module main import os import net.urllib import strings import markdown import regex import v.scanner import v.ast import v.token import v.doc import v.pref const ( css_js_assets = ['doc.css', 'normalize.css', 'doc.js', 'dark-mode.js'] default_theme = os.resource_abs_path('theme') link_svg = '' single_quote = "'" double_quote = '"' no_quotes_replacement = [single_quote, '', double_quote, ''] ) enum HighlightTokenTyp { unone boolean builtin char comment function keyword name number operator punctuation string symbol none_ module_ prefix } struct SearchModuleResult { description string link string } struct SearchResult { prefix string badge string description string link string } fn (vd VDoc) render_search_index(out Output) { mut js_search_index := strings.new_builder(200) mut js_search_data := strings.new_builder(200) js_search_index.write_string('var searchModuleIndex = [') js_search_data.write_string('var searchModuleData = [') for i, title in vd.search_module_index { data := vd.search_module_data[i] js_search_index.write_string('"${title}",') js_search_data.write_string('["${data.description}","${data.link}"],') } js_search_index.writeln('];') js_search_index.write_string('var searchIndex = [') js_search_data.writeln('];') js_search_data.write_string('var searchData = [') for i, title in vd.search_index { data := vd.search_data[i] js_search_index.write_string('"${title}",') // array instead of object to reduce file size js_search_data.write_string('["${data.badge}","${data.description}","${data.link}","${data.prefix}"],') } js_search_index.writeln('];') js_search_data.writeln('];') out_file_path := os.join_path(out.path, 'search_index.js') os.write_file(out_file_path, js_search_index.str() + js_search_data.str()) or { panic(err) } } fn (mut vd VDoc) render_static_html(out Output) { vd.assets = { 'doc_css': vd.get_resource(css_js_assets[0], out) 'normalize_css': vd.get_resource(css_js_assets[1], out) 'doc_js': vd.get_resource(css_js_assets[2], out) 'dark_mode_js': vd.get_resource(css_js_assets[3], out) 'light_icon': vd.get_resource('light.svg', out) 'dark_icon': vd.get_resource('dark.svg', out) 'menu_icon': vd.get_resource('menu.svg', out) 'arrow_icon': vd.get_resource('arrow.svg', out) } } fn (vd VDoc) get_resource(name string, out Output) string { cfg := vd.cfg path := os.join_path(cfg.theme_dir, name) mut res := os.read_file(path) or { panic('vdoc: could not read ${path}') } /* if minify { if name.ends_with('.js') { res = js_compress(res) } else { res = res.split_into_lines().map(it.trim_space()).join('') } } */ // TODO: Make SVG inline for now if cfg.inline_assets || path.ends_with('.svg') { return res } else { output_path := os.join_path(out.path, name) if !os.exists(output_path) { println('Generating ${out.typ} in "${output_path}"') os.write_file(output_path, res) or { panic(err) } } return name } } fn (mut vd VDoc) collect_search_index(out Output) { cfg := vd.cfg for doc in vd.docs { mod := doc.head.name vd.search_module_index << mod comments := if cfg.include_examples { doc.head.merge_comments() } else { doc.head.merge_comments_without_examples() } vd.search_module_data << SearchModuleResult{ description: trim_doc_node_description(comments) link: vd.get_file_name(mod, out) } for _, dn in doc.contents { vd.create_search_results(mod, dn, out) } } } fn (mut vd VDoc) create_search_results(mod string, dn doc.DocNode, out Output) { cfg := vd.cfg if