From 4d5ee2ead874c7a436faa8754eb99215927ea94d Mon Sep 17 00:00:00 2001 From: Ralph Amissah Date: Sun, 27 Jun 2021 16:59:44 -0400 Subject: dir rename ./sundry (from ./misc) --- misc/util/d/cgi/search/README | 11 - misc/util/d/cgi/search/dub.sdl | 16 - .../d/cgi/search/src/spine_cgi_sqlite_search.d | 963 --------------------- misc/util/d/tools/markup_conversion/README | 1 - .../endnotes_inline_from_binary.d | 123 --- .../d/tools/markup_conversion/markup_changes.d | 136 --- .../markup_changes_header_and_content.d | 244 ------ ...arkup_conversion_from_sisu_ruby_to_sisu_spine.d | 367 -------- misc/util/d/tools/spine_scaffold.d | 134 --- 9 files changed, 1995 deletions(-) delete mode 100644 misc/util/d/cgi/search/README delete mode 100644 misc/util/d/cgi/search/dub.sdl delete mode 100644 misc/util/d/cgi/search/src/spine_cgi_sqlite_search.d delete mode 100644 misc/util/d/tools/markup_conversion/README delete mode 100755 misc/util/d/tools/markup_conversion/endnotes_inline_from_binary.d delete mode 100644 misc/util/d/tools/markup_conversion/markup_changes.d delete mode 100755 misc/util/d/tools/markup_conversion/markup_changes_header_and_content.d delete mode 100755 misc/util/d/tools/markup_conversion/markup_conversion_from_sisu_ruby_to_sisu_spine.d delete mode 100755 misc/util/d/tools/spine_scaffold.d (limited to 'misc/util/d') diff --git a/misc/util/d/cgi/search/README b/misc/util/d/cgi/search/README deleted file mode 100644 index eb8fcde..0000000 --- a/misc/util/d/cgi/search/README +++ /dev/null @@ -1,11 +0,0 @@ -change db name to match name of db you create -cv.db_selected = "spine.search.sql.db"; - -~dr/bin/spine-ldc -v --sqlite-db-create --sqlite-db-filename="spine.search.db" --cgi-sqlite-search-filename="spine-search" --output=/var/www ~grotto/repo/git.repo/code/project-spine/doc-reform-markup/markup_samples/markup/pod/* - -~dr/bin/spine-ldc -v --sqlite-update --sqlite-db-filename="spine.search.db" --output=/var/www ~grotto/repo/git.repo/code/project-spine/doc-reform-markup/markup_samples/markup/pod/* - -cd util/d/cgi/search/src -dub --force --compiler=ldc2 && sudo cp -v cgi-bin/spine-search /usr/lib/cgi-bin/. - -http://localhost/cgi-bin/spine-search? diff --git a/misc/util/d/cgi/search/dub.sdl b/misc/util/d/cgi/search/dub.sdl deleted file mode 100644 index 30b076d..0000000 --- a/misc/util/d/cgi/search/dub.sdl +++ /dev/null @@ -1,16 +0,0 @@ -name "spine_search" -description "A minimal D application." -authors "ralph" -copyright "Copyright © 2021, ralph" -license "GPL-3.0+" -dependency "d2sqlite3" version="~>0.18.3" -dependency "arsd-official": "7.2.0" - subConfiguration "arsd-official:cgi" "cgi" -targetType "executable" -targetPath "./cgi-bin" -mainSourceFile "src/spine_cgi_sqlite_search.d" -configuration "default" { - targetType "executable" - targetName "spine-search" - postGenerateCommands "notify-send -t 0 'D executable ready' 'spine cgi sqlite search d'" -} diff --git a/misc/util/d/cgi/search/src/spine_cgi_sqlite_search.d b/misc/util/d/cgi/search/src/spine_cgi_sqlite_search.d deleted file mode 100644 index 987c319..0000000 --- a/misc/util/d/cgi/search/src/spine_cgi_sqlite_search.d +++ /dev/null @@ -1,963 +0,0 @@ -/+ -- Name: Spine, Doc Reform [a part of] - - Description: documents, structuring, processing, publishing, search - - static content generator - - - Author: Ralph Amissah - [ralph.amissah@gmail.com] - - - Copyright: (C) 2015 - 2021 Ralph Amissah, All Rights - Reserved. - - - License: AGPL 3 or later: - - Spine (SiSU), a framework for document structuring, publishing and - search - - Copyright (C) Ralph Amissah - - This program is free software: you can redistribute it and/or modify it - under the terms of the GNU AFERO General Public License as published by the - Free Software Foundation, either version 3 of the License, or (at your - option) any later version. - - This program is distributed in the hope that it will be useful, but WITHOUT - ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or - FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for - more details. - - You should have received a copy of the GNU General Public License along with - this program. If not, see [https://www.gnu.org/licenses/]. - - If you have Internet connection, the latest version of the AGPL should be - available at these locations: - [https://www.fsf.org/licensing/licenses/agpl.html] - [https://www.gnu.org/licenses/agpl.html] - - - Spine (by Doc Reform, related to SiSU) uses standard: - - docReform markup syntax - - standard SiSU markup syntax with modified headers and minor modifications - - docReform object numbering - - standard SiSU object citation numbering & system - - - Hompages: - [https://www.doc_reform.org] - [https://www.sisudoc.org] - - - Git - [https://git.sisudoc.org/projects/?p=software/spine.git;a=summary] - -+/ -/+ dub.sdl - name "spine search" - description "spine cgi search" -+/ -import std.format; -import std.range; -import std.regex; -import arsd.cgi; -import d2sqlite3; -import std.process : environment; -void cgi_function_intro(Cgi cgi) { - string header; - string table; - string form; - struct Config { - string http_request_type; - string http_host; - // string server_name; - string doc_root; - string cgi_root; - string cgi_script; - string data_path_html; - string db_path; - string query_string; - string http_url; - string request_method; - } - auto conf = Config(); - conf.http_request_type = environment.get("REQUEST_SCHEME", "http"); - conf.http_host = environment.get("HTTP_HOST", "localhost"); - // conf.server_name = environment.get("SERVER_NAME", "localhost"); - conf.doc_root = environment.get("DOCUMENT_ROOT", "/var/www/html"); - conf.cgi_root = environment.get("CONTEXT_DOCUMENT_ROOT", "/usr/lib/cgi-bin/"); - // conf.cgi_script = environment.get("SCRIPT_NAME", "/cgi-bin/spine-search"); - conf.query_string = environment.get("QUERY_STRING", ""); - conf.http_url = environment.get("HTTP_REFERER", conf.http_request_type ~ "://" ~ conf.http_host ~ conf.cgi_script ~ "?" ~ conf.query_string); - conf.db_path = "/var/www/html/sqlite/"; // conf.http_host ~ "/sqlite/"; - conf.request_method = environment.get("REQUEST_METHOD", "POST"); - struct CGI_val { - string db_selected = ""; - string sql_match_limit = ""; // radio: ( 1000 | 2500 ) - string sql_match_offset = ""; - string search_text = ""; - string results_type = ""; // index - bool checked_echo = false; - bool checked_stats = false; - bool checked_url = false; - bool checked_searched = false; - bool checked_tip = false; - bool checked_sql = false; - } - auto cv = CGI_val(); - cv.db_selected = "spine.search.db"; // config, set db name - auto text_fields() { - string canned_query_str = environment.get("QUERY_STRING", ""); - if ("query_string" in cgi.post) { - canned_query_str = environment.get("QUERY_STRING", ""); - } - string[string] canned_query; - if (conf.request_method == "POST") { - } else if (conf.request_method == "GET") { - foreach (pair_str; canned_query_str.split("&")) { - // cgi.write(pair_str ~ "
"); - string[] pair = pair_str.split("="); - canned_query[pair[0]] = pair[1]; - } - // foreach (field, content; canned_query) { - // cgi.write(field ~ ": " ~ content ~ "
