86d49acdb0
A new file src/data/generate-data.py has been added. It is written in Python 3 and has some new features like: - It doesn't need Chrome to work, the script fetches the page automatically; - It does the translation of HTML tags for better user experiences in editors. In addition, a new section has been added to README.md to reflect this change.
106 lines
3.6 KiB
Python
106 lines
3.6 KiB
Python
#!/usr/bin/env python3
|
|
import urllib.request
|
|
import re
|
|
import minify_html
|
|
|
|
zig_version = 'master'
|
|
|
|
|
|
def fix_ul(s):
|
|
l = s.split('<li>')
|
|
l.insert(0, '')
|
|
return '\n - '.join(l)
|
|
|
|
|
|
url = f'https://raw.githubusercontent.com/ziglang/zig/{zig_version}/doc/langref.html.in'
|
|
res = urllib.request.urlopen(url)
|
|
page = res.read().decode('utf-8')
|
|
print('''const Builtin = struct {
|
|
name: []const u8,
|
|
signature: []const u8,
|
|
snippet: []const u8,
|
|
documentation: []const u8,
|
|
arguments: []const []const u8,
|
|
};
|
|
|
|
pub const builtins = [_]Builtin{''')
|
|
pattern = r'{#header_open\|(@\S+?)#}(.+?){#header_close#}'
|
|
for match in re.finditer(pattern, page, re.M | re.S):
|
|
blk = match[2].strip(' \n')
|
|
name = match[1]
|
|
signature = re.search(r'<pre>{#syntax#}(.+?){#endsyntax#}</pre>', blk,
|
|
re.M | re.S)[1].replace('\n ', '').replace('"', '\\"')
|
|
snippet = name
|
|
if f'{name}()' in signature:
|
|
params = None
|
|
snippet += '()'
|
|
else:
|
|
params = []
|
|
i = signature.index('(') + 1
|
|
level = 1
|
|
j = i
|
|
while i < len(signature):
|
|
if signature[i] == '(':
|
|
level += 1
|
|
elif signature[i] == ')':
|
|
level -= 1
|
|
if signature[i] == ',' and level == 1:
|
|
params.append(signature[j:i])
|
|
j = i + 2
|
|
if level == 0:
|
|
break
|
|
i += 1
|
|
params.append(signature[j:i])
|
|
snippet += '(${'
|
|
i = 1
|
|
for param in params:
|
|
snippet += f'{i}:{param}}}, ${{'
|
|
i += 1
|
|
snippet = snippet[:-4] + ')'
|
|
docs = re.sub(r'{#see_also\|[^#]+#}', '', blk)
|
|
docs = re.sub(
|
|
r' {#code_begin\|(obj|syntax|(test(\|(call|truncate))?))#}\n', ' <pre>{#syntax#}', docs)
|
|
docs = re.sub(
|
|
r' {#code_begin\|test_(err|safety)\|[^#]+#}\n', ' <pre>{#syntax#}', docs)
|
|
docs = docs.replace(' {#code_release_fast#}\n', '')
|
|
docs = docs.replace(' {#code_end#}', '{#endsyntax#}</pre>')
|
|
docs = docs.replace('\n{#endsyntax#}</pre>', '{#endsyntax#}</pre>')
|
|
docs = minify_html.minify(docs)
|
|
prefix = '</pre><p>'
|
|
docs = docs[docs.index(prefix)+len(prefix):]
|
|
docs = docs.replace('<p>', '\n\n')
|
|
docs = re.sub(r'{#(end)?syntax#}', '`', docs)
|
|
# @cDefine
|
|
docs = re.sub(r'<pre><code[^>]+>([^<]+)</code></pre>', '`\\1`', docs)
|
|
docs = re.sub(r'</?code>', '`', docs)
|
|
docs = docs.replace('<pre>`', '\n\n```zig\n')
|
|
docs = docs.replace('`</pre>', '\n```')
|
|
# @setFloatMode
|
|
docs = docs.replace('```<', '```\n<')
|
|
# @TypeOf
|
|
docs = re.sub(r'</?em>', '*', docs)
|
|
docs = re.sub(r'<a href=([^>]+)>([^<]+)</a>', '[\\2](\\1)', docs)
|
|
docs = re.sub(r'{#link\|([^|#]+)\|([^|#]+)#}',
|
|
lambda m: f'[{m[1]}](https://ziglang.org/documentation/{zig_version}/#{m[2].replace(" ","-")})', docs)
|
|
docs = re.sub(
|
|
r'{#link\|([^|#]+)#}', lambda m: f'[{m[1]}](https://ziglang.org/documentation/{zig_version}/#{m[1].replace(" ","-").replace("@","")})', docs)
|
|
docs = re.sub(r'<ul><li>(.+?)</ul>', lambda m: fix_ul(m[1]), docs)
|
|
|
|
print(' .{')
|
|
print(f' .name = "{name}",')
|
|
print(f' .signature = "{signature}",')
|
|
print(f' .snippet = "{snippet}",')
|
|
print(' .documentation =')
|
|
for line in docs.splitlines():
|
|
print(r' \\' + line)
|
|
print(' ,')
|
|
if params is None:
|
|
print(' .arguments = &.{},')
|
|
else:
|
|
print(' .arguments = &.{')
|
|
for param in params:
|
|
print(f' "{param}",')
|
|
print(' },')
|
|
print(' },')
|
|
print('};')
|