dn.kind == .const_group { return } comments := if cfg.include_examples { dn.merge_comments() } else { dn.merge_comments_without_examples() } dn_description := trim_doc_node_description(comments) vd.search_index << dn.name vd.search_data << SearchResult{ prefix: if dn.parent_name != '' { '${dn.kind} (${dn.parent_name})' } else { '${dn.kind} ' } description: dn_description badge: mod link: vd.get_file_name(mod, out) + '#' + get_node_id(dn) } for child in dn.children { vd.create_search_results(mod, child, out) } } fn (vd VDoc) write_content(cn &doc.DocNode, d &doc.Doc, mut hw strings.Builder) { cfg := vd.cfg base_dir := os.dir(os.real_path(cfg.input_path)) file_path_name := if cfg.is_multi { cn.file_path.replace('${base_dir}/', '') } else { os.file_name(cn.file_path) } src_link := get_src_link(vd.manifest.repo_url, file_path_name, cn.pos.line_nr + 1) if cn.content.len != 0 || cn.name == 'Constants' { hw.write_string(doc_node_html(cn, src_link, false, cfg.include_examples, d.table)) } for child in cn.children { child_file_path_name := child.file_path.replace('${base_dir}/', '') child_src_link := get_src_link(vd.manifest.repo_url, child_file_path_name, child.pos.line_nr + 1) hw.write_string(doc_node_html(child, child_src_link, false, cfg.include_examples, d.table)) } } fn (vd VDoc) gen_html(d doc.Doc) string { cfg := vd.cfg mut symbols_toc := strings.new_builder(200) mut modules_toc := strings.new_builder(200) mut contents := strings.new_builder(200) dcs_contents := d.contents.arr() // generate toc first contents.writeln(doc_node_html(d.head, '', true, cfg.include_examples, d.table)) if is_module_readme(d.head) { write_toc(d.head, mut symbols_toc) } for cn in dcs_contents { vd.write_content(&cn, &d, mut contents) write_toc(cn, mut symbols_toc) } // write head // write css mut version := if vd.manifest.version.len != 0 { vd.manifest.version } else { '' } version = [version, @VHASH].join(' ') header_name := if cfg.is_multi && vd.docs.len > 1 { os.file_name(os.real_path(cfg.input_path)) } else { d.head.name } // write nav1 if cfg.is_multi || vd.docs.len > 1 { mut submod_prefix := '' for i, dc in vd.docs { if i - 1 >= 0 && dc.head.name.starts_with(submod_prefix + '.') { continue } names := dc.head.name.split('.') submod_prefix = if names.len > 1 { names[0] } else { dc.head.name } mut href_name := './${dc.head.name}.html' if (cfg.is_vlib && dc.head.name == 'builtin' && !cfg.include_readme) || dc.head.name == 'README' { href_name = './index.html' } else if submod_prefix !in vd.docs.map(it.head.name) { href_name = '#' } submodules := vd.docs.filter(it.head.name.starts_with(submod_prefix + '.')) dropdown := if submodules.len > 0 { vd.assets['arrow_icon'] } else { '' } active_class := if dc.head.name == d.head.name { ' active' } else { '' } modules_toc.write_string('
  • ') for j, cdoc in submodules { if j == 0 { modules_toc.write_string('') } } modules_toc.write_string('
  • ') } } modules_toc_str := modules_toc.str() symbols_toc_str := symbols_toc.str() result := (os.read_file(os.join_path(cfg.theme_dir, 'index.html')) or { panic(err) }).replace('{{ title }}', d.head.name).replace('{{ head_name }}', header_name).replace('{{ version }}', version).replace('{{ light_icon }}', vd.assets['light_icon']).replace('{{ dark_icon }}', vd.assets['dark_icon']).replace('{{ menu_icon }}', vd.assets['menu_icon']).replace('{{ head_assets }}', if cfg.inline_assets { '\n${tabs[0]}\n${tabs[0]}\n${tabs[0]}' } else { '\n${tabs[0]}\n${tabs[0]}\n${tabs[0]}' }).replace('{{ toc_links }}', if cfg.is_multi || vd.docs.len > 1 { modules_toc_str } else { symbols_toc_str }).replace('{{ contents }}', contents.str()).replace('{{ right_content }}', if cfg.is_multi && d.head.name != 'README' { '