"); - // } - } - static struct Rgx { - // static canned_query = ctRegex!(`\A(?P.+)\Z`, "m"); - static search_text_area = ctRegex!(`\A(?P.+)\Z`, "m"); - // static fulltext = ctRegex!(`\A(?P.+)\Z`, "m"); - static line = ctRegex!(`^(?P.+?)(?: ~|$)`, "m"); - static text = ctRegex!(`(?:^|\s~\s*)text:\s+(?P.+?)(?: ~|$)`, "m"); - static author = ctRegex!(`(?:^|\s~\s*)author:\s+(?P.+)$`, "m"); - static title = ctRegex!(`(?:^|\s~\s*)title:\s+(?P.+)$`, "m"); - static uid = ctRegex!(`(?:^|\s~\s*)uid:\s+(?P.+)$`, "m"); - static fn = ctRegex!(`(?:^|\s~\s*)fn:\s+(?P.+)$`, "m"); - static keywords = ctRegex!(`(?:^|\s~\s*)keywords:\s+(?P.+)$`, "m"); - static topic_register = ctRegex!(`(?:^|\s~\s*)topic_register:\s+(?P.+)$`, "m"); - static subject = ctRegex!(`(?:^|\s~\s*)subject:\s+(?P.+)$`, "m"); - static description = ctRegex!(`(?:^|\s~\s*)description:\s+(?P.+)$`, "m"); - static publisher = ctRegex!(`(?:^|\s~\s*)publisher:\s+(?P.+)$`, "m"); - static editor = ctRegex!(`(?:^|\s~\s*)editor:\s+(?P.+)$`, "m"); - static contributor = ctRegex!(`(?:^|\s~\s*)contributor:\s+(?P.+)$`, "m"); - static date = ctRegex!(`(?:^|\s~\s*)date:\s+(?P.+)$`, "m"); - static results_type = ctRegex!(`(?:^|\s~\s*)type:\s+(?P.+)$`, "m"); - static format = ctRegex!(`(?:^|\s~\s*)format:\s+(?P.+)$`, "m"); - static identifier = ctRegex!(`(?:^|\s~\s*)identifier:\s+(?P.+)$`, "m"); - static source = ctRegex!(`(?:^|\s~\s*)source:\s+(?P.+)$`, "m"); - static language = ctRegex!(`(?:^|\s~\s*)language:\s+(?P.+)$`, "m"); - static relation = ctRegex!(`(?:^|\s~\s*)relation:\s+(?P.+)$`, "m"); - static coverage = ctRegex!(`(?:^|\s~\s*)coverage:\s+(?P.+)$`, "m"); - static rights = ctRegex!(`(?:^|\s~\s*)rights:\s+(?P.+)$`, "m"); - static comment = ctRegex!(`(?:^|\s~\s*)comment:\s+(?P.+)$`, "m"); - // static abstract_ = ctRegex!(`(?:^|\s~\s*)abstract:\s+(?P.+)$`, "m"); - static src_filename_base = ctRegex!(`^src_filename_base:\s+(?P.+)$`, "m"); - } - struct searchFields { - string canned_query = ""; // GET canned_query == cq - string search_text_area = ""; // POST search_text_area == tsa - string text = ""; // text == txt - string author = ""; // author == au - string title = ""; // title == ti - string uid = ""; // uid == uid - string fn = ""; // fn == fn - string keywords = ""; // keywords == kw - string topic_register = ""; // topic_register == tr - string subject = ""; // subject == su - string description = ""; // description == de - string publisher = ""; // publisher == pb - string editor = ""; // editor == ed - string contributor = ""; // contributor == ct - string date = ""; // date == dt - string format = ""; // format == fmt - string identifier = ""; // identifier == id - string source = ""; // source == src sfn - string language = ""; // language == lng - string relation = ""; // relation == rl - string coverage = ""; // coverage == cv - string rights = ""; // rights == rgt - string comment = ""; // comment == cmt - // string abstract = ""; - string src_filename_base = ""; // src_filename_base == bfn - string results_type = ""; // results_type == rt radio - string sql_match_limit = ""; // sql_match_limit == sml radio - string sql_match_offset = ""; // sql_match_offset == smo - string stats = ""; // stats == sts checked - string echo = ""; // echo == ec checked - string url = ""; // url == url checked - string searched = ""; // searched == se checked - string sql = ""; // sql == sql checked - } - auto rgx = Rgx(); - auto got = searchFields(); - if (environment.get("REQUEST_METHOD", "POST") == "POST") { - if ("sf" in cgi.post) { - got.search_text_area = cgi.post["sf"]; - if (auto m = got.search_text_area.matchFirst(rgx.text)) { - got.text = m["matched"]; - got.canned_query ~= "sf=" ~ m["matched"]; - } else if (auto m = got.search_text_area.matchFirst(rgx.line)) { - if ( - !(m["matched"].matchFirst(rgx.author)) - && !(m["matched"].matchFirst(rgx.title)) - ) { - got.text = m["matched"]; - got.canned_query ~= "sf=" ~ m["matched"]; - } - } - if (auto m = got.search_text_area.matchFirst(rgx.author)) { - got.author = m["matched"]; - got.canned_query ~= "&au=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.title)) { - got.title = m["matched"]; - got.canned_query ~= "&ti=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.uid)) { - got.uid = m["matched"]; - got.canned_query ~= "&uid=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.fn)) { - got.fn = m["matched"]; - got.canned_query ~= "&fn=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.keywords)) { - got.keywords = m["matched"]; - got.canned_query ~= "&kw=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.topic_register)) { - got.topic_register = m["matched"]; - got.canned_query ~= "&tr=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.subject)) { - got.subject = m["matched"]; - got.canned_query ~= "&su=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.description)) { - got.description = m["matched"]; - got.canned_query ~= "&de=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.publisher)) { - got.publisher = m["matched"]; - got.canned_query ~= "&pb=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.editor)) { - got.editor = m["matched"]; - got.canned_query ~= "&ed=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.contributor)) { - got.contributor = m["matched"]; - got.canned_query ~= "&ct=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.date)) { - got.date = m["matched"]; - got.canned_query ~= "&dt=" ~ m["matched"]; - } - // if (auto m = got.search_text_area.matchFirst(rgx.results_type)) { - // got.results_type = m["matched"]; - // got.canned_query ~= "&rt=" ~ m["matched"]; - // } - if (auto m = got.search_text_area.matchFirst(rgx.format)) { - got.format = m["matched"]; - got.canned_query ~= "&fmt=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.identifier)) { - got.identifier = m["matched"]; - got.canned_query ~= "&id=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.source)) { - got.source = m["matched"]; - got.canned_query ~= "&src=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.language)) { - got.language = m["matched"]; - got.canned_query ~= "&lng=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.relation)) { - got.relation = m["matched"]; - got.canned_query ~= "&rl=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.coverage)) { - got.coverage = m["matched"]; - got.canned_query ~= "&cv=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.rights)) { - got.rights = m["matched"]; - got.canned_query ~= "&rgt=" ~ m["matched"]; - } - if (auto m = got.search_text_area.matchFirst(rgx.comment)) { - got.comment = m["matched"]; - got.canned_query ~= "&cmt=" ~ m["matched"]; - } - // if (auto m = search_text_area.matchFirst(rgx.abstract)) { - // got.abstract = m["matched"]; - // } - if (auto m = got.search_text_area.matchFirst(rgx.src_filename_base)) { - got.src_filename_base = m["matched"]; - got.canned_query ~= "&bfn=" ~ m["matched"]; - } - } - if ("fn" in cgi.post) { - got.fn = cgi.post["fn"]; - got.canned_query ~= "&fn=" ~ cgi.post["fn"]; - } - if ("rt" in cgi.post) { - got.results_type = cgi.post["rt"]; - got.canned_query ~= "&rt=" ~ cgi.post["rt"]; - } - if ("sts" in cgi.post) { - got.stats = cgi.post["sts"]; - got.canned_query ~= "&sts=" ~ cgi.post["sts"]; - } - if ("ec" in cgi.post) { - got.echo = cgi.post["ec"]; - got.canned_query ~= "&ec=" ~ cgi.post["ec"]; - } - if ("url" in cgi.post) { - got.url = cgi.post["url"]; - got.canned_query ~= "&url=" ~ cgi.post["url"]; - } - if ("se" in cgi.post) { - got.searched = cgi.post["se"]; - got.canned_query ~= "&se=" ~ cgi.post["se"]; - } - if ("sql" in cgi.post) { - got.sql = cgi.post["sql"]; - got.canned_query ~= "&sql=" ~ cgi.post["sql"]; - } - if ("sml" in cgi.post) { - got.sql_match_limit = cgi.post["sml"]; - got.canned_query ~= "&sml=" ~ cgi.post["sml"]; - } - if ("smo" in cgi.post) { - got.sql_match_offset = "0"; // cgi.post["smo"]; - got.canned_query ~= "&smo=0"; // ~ cgi.post["smo"]; - } - got.canned_query = got.canned_query.strip.split(" ").join("%20"); - conf.query_string = got.canned_query; - // cgi.write("f.canned_query: " ~ got.canned_query ~ "
"); - } else if (environment.get("REQUEST_METHOD", "POST") == "GET") { - got.canned_query = environment.get("QUERY_STRING", ""); - // cgi.write("f.canned_query: " ~ got.canned_query ~ "
"); - got.search_text_area = ""; - if ("sf" in canned_query && !(canned_query["sf"]).empty) { - got.text = canned_query["sf"].split("%20").join(" "); - got.search_text_area ~= "text: " ~ got.text ~ "\n"; - } - if ("au" in canned_query && !(canned_query["au"]).empty) { - got.author = canned_query["au"].split("%20").join(" "); - got.search_text_area ~= "author: " ~ got.author ~ "\n"; - } - if ("ti" in canned_query && !(canned_query["ti"]).empty) { - got.title = canned_query["ti"].split("%20").join(" "); - got.search_text_area ~= "title: " ~ got.title ~ "\n"; - } - if ("uid" in canned_query && !(canned_query["uid"]).empty) { - got.uid = canned_query["uid"].split("%20").join(" "); - got.search_text_area ~= "uid: " ~ got.uid ~ "\n"; - } - if ("fn" in canned_query && !(canned_query["fn"]).empty) { - got.fn = canned_query["fn"].split("%20").join(" "); - got.search_text_area ~= "fn: " ~ got.fn ~ "\n"; - } - if ("kw" in canned_query && !(canned_query["kw"]).empty) { - got.keywords = canned_query["kw"].split("%20").join(" "); - got.search_text_area ~= "keywords: " ~ got.keywords ~ "\n"; - } - if ("tr" in canned_query && !(canned_query["tr"]).empty) { - got.topic_register = canned_query["tr"].split("%20").join(" "); - got.search_text_area ~= "topic_register: " ~ got.topic_register ~ "\n"; - } - if ("su" in canned_query && !