(match[2].replace("\\x", ""))
|
|
|
|
out_pattern = "(static constexpr ImWchar range_fa\[\] = \{)[0-9A-Z_a-z, \n]+(\};)"
|
|
out_pf_pattern = "(static constexpr ImWchar range_pf\[\] = \{)[0-9A-Z_a-z, \n]+(\};)"
|
|
|
|
def get_pairs(tokens):
|
|
codepoints = list()
|
|
for token in tokens:
|
|
u8_bytes = u8_encodings[token]
|
|
u8 = str(u8_bytes, "utf-8")
|
|
u32 = u8.encode("utf-32le")
|
|
if len(u32) > 4:
|
|
raise ValueError("{} {} too long".format(u8_bytes, token))
|
|
|
|
codepoint = int.from_bytes(u32, byteorder="little", signed=False)
|
|
codepoints.append(codepoint)
|
|
codepoints.sort()
|
|
codepoints.append(0) # null terminator
|
|
|
|
startc = codepoints[0]
|
|
endc = None
|
|
pairs = [startc]
|
|
for codepoint in codepoints:
|
|
if endc is not None and (endc + 1) != codepoint:
|
|
pairs.append(endc)
|
|
pairs.append(codepoint)
|
|
startc = codepoint
|
|
endc = codepoint
|
|
else:
|
|
endc = codepoint
|
|
pairs.append(endc)
|
|
|
|
pairs_str = ",".join(list(map(lambda x: "0x{:x}".format(x), pairs)))
|
|
return pairs_str
|
|
|
|
with open(dst_file, "r") as f:
|
|
original = f.read()
|
|
updated = re.sub(out_pattern, "\\1 " + get_pairs(tokens) + " \\2", original)
|
|
updated = re.sub(out_pf_pattern, "\\1 " + get_pairs(pf_tokens) + " \\2", updated)
|
|
if original != updated:
|
|
with open(dst_file, "w") as f:
|
|
f.write(updated)
|
|
print("Updated {}".format(dst_file))
|
|
else:
|
|
print("Skipping updating {}".format(dst_file))
|
|
|