    ' } else { '' }).replace('{{ footer_content }}', gen_footer_text(d, !cfg.no_timestamp)).replace('{{ footer_assets }}', if cfg.inline_assets { '' } else { '' }) return result } fn get_src_link(repo_url string, file_name string, line_nr int) string { mut url := urllib.parse(repo_url) or { return '' } if url.path.len <= 1 || file_name.len == 0 { return '' } url.path = url.path.trim_right('/') + match url.host { 'github.com' { '/blob/master/${file_name}' } 'gitlab.com' { '/-/blob/master/${file_name}' } 'git.sir.ht' { '/tree/master/${file_name}' } else { '' } } if url.path == '/' { return '' } url.fragment = 'L${line_nr}' return url.str() } fn html_highlight(code string, tb &ast.Table) string { builtin := ['bool', 'string', 'i8', 'i16', 'int', 'i64', 'i128', 'byte', 'u16', 'u32', 'u64', 'u128', 'rune', 'f32', 'f64', 'int_literal', 'float_literal', 'byteptr', 'voidptr', 'any'] highlight_code := fn (tok token.Token, typ HighlightTokenTyp) string { lit := if typ in [.unone, .operator, .punctuation] { tok.kind.str() } else if typ == .string { "'${tok.lit}'" } else if typ == .char { '`${tok.lit}`' } else if typ == .comment { if tok.lit != '' && tok.lit[0] == 1 { '//${tok.lit[1..]}' } else { '//${tok.lit}' } } else { tok.lit } if typ in [.unone, .name] { return lit } return '${lit}' } mut s := scanner.new_scanner(code, .parse_comments, &pref.Preferences{}) mut tok := s.scan() mut next_tok := s.scan() mut buf := strings.new_builder(200) mut i := 0 for i < code.len { if i == tok.pos { mut tok_typ := HighlightTokenTyp.unone match tok.kind { .name { if tok.lit in builtin || tb.known_type(tok.lit) { tok_typ = .builtin } else if next_tok.kind == .lcbr { tok_typ = .symbol } else if next_tok.kind == .lpar || (!tok.lit[0].is_capital() && next_tok.kind == .lt && next_tok.pos == tok.pos + tok.lit.len) { tok_typ = .function } else { tok_typ = .name } } .comment { tok_typ = .comment } .chartoken { tok_typ = .char } .string { tok_typ = .string } .number { tok_typ = .number } .key_true, .key_false { tok_typ = .boolean } .lpar, .lcbr, .rpar, .rcbr, .lsbr, .rsbr, .semicolon, .colon, .comma, .dot, .dotdot, .ellipsis { tok_typ = .punctuation } else { if token.is_key(tok.lit) || token.is_decl(tok.kind) { tok_typ = .keyword } else if tok.kind == .decl_assign || tok.kind.is_assign() || tok.is_unary() || tok.kind.is_relational() || tok.kind.is_infix() || tok.kind.is_postfix() { tok_typ = .operator } } } buf.write_string(highlight_code(tok, tok_typ)) if next_tok.kind != .eof { i = tok.pos + tok.len tok = next_tok next_tok = s.scan() } else { break } } else { buf.write_u8(code[i]) i++ } } return buf.str() } fn doc_node_html(dn doc.DocNode, link string, head bool, include_examples bool, tb &ast.Table) string { mut dnw := strings.new_builder(200) head_tag := if head { 'h1' } else { 'h2' } comments := dn.merge_comments_without_examples() // Allow README.md to go through unescaped except for script tags escaped_html := if head && is_module_readme(dn) { // Strip markdown [TOC] directives, since we generate our own. stripped := comments.replace('[TOC]', '') markdown_escape_script_tags(stripped) } else { html_tag_escape(comments) } md_content := markdown.to_html(escaped_html) highlighted_code := html_highlight(dn.content, tb) node_class := if dn.kind == .const_group { ' const' } else { '' } sym_name := get_sym_name(dn) mut deprecated_tags := dn.tags.filter(it.starts_with('deprecated')) if doc.should_sort { deprecated_tags.sort() } mut tags := dn.tags.filter(!it.starts_with('deprecated')) if doc.should_sort { tags.sort() } mut node_id := get_node_id(dn) mut hash_link := if !head { ' #' } else { '' } if head && is_module_readme(dn) { node_id = 'readme_${node_id}' hash_link = ' #' } dnw.writeln('${tabs[1]}
    ') if dn.name.len > 0 { if dn.kind == .const_group { dnw.write_string('${tabs[2]}
    <${head_tag}>${sym_name}${hash_link}') } else { dnw.write_string('${tabs[2]}
    <${head_tag}>${dn.kind} ${sym_name}${hash_link}') } if link.len != 0 { dnw.write_string('${link_svg}') } dnw.write_string('
    ') } if deprecated_tags.len > 0 { attributes := deprecated_tags.map('
    ${no_quotes(it)}
    ').join('') dnw.writeln('
    ${attributes}
    ') } if tags.len > 0 { attributes := tags.map('
    ${it}
    ').join('') dnw.writeln('
    ${attributes}
    ') } if !head && dn.content.len > 0 { dnw.writeln('
    ${highlighted_code}
    ') } // do not mess with md_content further, its formatting is important, just output it 1:1 ! dnw.writeln('${md_content}\n') // Write examples if any found examples := dn.examples() if include_examples && examples.len > 0 { example_title := if examples.len > 1 { 'Examples' } else { 'Example' } dnw.writeln('

    ${example_title}

    ') for example in examples { hl_example := html_highlight(example, tb) dnw.writeln('
    ${hl_example}
    ') } dnw.writeln('
    ') } dnw.writeln('
    ') dnw_str := dnw.str() return dnw_str } fn html_tag_escape(str string) string { excaped_string := str.replace_each(['<', '<', '>', '>']) mut re := regex.regex_opt(r'`.+[(<)(>)].+`') or { regex.RE{} } if re.find_all_str(excaped_string).len > 0 { return str } return excaped_string } /* fn js_compress(str string) string { mut js := strings.new_builder(200) lines := str.split_into_lines() rules := [') {', ' = ', ', ', '{ ', ' }', ' (', '; ', ' + ', ' < ', ' - ', ' || ', ' var', ': ', ' >= ', ' && ', ' else if', ' === ', ' !== ', ' else '] clean := ['){', '=', ',', '{', '}', '(', ';', '+', '<', '-', '||', 'var', ':', '>=', '&&', 'else if', '===', '!==', 'else'] for line in lines { mut trimmed := line.trim_space() if trimmed.starts_with('//') || (trimmed.starts_with('/*') && trimmed.ends_with('*/')) { continue } for i in 0 .. rules.len - 1 { trimmed = trimmed.replace(rules[i], clean[i]) } js.write_string(trimmed) } js_str := js.str() return js_str } */ fn write_toc(dn doc.DocNode, mut toc strings.Builder) { mut toc_slug := if dn.name.len == 0 || dn.content.len == 0 { '' } else { slug(dn.name) } if toc_slug == '' && dn.children.len > 0 { if dn.children[0].name == '' { toc_slug = slug(dn.name) } else { toc_slug = slug(dn.name + '.' + dn.children[0].name) } } if is_module_readme(dn) { if dn.comments.len == 0 || (dn.comments.len > 0 && dn.comments[0].text.len == 0) { return } toc.write_string('
  • README') } else if dn.name != 'Constants' { toc.write_string('
  • ${dn.kind} ${dn.name}') toc.writeln(' ') } else { toc.write_string('
  • ${dn.name}') } toc.writeln('
  • ') } fn no_quotes(s string) string { return s.replace_each(no_quotes_replacement) }