(canned_query["su"]).empty) { - got.subject = canned_query["su"].split("%20").join(" "); - got.search_text_area ~= "subject: " ~ got.subject ~ "\n"; - } - if ("de" in canned_query && !(canned_query["de"]).empty) { - got.description = canned_query["de"].split("%20").join(" "); - got.search_text_area ~= "description: " ~ got.description ~ "\n"; - } - if ("pb" in canned_query && !(canned_query["pb"]).empty) { - got.publisher = canned_query["pb"].split("%20").join(" "); - got.search_text_area ~= "publisher: " ~ got.publisher ~ "\n"; - } - if ("ed" in canned_query && !(canned_query["ed"]).empty) { - got.editor = canned_query["ed"].split("%20").join(" "); - got.search_text_area ~= "editor: " ~ got.editor ~ "\n"; - } - if ("ct" in canned_query && !(canned_query["ct"]).empty) { - got.contributor = canned_query["ct"].split("%20").join(" "); - got.search_text_area ~= "contributor: " ~ got.contributor ~ "\n"; - } - if ("dt" in canned_query && !(canned_query["dt"]).empty) { - got.date = canned_query["dt"].split("%20").join(" "); - got.search_text_area ~= "date: " ~ got.date ~ "\n"; - } - if ("rt" in canned_query && !(canned_query["rt"]).empty) { - got.results_type = canned_query["rt"].split("%20").join(" "); - // got.search_text_area ~= "results_type: " ~ got.results_type ~ "\n"; - } - if ("fmt" in canned_query && !(canned_query["fmt"]).empty) { - got.format = canned_query["fmt"].split("%20").join(" "); - got.search_text_area ~= "format: " ~ got.format ~ "\n"; - } - if ("id" in canned_query && !(canned_query["id"]).empty) { - got.identifier = canned_query["id"].split("%20").join(" "); - got.search_text_area ~= "identifier: " ~ got.identifier ~ "\n"; - } - if ("src" in canned_query && !(canned_query["src"]).empty) { - got.source = canned_query["src"].split("%20").join(" "); - got.search_text_area ~= "source: " ~ got.source ~ "\n"; - } - if ("lng" in canned_query && !(canned_query["lng"]).empty) { - got.language = canned_query["lng"].split("%20").join(" "); - got.search_text_area ~= "language: " ~ got.language ~ "\n"; - } - if ("rl" in canned_query && !(canned_query["rl"]).empty) { - got.relation = canned_query["rl"].split("%20").join(" "); - got.search_text_area ~= "relation: " ~ got.relation ~ "\n"; - } - if ("cv" in canned_query && !(canned_query["cv"]).empty) { - got.coverage = canned_query["cv"].split("%20").join(" "); - got.search_text_area ~= "coverage: " ~ got.coverage ~ "\n"; - } - if ("rgt" in canned_query && !(canned_query["rgt"]).empty) { - got.rights = canned_query["rgt"].split("%20").join(" "); - got.search_text_area ~= "rights: " ~ got.rights ~ "\n"; - } - if ("cmt" in canned_query && !(canned_query["cmt"]).empty) { - got.comment = canned_query["cmt"].split("%20").join(" "); - got.search_text_area ~= "comment: " ~ got.comment ~ "\n"; - } - // if ("abstract" in canned_query && !(canned_query["abstract"]).empty) { - // got.abstract = canned_query["abstract"]; - // } - if ("bfn" in canned_query && !(canned_query["bfn"]).empty) { // search_field - got.src_filename_base = canned_query["bfn"].split("%20").join(" "); - got.search_text_area ~= "src_filename_base: " ~ got.src_filename_base ~ "\n"; - } - if ("sml" in canned_query && !(canned_query["sml"]).empty) { - got.sql_match_limit = canned_query["sml"].split("%20").join(" "); - // got.search_text_area ~= "sql_match_limit: " ~ got.sql_match_limit ~ "\n"; - } - // cgi.write("f.search_text_area: " ~ got.search_text_area ~ "
"); - } - return got; - } - auto tf = text_fields; // - struct SQL_select { - string the_body = ""; - string the_range = ""; - } - auto sql_select = SQL_select(); - string base ; // = ""; - string tip ; // = ""; - string search_note ; // = ""; - uint sql_match_offset_count = 0; - string previous_next () { - static struct Rgx { - static track_offset = ctRegex!(`(?P[&]smo=)(?P[0-9]+)`, "m"); - } - auto rgx = Rgx(); - string _previous_next = ""; - int _current_offset_value = 0; - string _set_offset_next = ""; - string _set_offset_previous = ""; - string _url = ""; - string _url_previous = ""; - string _url_next = ""; - string arrow_previous = ""; - string arrow_next = ""; - if (environment.get("REQUEST_METHOD", "POST") == "POST") { - _url = conf.http_request_type ~ "://" ~ conf.http_host ~ conf.cgi_script ~ "?" ~ tf.canned_query; - } else if (environment.get("REQUEST_METHOD", "POST") == "GET") { - _url = conf.http_request_type ~ "://" ~ conf.http_host ~ conf.cgi_script ~ "?" ~ environment.get("QUERY_STRING", ""); - } - if (auto m = _url.matchFirst(rgx.track_offset)) { - _current_offset_value = m.captures["offset_val"].to!int; - _set_offset_next = m.captures["offset_key"] ~ ((m.captures["offset_val"]).to!int + cv.sql_match_limit.to!int).to!string; - _url_next = _url.replace(rgx.track_offset, _set_offset_next); - if (_current_offset_value < cv.sql_match_limit.to!int) { - _url_previous = ""; - } else { - _url_previous = ""; - _set_offset_previous = m.captures["offset_key"] ~ ((m.captures["offset_val"]).to!int - cv.sql_match_limit.to!int).to!string; - _url_previous = _url.replace(rgx.track_offset, _set_offset_previous); - } - } else {// _current_offset_value = 0; - _url_next = _url ~= "&smo=" ~ cv.sql_match_limit.to!string; - } - if (_url_previous.empty) { - arrow_previous = ""; - } else { - arrow_previous = - "" - ~ "" - ~ "<< prev" - ~ " || "; - } - arrow_next = - "" - ~ "" - ~ "next >>" - ~ ""; - _previous_next = "
" ~ arrow_previous ~ arrow_next; - return _previous_next; - } - { - header = format(q"┃ - - - - - - SiSU spine search form (sample) - - - - - - - -┃", - conf.http_host, - ); - } - { - table = format(q"┃ - - - -
- - -
-
- SiSU - -
- git - -
-
- -
- ┃"); - } - { - string post_value(string field_name, string type="box", string set="on") { - string val = ""; - switch (type) { - case "field": - val = ((field_name in cgi.post && !(cgi.post[field_name]).empty) - ? cgi.post[field_name] - : (field_name in cgi.get) - ? cgi.get[field_name] - : ""); - val = tf.search_text_area; - break; - case "box": // generic for checkbox or radio; checkbox set == "on" radio set == "name set" - val = ((field_name in cgi.post && !(cgi.post[field_name]).empty) - ? (cgi.post[field_name] == set ? "checked" : "off") - : (field_name in cgi.get) - ? (cgi.get[field_name] == set ? "checked" : "off") - : "off"); - break; - case "radio": // used generic bo - val = ((field_name in cgi.post && !(cgi.post[field_name]).empty) - ? (cgi.post[field_name] == set ? "checked" : "off") - : (field_name in cgi.get) - ? (cgi.get[field_name] == set ? "checked" : "off") - : "checked"); - break; - case "checkbox": // used generic bo - val = ((field_name in cgi.post && !(cgi.post[field_name]).empty) - ? (cgi.post[field_name] == set ? "checked" : "off") - : (field_name in cgi.get) - ? (cgi.get[field_name] == set ? "checked" : "off") - : "checked"); - break; - default: - } - return val; - } - string the_can(string fv) { - string show_the_can = post_value("url"); - string _the_can = ""; - if (show_the_can == "checked") { - tf = text_fields; - string method_get_url = conf.http_request_type ~ "://" ~ conf.http_host ~ conf.cgi_script ~ "?" ~ environment.get("QUERY_STRING", ""); - string method_post_url_construct = conf.http_request_type ~ "://" ~ conf.http_host ~ conf.cgi_script ~ "?" ~ tf.canned_query; - // assert(method_get_url == environment.get("HTTP_REFERER", conf.http_request_type ~ "://" ~ conf.http_host ~ conf.cgi_script ~ "?" ~ conf.query_string)); - if (conf.request_method == "POST") { - _the_can = - "" - ~ "POST: " - ~ "" - ~ method_post_url_construct - ~ "" - ~ "
"; - } else if (conf.request_method == "GET") { - _the_can = - "" - ~ "GET:  " - ~ "" - ~ method_get_url - ~ ""; - } - conf.http_url = conf.http_request_type ~ "://" ~ conf.http_host ~ conf.cgi_script ~ tf.canned_query; - } - return _the_can; - } - string provide_tip() { - string searched_tip = post_value("se"); - string tip = ""; - if (searched_tip == "checked") { - string search_field = post_value("sf", "field"); - tf = text_fields; - tip = format(q"┃ - -database: %s; selected view: index -search string: %s %s %s %s %s %s
-%s %s %s %s %s %s -
-┃", - cv.db_selected, - (tf.text.empty ? "" : "\"text: " ~ tf.text ~ "; "), - (tf.title.empty ? "" : "\"title: " ~ tf.title ~ "; "), - (tf.author.empty ? "" : "\"author: " ~ tf.author ~ "; "), - (tf.date.empty ? "" : "\"date " ~ tf.date ~ "; "), - (tf.uid.empty ? "" : "\"uid: " ~ tf.uid ~ "; "), - (tf.fn.empty ? "" : "\"fn: " ~ tf.fn ~ "; "), - (tf.text.empty ? "" : "text: " ~ tf.text ~ "
