tartrazine/scripts/token_abbrevs.py

55 lines
1.8 KiB
Python

# Script to generate abbreviations for tokens. Parses all lexers
# and styles files to find all token names and generate a unique
# abbreviation for each one. The abbreviations are generated by
# taking the uppercase letters of the token name and converting
# them to lowercase. If the abbreviation is not unique, the script
# will print a warning and exit.
import sys
import string
import glob
tokens = {"Highlight"}
abbrevs = {"Highlight": "hl"}
def abbr(line):
return "".join(c for c in line if c in string.ascii_uppercase).lower()
def check_abbrevs():
if len(abbrevs) != len(tokens):
print("Warning: Abbreviations are not unique")
print(len(abbrevs), len(tokens))
sys.exit(1)
# Processes all files in lexers looking for token names
for fname in glob.glob("lexers/*.xml"):
with open(fname) as f:
for line in f:
if "<token" not in line:
continue
line = line.strip()
line = line.split('<token ',1)[-1]
line = line.split('"')[1]
abbrevs[line] = abbr(line)
tokens.add(line)
check_abbrevs()
# Processes all files in styles looking for token names too
for fname in glob.glob("styles/*.xml"):
with open(fname) as f:
for line in f:
if "<entry" not in line:
continue
line = line.strip()
line = line.split('type=',1)[-1]
line = line.split('"')[1]
abbrevs[line] = abbr(line)
tokens.add(line)
check_abbrevs()
with open ("src/constants/token_abbrevs.cr", "w") as outf:
outf.write("module Tartrazine\n")
outf.write(" Abbreviations = {\n")
for k in sorted(abbrevs.keys()):
outf.write(f' "{k}" => "{abbrevs[k]}",\n')
outf.write(" }\nend\n")