"), - (tf.title.empty ? "" : "title: " ~ tf.title ~ "
"), - (tf.author.empty ? "" : "author: " ~ tf.author ~ "
"), - (tf.date.empty ? "" : "date: " ~ tf.date ~ "
"), - (tf.uid.empty ? "" : "\"uid: " ~ tf.uid ~ "; "), - (tf.fn.empty ? "" : "\"fn: " ~ tf.fn ~ "; "), - ); - } - return tip; - } - form = format(q"┃ -
- - -
- - - %s - %s - %s -
- - - to search: select which database to search (drop-down menu below); enter your search query (in the form above); and click on the search button (below) -
- - - index - text / grep; - match limit: - 1,000 - 2,500 -
- echo query - result stats - search url - searched - available fields - sql statement - -
-
- - -
-┃", - "spine-search", - (post_value("ec") == "checked") ? post_value("sf", "field") : "", - provide_tip, - search_note, - the_can(post_value("sf", "field")), - cv.db_selected, - cv.db_selected, - post_value("rt", "box", "idx"), - post_value("rt", "box", "txt"), - post_value("sml", "box", "1000"), - post_value("sml", "box", "2500"), - post_value("ec"), - post_value("sts"), - post_value("url"), - post_value("se"), - post_value("tip"), - post_value("sql"), - ); - { - string set_value(string field_name, string default_val) { - string val; - if (field_name in cgi.post) { - val = cgi.post[field_name]; - } else if (field_name in cgi.get) { - val = cgi.get[field_name]; - } else { val = default_val; } - return val; - } - bool set_bool(string field_name) { - bool val; - if (field_name in cgi.post - && cgi.post[field_name] == "on") { - val = true; - } else if (field_name in cgi.get - && cgi.get[field_name] == "on") { - val = true; - } else { val = false; } - return val; - } - cv.db_selected = set_value("selected_db", "spine.search.db"); // selected_db == db - cv.sql_match_limit = set_value("sml", "1000"); - cv.sql_match_offset = set_value("smo", "0"); - cv.search_text = set_value("sf", "test"); // remove test - cv.results_type = set_value("rt", "idx"); - cv.checked_echo = set_bool("ec"); - cv.checked_stats = set_bool("sts"); - cv.checked_url = set_bool("url"); - cv.checked_searched = set_bool("se"); - cv.checked_tip = set_bool("tip"); - cv.checked_sql = set_bool("sql"); - tf = text_fields; - } - } - { - cgi.write(header); - cgi.write(table); - cgi.write(form); - // cgi.write(previous_next); - { // debug environment - // foreach (k, d; environment.toAA) { - // cgi.write(k ~ ": " ~ d ~ "
"); - // } - } - { // debug cgi info - // cgi.write("db_selected: " ~ cv.db_selected ~ "
\n"); - // cgi.write("search_text: " ~ cv.search_text ~ "
\n"); - // cgi.write("sql_match_limit: " ~ cv.sql_match_limit ~ ";\n"); - // cgi.write("sql_match_offset: " ~ cv.sql_match_offset ~ ";\n"); - // cgi.write("results_type: " ~ cv.results_type ~ "
\n"); - // cgi.write("cv.checked_echo: " ~ (cv.checked_echo ? "checked" : "off") ~ "; \n"); - // cgi.write("cv.checked_stats: " ~ (cv.checked_stats ? "checked" : "off") ~ "; \n"); - // cgi.write("cv.checked_url: " ~ (cv.checked_url ? "checked" : "off") ~ "; \n"); - // cgi.write("cv.checked_searched: " ~ (cv.checked_searched ? "checked" : "off") ~ ";
\n"); - // cgi.write("cv.checked_tip: " ~ (cv.checked_tip ? "checked" : "off") ~ "; \n"); - // cgi.write("cv.checked_sql: " ~ (cv.checked_sql ? "checked" : "off") ~ "
\n"); - } - } - auto db = Database(conf.db_path ~ cv.db_selected); - { - uint sql_match_offset_counter(T)(T cv) { - sql_match_offset_count += cv.sql_match_limit.to!uint; - return sql_match_offset_count; - } - void sql_search_query() { - string select_field_like(string db_field, string search_field) { - string where_ = ""; - if (!(search_field.empty)) { - string _sf = search_field.strip.split("%20").join(" "); - if (_sf.match(r" OR ")) { - _sf = _sf.split(" OR ").join("%' OR " ~ db_field ~ " LIKE '%"); - } - if (_sf.match(r" AND ")) { - _sf = _sf.split(" AND ").join("%' AND " ~ db_field ~ " LIKE '%"); - } - _sf = "( " ~ db_field ~ " LIKE\n '%" ~ _sf ~ "%' )"; - where_ ~= format(q"┃ - %s -┃", - _sf - ); - } - return where_; - } - string[] _fields; - _fields ~= select_field_like("doc_objects.clean", tf.text); - _fields ~= select_field_like("metadata_and_text.title", tf.title); - _fields ~= select_field_like("metadata_and_text.creator_author", tf.author); - _fields ~= select_field_like("metadata_and_text.uid", tf.uid); - _fields ~= select_field_like("metadata_and_text.src_filename_base", tf.fn); - _fields ~= select_field_like("metadata_and_text.src_filename_base", tf.src_filename_base); - _fields ~= select_field_like("metadata_and_text.language_document_char", tf.language); - _fields ~= select_field_like("metadata_and_text.date_published", tf.date); - _fields ~= select_field_like("metadata_and_text.classify_keywords", tf.keywords); - _fields ~= select_field_like("metadata_and_text.classify_topic_register", tf.topic_register); - string[] fields; - foreach (f; _fields) { - if (!(f.empty)) { fields ~= f; } - } - string fields_str = ""; - fields_str ~= fields.join(" AND "); - sql_select.the_body ~= format(q"┃ -SELECT - metadata_and_text.uid, - metadata_and_text.title, - metadata_and_text.creator_author_last_first, - metadata_and_text.creator_author, - metadata_and_text.src_filename_base, - metadata_and_text.language_document_char, - metadata_and_text.date_published, - metadata_and_text.classify_keywords, - metadata_and_text.classify_topic_register, - doc_objects.body, - doc_objects.seg_name, - doc_objects.ocn, - metadata_and_text.uid -FROM - doc_objects, - metadata_and_text -WHERE ( - %s - ) -AND - doc_objects.uid_metadata_and_text = metadata_and_text.uid -ORDER BY - metadata_and_text.creator_author_last_first, - metadata_and_text.date_published DESC, - metadata_and_text.title, - metadata_and_text.language_document_char, - metadata_and_text.src_filename_base, - doc_objects.ocn -LIMIT %s OFFSET %s -;┃", - fields_str, - cv.sql_match_limit, - cv.sql_match_offset, - ); - (cv.checked_sql) - ? cgi.write(previous_next ~ "
" ~ sql_select.the_body.split("\n ").join(" ").split("\n").join("
") ~ "
\n") - : ""; - cgi.write(previous_next); - auto select_query_results = db.execute(sql_select.the_body).cached; - string _old_uid = ""; - if (!select_query_results.empty) { - foreach (row; select_query_results) { - if (row["uid"].as!string != _old_uid) { - _old_uid = row["uid"].as!string; - auto m = (row["date_published"].as!string).match(regex(r"^([0-9]{4})")); // breaks if row missing or no match? - cgi.write( - "
\"" - ~ row["title"].as!string ~ "\"" - ~ " (" - ~ m.hit - ~ ") " - ~ "[" - ~ row["language_document_char"].as!string - ~ "] " - ~ row["creator_author_last_first"].as!string - ~ ":
\n" - ); - } - if (cv.results_type == "txt") { - cgi.write( - "
" - ~ row["ocn"].as!string - ~ "" - ~ "
" - ~ row["body"].as!string - ); - } else { - cgi.write( - "" - ~ row["ocn"].as!string - ~ ", " - ); - } - } - cgi.write( previous_next); - } else { // offset_not_beyond_limit = false; - cgi.write("select_query_results empty

\n"); - } - } - sql_search_query; - } - { - db.close; - } - { - string tail = format(q"┃ - -┃"); - cgi.write(tail); - } -} -mixin GenericMain!cgi_function_intro; diff --git a/misc/util/d/tools/markup_conversion/README b/misc/util/d/tools/markup_conversion/README deleted file mode 100644 index 8b13789..0000000 --- a/misc/util/d/tools/markup_conversion/README +++ /dev/null @@ -1 +0,0 @@ - diff --git a/misc/util/d/tools/markup_conversion/endnotes_inline_from_binary.d b/misc/util/d/tools/markup_conversion/endnotes_inline_from_binary.d deleted file mode 100755 index b084052..0000000 --- a/misc/util/d/tools/markup_conversion/endnotes_inline_from_binary.d +++ /dev/null @@ -1,123 +0,0 @@ -#!/usr/bin/env rdmd -/+ - - read in file .sst .ssi .ssm - - loop twice - - first - - check for and skip code blocks - - use unique code marker for endnote markers in text and give an endnote - number ★1, increment - - extract all endnotes in array - - second - - check that the footnote marker number count matches the number of notes - in the array - - if they match either: - - substitute each endnote marker with the array footnote[number-1] - - substitute each endnote marker with footnote - as inlined footnote markup (footnote number not needed) - - if they do not match exit - - check whether changes have been made - - if so write file with inline footnotes in sub-directory converted_output_/ - using the same name as the original file - - else, exit -+/ -import std.stdio; -import std.file; -import std.array : split; -import std.exception; -import core.stdc.errno; -import std.regex; -import std.format; -import std.conv; -void main(string[] args) { - static comment = ctRegex!(`^%+ `); - static block_tic_code_open = ctRegex!("^`{3} code(?:[.](?P[a-z][0-9a-z#+_]+))?(?:[(](?P[ a-zA-Z0-9;:,]*)[)])?"); - static block_tic_close = ctRegex!("^(`{3})$","m"); - static block_curly_code_open = ctRegex!(`^(?:code(?:[.](?P[a-z][0-9a-z_]+))?(?:[(](?P[ a-zA-Z0-9;:,]*)[)])?[{][ ]*$)`); - static block_curly_code_close = ctRegex!(`^([}]code)`); - auto rgx_endnote_ref = ctRegex!(`([~]\^)(?P[)\]]? |$)`, "gm"); - auto rgx_endnote = ctRegex!(`^\^~\s+(.+|\n)`, "gm"); - foreach(arg; args[1..$]) { - if ( - !(arg.match(regex(r"--\w+"))) - && arg.match(regex(r"\w+?\.ss[itm]")) - ) { - writeln(arg); - string filename = arg; - try { - string[] contents, endnotes, endnote_refs; - string text = filename.readText; - string[] paragraphs = text.split("\n\n"); - int endnote_ref_count = 0; - int code_block_status = 0; - enum codeBlock { off, curly, tic, } - foreach (paragraph; paragraphs) { /+ loop to gather binary endnotes +/ - if (code_block_status == codeBlock.off - && paragraph.match(rgx_endnote) - ) { - endnotes ~= replaceAll!(m => m[1]) - (paragraph, rgx_endnote); - } else { - if ((code_block_status == codeBlock.curly - && paragraph.matchFirst(block_curly_code_close)) - || ((code_block_status == codeBlock.tic - && paragraph.matchFirst(block_tic_close)) - ) { - code_block_status = codeBlock.off; - } else if ( type["curly_code"] == 1 || type["tic_code"] == 1) { - // skip, prevent search for endnotes - } else if (paragraph.matchFirst(block_curly_code_open)) { - code_block_status = codeBlock.curly; - } else if (paragraph.matchFirst(block_tic_code_open)) { - code_block_status = codeBlock.tic; - } else if (auto m = paragraph.matchAll(rgx_endnote_ref)) { - foreach (n; m) { - endnote_ref_count++; // endnote_refs ~= (n.captures[1]); - } - } - contents ~= paragraph; - } - } - if (endnotes.length == endnote_ref_count) { - import std.outbuffer; - writeln("endnote ref count: ", endnote_ref_count); - writeln("number of binary endnotes: ", endnotes.length); - int endnote_count = -1; - auto buffer = new OutBuffer(); - foreach (content; contents) { /+ loop to inline endnotes +/ - content = replaceAll!(m => "~{ " ~ endnotes[++endnote_count] ~ " }~" ~ m["tail"] ) - (content, rgx_endnote_ref); - buffer.write(content ~ "\n\n"); - } - if (buffer) { - try { - string dir_out = "converted_output_"; - string path_and_file_out = dir_out ~ "/" ~ filename; - dir_out.mkdirRecurse; - auto f = File(path_and_file_out, "w"); - f.write(buffer); - writeln("wrote: ", path_and_file_out); - } catch (FileException ex) { - writeln("did not write file"); - // Handle errors - } - } - } else { - writeln("ERROR binary endnote mismatch, check markup,\nmisatch in the number of endnotes & endnote references!"); - writeln(" number of endnotes: ", endnotes.length); - writeln(" number of endnote refs: ", endnote_ref_count); // endnote_refs.length, - } - // assert(endnotes.length == endnote_ref_count); - } catch (ErrnoException ex) { - switch(ex.errno) { - case EPERM: - case EACCES: // Permission denied - break; - case ENOENT: // File does not exist - break; - default: // Handle other errors - break; - } - } - } - } -} diff --git a/misc/util/d/tools/markup_conversion/markup_changes.d b/misc/util/d/tools/markup_conversion/markup_changes.d deleted file mode 100644 index 4274f78..0000000 --- a/misc/util/d/tools/markup_conversion/markup_changes.d +++ /dev/null @@ -1,136 +0,0 @@ -#!/usr/bin/env rdmd -/+ - - read in file .sst .ssi .ssm - - loop twice - - first - - check for and skip code blocks - - use unique code marker for endnote markers in text and give an endnote - number ★1, increment - - extract all endnotes in array - - second - - check that the footnote marker number count matches the number of notes - in the array - - if they match either: - - substitute each endnote marker with the array footnote[number-1] - - substitute each endnote marker with footnote - as inlined footnote markup (footnote number not needed) - - if they do not match exit - - check whether changes have been made - - if so write file with inline footnotes in sub-directory converted_output_/ - using the same name as the original file - - else, exit -+/ -import std.stdio; -import std.file; -import std.array : split; -import std.exception; -import core.stdc.errno; -import std.regex; -import std.format; -import std.conv; -void main(string[] args) { - static comment = ctRegex!(`^%+ `); - static block_tic_code_open = ctRegex!("^`{3} code(?:[.](?P[a-z][0-9a-z#+_]+))?(?:[(](?P[ a-zA-Z0-9;:,]*)[)])?"); - static block_tic_close = ctRegex!("^(`{3})$","m"); - static block_curly_code_open = ctRegex!(`^(?:code(?:[.](?P[a-z][0-9a-z_]+))?(?:[(](?P[ a-zA-Z0-9;:,]*)[)])?[{][ ]*$)`); - static block_curly_code_close = ctRegex!(`^([}]code)`); - auto rgx_endnote_ref = ctRegex!(`([~]\^)(?P[)\]]? |$)`, "gm"); - auto rgx_endnote = ctRegex!(`^\^~\s+(.+|\n)`, "gm"); - foreach(arg; args[1..$]) { - if ( - !(arg.match(regex(r"--\w+"))) - && arg.match(regex(r"\w+?\.ss[itm]")) - ) { - writeln(arg); - string filename = arg; - try { - string[] contents, endnotes, endnote_refs; - string text = filename.readText; - string[] paragraphs = text.split("\n\n"); - int endnote_ref_count = 0; - int[string] type = [ - "curly_code" : 0, - "tic_code" : 0, - ]; - foreach (paragraph; paragraphs) { /+ loop to gather binary endnotes +/ - if ( !( type["curly_code"] == 1 || type["tic_code"] == 1) - && paragraph.match(rgx_endnote) - ) { - endnotes ~= replaceAll!(m => m[1]) - (paragraph, rgx_endnote); - } else { - if ( type["curly_code"] == 1 || type["tic_code"] == 1 - || paragraph.matchFirst(block_curly_code_open) - || paragraph.matchFirst(block_tic_code_open) - ) { /+ code blocks identified, no munging +/ - if ( type["curly_code"] == 1 - && paragraph.matchFirst(block_curly_code_close) - ) { - type["curly_code"] = 0; - } else if (type["tic_code"] == 1 - && paragraph.matchFirst(block_tic_close) - ) { - type["tic_code"] = 0; - } else if (paragraph.matchFirst(block_curly_code_open)) { - type["curly_code"] = 1; - } else if (paragraph.matchFirst(block_tic_code_open)) { - type["tic_code"] = 1; - } - contents ~= paragraph; - } else { /+ regular content, not a code block +/ - if (auto m = paragraph.matchAll(rgx_endnote_ref)) { - foreach (n; m) { - endnote_ref_count++; // endnote_refs ~= (n.captures[1]); - } - } - paragraph = replaceAll!(m => " \\\\ " ) - (paragraph, regex(r"\s*<(?:/\s*|:)?br>\s*")); // (paragraph, regex(r"(
)")); - contents ~= paragraph; - } - } - } - { - import std.outbuffer; - auto buffer = new OutBuffer(); - if (endnotes.length == endnote_ref_count) { - // writeln("endnote ref count: ", endnote_ref_count); - // writeln("number of binary endnotes: ", endnotes.length); - int endnote_count = -1; - foreach (content; contents) { /+ loop to inline endnotes +/ - content = replaceAll!(m => "~{ " ~ endnotes[++endnote_count] ~ " }~" ~ m["tail"] ) - (content, rgx_endnote_ref); // endnote_ref cannot occur in a code block or else fail - buffer.write(content ~ "\n\n"); - } - if (buffer) { - try { - string dir_out = "converted_output_"; - string path_and_file_out = dir_out ~ "/" ~ filename; - dir_out.mkdirRecurse; - auto f = File(path_and_file_out, "w"); - f.write(buffer); - writeln("wrote: ", path_and_file_out); - } catch (FileException ex) { - writeln("did not write file"); - // Handle errors - } - } - } else { - foreach (content; contents) { /+ loop to inline endnotes +/ - buffer.write(content ~ "\n\n"); - } - } - } - } catch (ErrnoException ex) { - switch(ex.errno) { - case EPERM: - case EACCES: // Permission denied - break; - case ENOENT: // File does not exist - break; - default: // Handle other errors - break; - } - } - } - } -} diff --git a/misc/util/d/tools/markup_conversion/markup_changes_header_and_content.d b/misc/util/d/tools/markup_conversion/markup_changes_header_and_content.d deleted file mode 100755 index 86792ff..0000000 --- a/misc/util/d/tools/markup_conversion/markup_changes_header_and_content.d +++ /dev/null @@ -1,244 +0,0 @@ -#!/usr/bin/env rdmd -/+ - - read in file .sst .ssi .ssm - - loop twice - - first - - check for and skip code blocks - - use unique code marker for endnote markers in text and give an endnote - number ★1, increment - - extract all endnotes in array - - second - - check that the footnote marker number count matches the number of notes - in the array - - if they match either: - - substitute each endnote marker with the array footnote[number-1] - - substitute each endnote marker with footnote - as inlined footnote markup (footnote number not needed) - - if they do not match exit - - check whether changes have been made - - if so write file with inline footnotes in sub-directory converted_output_/ - using the same name as the original file - - else, exit -+/ -import std.stdio; -import std.file; -import std.array : split; -import std.exception; -// import std.range; -import core.stdc.errno; -import std.regex; -import std.format; -import std.conv; -void main(string[] args) { - static heading_a = ctRegex!(`^:?[A][~] `, "m"); - static comment = ctRegex!(`^%+ `); - static block_tic_code_open = ctRegex!("^`{3} code(?:[.](?P[a-z][0-9a-z#+_]+))?(?:[(](?P[ a-zA-Z0-9;:,]*)[)])?"); - static block_tic_close = ctRegex!("^(`{3})$","m"); - static block_curly_code_open = ctRegex!(`^(?:code(?:[.](?P[a-z][0-9a-z_]+))?(?:[(](?P[ a-zA-Z0-9;:,]*)[)])?[{][ ]*$)`); - static block_curly_code_close = ctRegex!(`^([}]code)`); - auto rgx_endnote_ref = ctRegex!(`([~]\^)(?P[)\]]? |$)`, "gm"); - auto rgx_endnote = ctRegex!(`^\^~\s+(.+|\n)`, "gm"); - char[][] header0Content1(in string src_text) { // cast(char[]) - /+ split string on _first_ match of "^:?A~\s" into [header, content] array/tuple +/ - char[][] header_and_content; - auto m = (cast(char[]) src_text).matchFirst(heading_a); - header_and_content ~= m.pre; - header_and_content ~= m.hit ~ m.post; - assert(header_and_content.length == 2, - "document markup is broken, header body split == " - ~ header_and_content.length.to!string - ~ "; (header / body array split should == 2 (split is on level A~))" - ); - return header_and_content; - } - foreach(arg; args[1..$]) { - if ( - !(arg.match(regex(r"--\w+"))) - && arg.match(regex(r"\w+?\.ss[itm]")) - ) { - writeln(arg); - string filename = arg; - try { - string[] munged_header, munged_contents, munged_endnotes, endnote_refs; - string text = filename.readText; - char[][] hc = header0Content1(text); - char[] src_header = hc[0]; - string[] headers = src_header.to!string.split("\n\n"); - char[] src_txt = hc[1]; - string[] paragraphs = src_txt.to!string.split("\n\n"); - int endnote_ref_count = 0; - int[string] type = [ - "curly_code" : 0, - "tic_code" : 0, - ]; - string _tmp_header; - foreach (h_; headers) { /+ loop to inline endnotes +/ - _tmp_header = ""; - if (h_.match(regex(r"^[@\[]?title[:\]]?"))) { // title - if (auto m = h_.match(regex(r"^@title:(?:\s+(?P.+)|$)"))) { // sisu bespoke markup - if (m.captures["c"].length == 0) { - _tmp_header ~= "title:"; - } else { - _tmp_header ~= "title:\n main: " ~ "\"" ~ m.captures["c"] ~ "\""; - } - } else if (auto m = h_.match(regex(r"^title\s*=\s*(?P.+)"))) { // toml? - if (m.captures["c"].length == 0) { - _tmp_header ~= "title:"; - } else { - _tmp_header ~= "title:\n main: " ~ "\"" ~ m.captures["c"] ~ "\""; - } - } else if (auto m = h_.match(regex(r"^\[title\]"))) { // toml markup - _tmp_header ~= "title:"; - } else if (auto m = h_.match(regex(r"^title(?:\s+(?P.+)|\s+\\$)"))) { // sdlang markup - if (m.captures["c"].length == 0) { - _tmp_header ~= "title:"; - } else { - _tmp_header ~= "title:\n main: " ~ "\"" ~ m.captures["c"] ~ "\""; - } - } - if (h_.match(regex(r"^\s*[:]?(?:main)[:= ]?", "m"))) { - if (auto m = h_.match(regex(r"^\s+(?P:main):(?:\s+(?P.+)|$)", "m"))) { // sisu bespoke markup - _tmp_header ~= " main: " ~ "\"" ~ m.captures["c"] ~ "\""; - } else if (auto m = h_.match(regex(r"^\s*(?Pmain)\s*=\s*(?P.+)", "m"))) { // toml? - _tmp_header ~= " main: " ~ m.captures["c"]; - } else if (auto m = h_.match(regex(r"^\s+(?Pmain)(?:\s*\s*(?P.+)|$)", "m"))) { // toml markup - _tmp_header ~= " main: " ~ "\"" ~ m.captures["c"] ~ "\""; - } else if (auto m = h_.match(regex(r"^\s+(?Pmain)(?:\s+(?P.+)|\s+\\$)", "m"))) { // sdlang markup - _tmp_header ~= " main: " ~ "\"" ~ m.captures["c"] ~ "\""; - } - } - if (h_.match(regex(r"^\s*[:]?(?:sub(title)?)[:= ]?", "m"))) { - if (auto m = h_.match(regex(r"^\s+:sub(?:title)?:(?:\s+(?P.+)|$)", "m"))) { // sisu bespoke markup - _tmp_header ~= " subtitle: " ~ "\"" ~ m.captures["c"] ~ "\""; - } else if (auto m = h_.match(regex(r"^\s*sub(?:title)?\s*=\s*(?P.+)$", "m"))) { // toml? - _tmp_header ~= " subtitle: " ~ m.captures["c"]; - } else if (auto m = h_.match(regex(r"^\s+(?:title)?(?:\s*\s*(?P.+)|$)", "m"))) { // toml markup - _tmp_header ~= " subtitle: " ~ "\"" ~ m.captures["c"] ~ "\""; - } else if (auto m = h_.match(regex(r"^\s+(?:title)?(?:\s+(?P.+)|\s+\\$)", "m"))) { // sdlang markup - _tmp_header ~= " subtitle: " ~ "\"" ~ m.captures["c"] ~ "\""; - } - } - } - if (h_.match(regex(r"^[@\[]?rights[:\]]?"))) { // rights - if (auto m = h_.match(regex(r"^@rights:[ ]+(?P.+)$"))) { // sisu bespoke markup - _tmp_header ~= "rights: \n copyright: \"" ~ m.captures["c"] ~ "\""; - } else if (auto m = h_.match(regex(r"^@rights:"))) { // sisu bespoke markup - _tmp_header ~= "rights:"; - } else if (auto m = h_.match(regex(r"^\[rights\]", "m"))) { // toml markup - _tmp_header ~= "rights:"; - } else if (auto m = h_.match(regex(r"^rights:"))) { // sdlang markup - _tmp_header ~= "rights:"; - } - if (h_.match(regex(r"^\s*[:]?copyright[:= ]?", "m"))) { - if (auto m = h_.match(regex(r"^\s+:copyright:(?:\s+(?P.+)|$)", "m"))) { // sisu bespoke markup - _tmp_header ~= " copyright: " ~ "\"" ~ m.captures["c"] ~ "\""; - } else if (auto m = h_.match(regex(r"^\s*copyright\s*=\s*(?P.+)", "m"))) { // toml? - _tmp_header ~= " copyright: " ~ m.captures["c"]; - } else if (auto m = h_.match(regex(r"^\s+copyright(?:\s*\s*(?P.+)|$)", "m"))) { // toml markup - _tmp_header ~= " copyright: " ~ "\"" ~ m.captures["c"] ~ "\""; - } else if (auto m = h_.match(regex(r"^\s+copyright(?:\s+(?P.+)|\s+\\$)", "m"))) { // sdlang markup - _tmp_header ~= " copyright: " ~ "\"" ~ m.captures["c"] ~ "\""; - } - } - if (h_.match(regex(r"^\s*[:]?licen[cs]e[:= ]?", "m"))) { - if (auto m = h_.match(regex(r"^\s+:licen[cs]e:(?:\s+(?P.+)|$)", "m"))) { // sisu bespoke markup - _tmp_header ~= " license: " ~ "\"" ~ m.captures["c"] ~ "\""; - } else if (auto m = h_.match(regex(r"^\s*licen[cs]e\s*=\s*(?P.+)$", "m"))) { // toml? - _tmp_header ~= " license: " ~ m.captures["c"]; - } else if (auto m = h_.match(regex(r"^\s+licen[cs]e(?:\s*\s*(?P.+)|$)", "m"))) { // toml markup - _tmp_header ~= " license: " ~ "\"" ~ m.captures["c"] ~ "\""; - } else if (auto m = h_.match(regex(r"^\s+licen[cs]e(?:\s+(?P.+)|\s+\\$)", "m"))) { // sdlang markup - _tmp_header ~= " license: " ~ "\"" ~ m.captures["c"] ~ "\""; - } - } - } - if (_tmp_header.length > 0) { - munged_header ~= _tmp_header; - } else { - munged_header ~= h_; - } - } - writeln(munged_header); - foreach (paragraph; paragraphs) { /+ loop to gather binary endnotes +/ - if ( !( type["curly_code"] == 1 || type["tic_code"] == 1) - && paragraph.match(rgx_endnote) - ) { - munged_endnotes ~= replaceAll!(m => m[1]) - (paragraph, rgx_endnote); - } else { - if ( type["curly_code"] == 1 || type["tic_code"] == 1 - || paragraph.matchFirst(block_curly_code_open) - || paragraph.matchFirst(block_tic_code_open) - ) { /+ code blocks identified, no munging +/ - if ( type["curly_code"] == 1 - && paragraph.matchFirst(block_curly_code_close) - ) { - type["curly_code"] = 0; - } else if (type["tic_code"] == 1 - && paragraph.matchFirst(block_tic_close) - ) { - type["tic_code"] = 0; - } else if (paragraph.matchFirst(block_curly_code_open)) { - type["curly_code"] = 1; - } else if (paragraph.matchFirst(block_tic_code_open)) { - type["tic_code"] = 1; - } - munged_contents ~= paragraph; - } else { /+ regular content, not a code block +/ - if (auto m = paragraph.matchAll(rgx_endnote_ref)) { - foreach (n; m) { - endnote_ref_count++; // endnote_refs ~= (n.captures[1]); - } - } - paragraph = replaceAll!(m => " \\\\ " ) - (paragraph, regex(r"\s*<(?:/\s*|:)?br>\s*")); // (paragraph, regex(r"(
)")); - munged_contents ~= paragraph; - } - } - } - { - import std.outbuffer; - auto buffer = new OutBuffer(); - foreach (header; munged_header) { /+ loop to inline endnotes +/ - buffer.write(header ~ "\n\n"); - } - if (munged_endnotes.length == endnote_ref_count) { - int endnote_count = -1; - foreach (content; munged_contents) { /+ loop to inline endnotes +/ - content = replaceAll!(m => "~{ " ~ munged_endnotes[++endnote_count] ~ " }~" ~ m["tail"] ) - (content, rgx_endnote_ref); // endnote_ref cannot occur in a code block or else fail - buffer.write(content ~ "\n\n"); - } - if (buffer) { - try { - string dir_out = "converted_output_"; - string path_and_file_out = dir_out ~ "/" ~ filename; - dir_out.mkdirRecurse; - auto f = File(path_and_file_out, "w"); - f.write(buffer); - // writeln("wrote: ", path_and_file_out); - } catch (FileException ex) { - writeln("did not write file"); - // Handle errors - } - } - } else { - foreach (content; munged_contents) { /+ loop to inline endnotes +/ - buffer.write(content ~ "\n\n"); - } - } - } - } catch (ErrnoException ex) { - switch(ex.errno) { - case EPERM: - case EACCES: // Permission denied - break; - case ENOENT: // File does not exist - break; - default: // Handle other errors - break; - } - } - } - } -} diff --git a/misc/util/d/tools/markup_conversion/markup_conversion_from_sisu_ruby_to_sisu_spine.d b/misc/util/d/tools/markup_conversion/markup_conversion_from_sisu_ruby_to_sisu_spine.d deleted file mode 100755 index 0ec541d..0000000 --- a/misc/util/d/tools/markup_conversion/markup_conversion_from_sisu_ruby_to_sisu_spine.d +++ /dev/null @@ -1,367 +0,0 @@ -#!/usr/bin/env rdmd -/+ - - read in file .sst .ssi .ssm - - loop twice - - first - - check for and skip code blocks - - use unique code marker for endnote markers in text and give an endnote - number ★1, increment - - extract all endnotes in array - - second - - check that the footnote marker number count matches the number of notes - in the array - - if they match either: - - substitute each endnote marker with the array footnote[number-1] - - substitute each endnote marker with footnote - as inlined footnote markup (footnote number not needed) - - if they do not match exit - - check whether changes have been made - - if so write file with inline footnotes in sub-directory converted_output_/ - using the same name as the original file - - else, exit -+/ -import std.stdio; -import std.file; -import std.array : split, join; -import std.exception; -// import std.range; -import core.stdc.errno; -import std.regex; -import std.format; -import std.conv; -void main(string[] args) { - static heading_a = ctRegex!(`^:?[A][~] `, "m"); - static comment = ctRegex!(`^%+ `); - static block_tic_code_open = ctRegex!("^`{3} code(?:[.](?P[a-z][0-9a-z#+_]+))?(?:[(](?P[ a-zA-Z0-9;:,]*)[)])?"); - static block_tic_close = ctRegex!("^(`{3})$","m"); - static block_curly_code_open = ctRegex!(`^(?:code(?:[.](?P[a-z][0-9a-z_]+))?(?:[(](?P[ a-zA-Z0-9;:,]*)[)])?[{][ ]*$)`); - static block_curly_code_close = ctRegex!(`^([}]code)`); - auto rgx_endnote_ref = ctRegex!(`([~]\^)(?P[)\]]? |$)`, "gm"); - auto rgx_endnote = ctRegex!(`^\^~\s+(.+|\n)`, "gm"); - char[][] header0Content1(in string src_text) { // cast(char[]) - /+ split string on _first_ match of "^:?A~\s" into [header, content] array/tuple +/ - char[][] header_and_content; - auto m = (cast(char[]) src_text).matchFirst(heading_a); - header_and_content ~= m.pre; - header_and_content ~= m.hit ~ m.post; - assert(header_and_content.length == 2, - "document markup is broken, header body split == " - ~ header_and_content.length.to!string - ~ "; (header / body array split should == 2 (split is on level A~))" - ); - return header_and_content; - } - string format_body_string(string s) { - string o; - o = s - .replaceAll(regex("^<(?:/[ ]*)?br>[ ]*"), " \\\\ ") - .replaceAll(regex("[ ]*<(?:/[ ]*)?br>$"), " \\\\") - .replaceAll(regex("[ ]*<(?:/[ ]*)?br>[ ]*"), " \\\\ "); - return o; - } - string format_header_string(string s) { - string o; - o = s - .replaceAll(regex("\""), "\\\"") - .replaceAll(regex("[ ]*<(?:/[ ]*)?br>$"), " \\\\") - .replaceAll(regex("[ ]*<(?:/[ ]*)?br>[ ]*"), " \\\\ "); - return o; - } - string format_main_header(string hm, string hs = "", string c = "") { - string o; - if (c.length == 0) { - o ~= hm ~ ":\n"; - } else { - o ~= hm ~ ":\n" - ~ " " ~ hs ~ ": " - ~ "\"" ~ format_header_string(c) ~ "\"\n"; - } - return o; - } - string format_sub_header(string hs, string c) { - string o; - o ~= " " ~ hs ~ ": " - ~ "\"" ~ format_header_string(c) ~ "\"\n"; - return o; - } - foreach(arg; args[1..$]) { - if ( - !(arg.match(regex(r"--\w+"))) - && arg.match(regex(r"\w+?\.ss[itm]")) - ) { - writeln(arg); - string filename = arg; - try { - string[] munged_header, munged_contents, munged_endnotes, endnote_refs; - char[][] hc; - char[] src_header; - string[] headers; - char[] src_txt; - string[] paragraphs; - enum codeBlock { off, curly, tic, } - string _tmp_header; - int endnote_ref_count = 0; - int code_block_status = codeBlock.off; - string text = filename.readText; - if (arg.match(regex(r"\w+?\.ss[tm]"))) { - hc = header0Content1(text); - src_header = hc[0]; - headers = src_header.to!string.split("\n\n"); - src_txt = hc[1]; - paragraphs = src_txt.to!string.split("\n\n"); - } else if (arg.match(regex(r"\w+?\.ssi"))) { - headers = []; - paragraphs = text.split("\n\n"); - } - if (headers.length > 0) { - headers[0] = headers[0].replaceFirst(regex(r"^%\s+SiSU.+", "i"), "# SiSU 8.0 spine (auto-conversion)"); - foreach (h_; headers) { - _tmp_header = ""; - if (auto m = h_.match(regex(r"^%\s*", "m"))) { - h_ = h_.replaceAll(regex(r"^%\s*", "m"), "# ") ~ "\n"; - } - if (h_.match(regex(r"^@title:|@subtitle"))) { - if (auto m = h_.match(regex(r"^@(?Ptitle):(?:[ ]+(?P.+)|\n)"))) { - _tmp_header ~= format_main_header(m.captures["h"], "main", m.captures["c"]); - } - if (auto m = h_.match(regex(r"^@(?Psubtitle):(?:[ ]+(?P.+)|$)"))) { - if (m.captures["c"].length == 0) { - } else { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - } - if (auto m = h_.match(regex(r"^\s+:(?Pmain):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:sub(?:title)?:(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header("subtitle", m.captures["c"]); - } - } else if (h_.match(regex(r"^@creator:|@author:"))) { - if (auto m = h_.match(regex(r"^(?:@creator:|@author:)(?:[ ]+(?P.+)|\n)"))) { - _tmp_header ~= format_main_header("creator", "author", m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pauthor):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - } else if (h_.match(regex(r"^@rights:"))) { - if (auto m = h_.match(regex(r"^@(?Prights):(?:[ ]+(?P.+)|\n)"))) { - _tmp_header ~= format_main_header(m.captures["h"], "copyright", m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pcopyright):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:licen[cs]e:(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header("license", m.captures["c"]); - } - } else if (h_.match(regex(r"^@date:|@date\."))) { - if (auto m = h_.match(regex(r"^@(?Pdate):(?:[ ]+(?P.+)|\n)"))) { - _tmp_header ~= format_main_header(m.captures["h"], "published", m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Ppublished):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pavailable):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pmodified):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pcreated):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pissued):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pvalid):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^@date\.(?Pavailable):[ ]+(?P.+)$"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^@date\.(?Pmodified):[ ]+(?P.+)$"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^@date\.(?Pcreated):[ ]+(?P.+)$"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^@date\.(?Pissued):[ ]+(?P.+)$"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^@date\.(?Pvalid):[ ]+(?P.+)$"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - } else if (h_.match(regex(r"^@classify:"))) { - if (auto m = h_.match(regex(r"^@classify:"))) { - _tmp_header ~= "classify:\n"; - } - if (auto m = h_.match(regex(r"^\s+:(?Ptopic_register):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:type:(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= "# type: " ~ "\"" ~ m.captures["c"] ~ "\"\n"; - } - } else if (h_.match(regex(r"^(?:@identifier:|@identify:)"))) { - if (auto m = h_.match(regex(r"^(?:@identifier:|@idenfify)"))) { - _tmp_header ~= "identify:\n"; - } - if (auto m = h_.match(regex(r"^\s+:(?Poclc):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pisbn):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pdewey):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - } else if (h_.match(regex(r"^@publisher:"))) { - if (auto m = h_.match(regex(r"^@publisher:[ ]+(?P.+)$"))) { - _tmp_header ~= "publisher: " ~ "\"" ~ m.captures["c"] ~ "\"\n"; - } - } else if (h_.match(regex(r"^@make:"))) { - // writeln(h_); - if (auto m = h_.match(regex(r"^@make:"))) { - _tmp_header ~= "make:\n"; - } - if (auto m = h_.match(regex(r"^\s+:(?Pbreaks):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pnum_top):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pheadings):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pitalics):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pbold):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pemphasis):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Psubstitute):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Ptexpdf_font):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Phome_button_text):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Phome_button_image):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pcover_image):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - if (auto m = h_.match(regex(r"^\s+:(?Pfooter):(?:[ ]+(?P.+)|$)", "m"))) { - _tmp_header ~= format_sub_header(m.captures["h"], m.captures["c"]); - } - // writeln(_tmp_header); - } else if (h_.match(regex(r"^@\w+:"))) { - _tmp_header ~= "# " ~ h_.split("\n").join("\n# ") ~ "\n"; - } else if (h_.match(regex(r"^\s+:\w+:", "m"))) { - if (auto m = h_.match(regex(r"^(?P\s+:\w+:.*)"))) { - _tmp_header ~= "# " ~ m.captures["g"] ~ "\n"; - } - } - if (h_.match(regex(r"^#", "m"))) { - if (auto m = h_.match(regex(r"^(?P#.*)", "m"))) { - _tmp_header ~= m.captures["g"] ~ "\n"; - } - } - if (_tmp_header.length > 0) { - munged_header ~= _tmp_header.split("\n\n"); - } else if (h_.length > 0) { - writeln("munging required: ", h_); - h_ = h_.replaceAll((regex(r"\n\n\n+", "m")), "\n\n"); - munged_header ~= h_; - } - } - // writeln(munged_header.join("\n")); - } - foreach (paragraph; paragraphs) { /+ loop to gather binary endnotes +/ - if (code_block_status == codeBlock.off - && paragraph.match(rgx_endnote) - ) { - munged_endnotes ~= replaceAll!(m => m[1]) - (paragraph, rgx_endnote); - } else { - if ( code_block_status != codeBlock.off - || paragraph.matchFirst(block_curly_code_open) - || paragraph.matchFirst(block_tic_code_open) - ) { /+ code blocks identified, no munging +/ - if ((code_block_status == codeBlock.curly - && paragraph.matchFirst(block_curly_code_close)) - || (code_block_status == codeBlock.tic - && paragraph.matchFirst(block_tic_close)) - ) { - code_block_status = codeBlock.off; - } else if (paragraph.matchFirst(block_curly_code_open)) { - code_block_status = codeBlock.curly; - } else if (paragraph.matchFirst(block_tic_code_open)) { - code_block_status = codeBlock.tic; - } - munged_contents ~= paragraph; - } else { /+ regular content, not a code block +/ - if (auto m = paragraph.matchAll(rgx_endnote_ref)) { - foreach (n; m) { - endnote_ref_count++; // endnote_refs ~= (n.captures[1]); - } - } - paragraph = format_body_string(paragraph); - // paragraph = replaceAll!(m => " \\\\ " ) - // (paragraph, regex(r"\s*<(?:/\s*|:)?br>\s*")); // (paragraph, regex(r"(
)")); - munged_contents ~= paragraph; - } - } - } - { - import std.outbuffer; - auto buffer = new OutBuffer(); - if (munged_header.length > 0) { - foreach (header; munged_header) { /+ loop to inline endnotes +/ - buffer.write(header ~ "\n"); - } - } - if (munged_endnotes.length == endnote_ref_count) { - int endnote_count = -1; - foreach (k, content; munged_contents) { /+ loop to inline endnotes +/ - content = replaceAll!(m => "~{ " ~ munged_endnotes[++endnote_count] ~ " }~" ~ m["tail"] ) - (content, rgx_endnote_ref); // endnote_ref cannot occur in a code block or else fail - buffer.write(content ~ ((k == munged_contents.length - 1) ? "" : "\n\n")); - } - if (buffer) { - try { - string dir_out = "converted_output_"; - string path_and_file_out = dir_out ~ "/" ~ filename; - dir_out.mkdirRecurse; - auto f = File(path_and_file_out, "w"); - f.write(buffer); - // writeln("wrote: ", path_and_file_out); - } catch (FileException ex) { - writeln("did not write file"); - // Handle errors - } - } - } else { - foreach (content; munged_contents) { /+ loop to inline endnotes +/ - buffer.write(content ~ "\n\n"); - } - } - } - } catch (ErrnoException ex) { - switch(ex.errno) { - case EPERM: - case EACCES: // Permission denied - break; - case ENOENT: // File does not exist - break; - default: // Handle other errors - break; - } - } - } - } -} diff --git a/misc/util/d/tools/spine_scaffold.d b/misc/util/d/tools/spine_scaffold.d deleted file mode 100755 index dbcc857..0000000 --- a/misc/util/d/tools/spine_scaffold.d +++ /dev/null @@ -1,134 +0,0 @@ -#!/usr/bin/env rund -void main( string[] args ) { - import std; - if (args.length > 1) { - string base_fn_path = args[1].expandTilde; - string base_fn = base_fn_path.baseName; - string sst_fn = base_fn ~ ".sst"; - string txt_for_pod_manifest = format(q"┃doc: - filename: %s - language: en -┃", - sst_fn - ); - string txt_for_sisu_document_make = format(q"┃ -┃", - ); - string txt_for_document_scaffold = format(q"┃# SiSU 8.0 - -title: "As Yet Unnamed" - -creator: - author: "Annon, Unnamed" - -:A~ @title @author - -1~ Summary - -To get you started, the first paragraph following a section or chapter heading. - -Spine / SiSU documents minimum requirements: - -_* a header containing document metadata that must at least contain the fields Title and Creator Author. - -_* text body, identified as starting by the A~ marker at the start of a line, followed by at least one level 1~ section heading with the text that follows it.~{ the document provided here would be a valid Spine document, and this text contained within the tilde and curly braces delimiters would be the first footnote/endnote }~ - -To generate this document to have html and epub output for example you would run: - -``` code -spine --html --epub --output=/tmp/spine-sample-output %s -``` - -1~ Conclusion - -This sample pod is provided to get you started. - -Good luck and good speed. -┃", - base_fn_path - ); - if (!exists(base_fn_path)) { - try { - base_fn_path.mkdirRecurse; - } catch (ErrnoException ex) { - writeln(ex); - } - if (exists(args[1].expandTilde)) { - try { - base_fn_path.buildPath("conf").mkdirRecurse; - } catch (ErrnoException ex) { - writeln(ex); - } - try { - base_fn_path.buildPath("media/text/en").mkdirRecurse; - } catch (ErrnoException ex) { - writeln(ex); - } - { - // 1 // create/write pod.manifest - string fn = base_fn_path.buildPath("pod.manifest"); - File(fn, "w").writeln(txt_for_pod_manifest); - string tell = format(q"┃OK - pod.manifest (yaml file containing filename and languages) - %s -%s -┃", - fn, - txt_for_pod_manifest.strip - ); - writeln(tell); - } - if (exists(base_fn_path.buildPath("conf"))) { - // 2 // create/write conf/sisu_document_make - string fn = base_fn_path.buildPath("conf/sisu_document_make"); - File(fn, "w").writeln(txt_for_sisu_document_make); - // auto f = File(fn, "w"); - // foreach (line; content_array) { - // f.writeln(line); - // } - string tell = format(q"┃OK - sisu_document_make - %s -┃", - fn - ); - writeln(tell); - } - if (exists(base_fn_path.buildPath("media/text/en"))) { - // 3 // create/write media/text/[lang code]/[filename].sst - string fn = base_fn_path.buildPath("media/text/en/" ~ sst_fn); - File(fn, "w").writeln(txt_for_document_scaffold); - // auto f = File(fn, "w"); - // foreach (line; content_array) { - // f.writeln(line); - // } - string tell = format(q"┃OK - .sst [document text content] - %s - - To start editing document (spine pod content): - ${EDITOR} %s - - To generate this document to have html and epub output for example you would run: - spine --html --epub --output=/tmp/spine-sample-output %s -┃", - fn, - fn, - base_fn_path - ); - writeln(tell); - } - } - /+ - pod/[filename] - │ - ├── conf - │   └── sisu_document_make - ├── media - │   └── text - │   └── en - │   └── [filename].charles_stross.sst - └── pod.manifest - +/ - } else { - writeln("requested output pod name with path already exists:\n ", args[1].expandTilde); - } - } else { - writeln( "please provide directory path to operate on, e.g.\n spine_scaffold.d ./pod/filetest" ); - } -} -- cgit v1.2.3