aboutsummaryrefslogtreecommitdiffstats
path: root/lib/docbuilder/src/docb_html_util.erl
diff options
context:
space:
mode:
Diffstat (limited to 'lib/docbuilder/src/docb_html_util.erl')
-rw-r--r--lib/docbuilder/src/docb_html_util.erl542
1 files changed, 0 insertions, 542 deletions
diff --git a/lib/docbuilder/src/docb_html_util.erl b/lib/docbuilder/src/docb_html_util.erl
deleted file mode 100644
index 02ce8b52a7..0000000000
--- a/lib/docbuilder/src/docb_html_util.erl
+++ /dev/null
@@ -1,542 +0,0 @@
-%% ``The contents of this file are subject to the Erlang Public License,
-%% Version 1.1, (the "License"); you may not use this file except in
-%% compliance with the License. You should have received a copy of the
-%% Erlang Public License along with this software. If not, it can be
-%% retrieved via the world wide web at http://www.erlang.org/.
-%%
-%% Software distributed under the License is distributed on an "AS IS"
-%% basis, WITHOUT WARRANTY OF ANY KIND, either expressed or implied. See
-%% the License for the specific language governing rights and limitations
-%% under the License.
-%%
-%% The Initial Developer of the Original Code is Ericsson Utvecklings AB.
-%% Portions created by Ericsson are Copyright 1999-2000, Ericsson
-%% Utvecklings AB. All Rights Reserved.''
-%%
-%% $Id$
-%%
--module(docb_html_util).
-
--export([attribute_cdata_to_html/1,
- element_cdata_to_html/1,
- pcdata_to_html/1, pcdata_to_html/2]).
--export([copy_pics/3]).
--export([extract_header_data/2, all_header_data/1]).
--export([make_uri/1,
- make_anchor_href/1, make_anchor_href_short/3,
- make_anchor_name_short/2,
- make_funcdef_short/2]).
--export([erl_include/2, code_include/2, erl_eval/1]).
--export([number/3, count_sections/1]).
--export([format_toc/1]).
--export([html_latin1_sort_order/1]).
-
-%%--Handle CDATA and PCDATA---------------------------------------------
-
-%% NB: Functions for transforming sgmls/XMerL data output to html.
-%% Do not use these for included text files (cf code_include and
-%% erl_include).
-
-attribute_cdata_to_html(Data) ->
- data2html(Data, false).
-
-element_cdata_to_html(Data) ->
- data2html(Data, false).
-
-pcdata_to_html(Data) ->
- data2html(Data, true).
-
-pcdata_to_html(Data, RmSp) ->
- data2html(Data, RmSp).
-
-%% PCDATA, CDATA: Replace entities, and optionally delete
-%% leading and multiple spaces. CDATA never contains entities to
-%% replace.
-
-%% data2html(Cs, RmSpace)
-data2html([246| Cs], RmSp) ->
- [$&, $#, $2, $4, $6, $;| data2html(Cs, RmSp)];
-data2html([$>| Cs], RmSp) ->
- [$&, $#, $6, $2, $;| data2html(Cs, RmSp)];
-data2html([$<| Cs], RmSp) ->
- [$&, $#, $6, $0, $;| data2html(Cs, RmSp)];
-data2html([$&| Cs], RmSp) ->
- [$&, $#, $3, $8, $;| data2html(Cs, RmSp)];
-data2html([$\"| Cs], RmSp) ->
- [$&, $#, $3, $4, $;| data2html(Cs, RmSp)];
-data2html([$\n| Cs], RmSp) ->
- data2html(Cs, RmSp);
-data2html([$\\, $n| Cs], false) ->
- [$\n| data2html(Cs, false)];
-data2html([$\\, $n| Cs], true) ->
- [$\n| data2html(delete_leading_space(Cs), true)];
-data2html([$ , $ | Cs], true) -> % delete multiple space
- [$ | data2html(delete_leading_space(Cs), true)];
-data2html([$\\, $|| Cs0], RmSp) ->
- {Ent, Cs1} = collect_entity(Cs0),
- [entity_to_html(Ent)| data2html(Cs1, RmSp)];
-data2html([$\\, $0, $1, $2| Cs], RmSp) ->
- data2html(Cs, RmSp);
-data2html([$\\, $\\, $n| Cs], RmSp) ->
- [$\\, $n| data2html(Cs, RmSp)];
-data2html([$\\, O1, O2, O3| Cs], RmSp)
- when O1 >= $0, O1 =< $7, O2 >= $0, O2 =< $7, O3 >= $0, O3 =< $7 ->
- case octal2dec(O1, O2, O3) of
- 173 -> % soft hyphen
- data2html(Cs, RmSp);
- C when C > 31, C < 256 ->
- Ent = io_lib:format("&#~w;", [C]),
- [Ent| data2html(Cs, RmSp)];
- C ->
- [C| data2html(Cs, RmSp)]
- end;
-data2html([$\\, $\\| Cs], RmSp) ->
- [$\\| data2html(Cs, RmSp)];
-data2html([C| Cs], RmSp) ->
- [C| data2html(Cs, RmSp)];
-data2html([], _) ->
- [].
-
-delete_leading_space([$ | Cs]) ->
- delete_leading_space(Cs);
-delete_leading_space(Cs) ->
- Cs.
-
-collect_entity(Data) ->
- collect_entity(Data, []).
-
-collect_entity([$\\, $|| Cs], Rs) ->
- {lists:reverse(Rs), Cs};
-collect_entity([C| Cs], Rs) ->
- collect_entity(Cs, [C| Rs]);
-collect_entity([], Rs) ->
- {[], lists:reverse(Rs)}.
-
-entity_to_html("&") -> "&#38;";
-entity_to_html("\"") -> "&#34;";
-entity_to_html("<") -> "&#60;";
-entity_to_html(">") -> "&#62;";
-entity_to_html([$\\, O1, O2, O3])
- when O1 >= $0, O1 =< $7, O2 >= $0, O2 =< $7, O3 >= $0, O3 =< $7 ->
- case octal2dec(O1, O2, O3) of
- 173 -> % soft hyphen
- "";
- Value ->
- io_lib:format("&#~w;", [Value])
- end;
-entity_to_html(Other) ->
- docb_html_util_iso:entity_to_html(Other).
-
-octal2dec(O1, O2, O3) ->
- (O1*8+O2)*8+O3-73*$0.
-
-%%--Copy images---------------------------------------------------------
-
-copy_pics(Src, Dest, Opts) ->
- Dir = code:lib_dir(docbuilder),
- InFile = filename:join([Dir, "etc", Src]),
- OutFile = docb_util:outfile(Dest, "", Opts),
- case filelib:last_modified(OutFile) of
- 0 -> % File doesn't exist
- file:copy(InFile, OutFile);
-
- OutMod2 ->
- InMod1s = calendar:datetime_to_gregorian_seconds(
- filelib:last_modified(InFile)),
- OutMod2s = calendar:datetime_to_gregorian_seconds(OutMod2),
- if
- InMod1s > OutMod2s -> % InFile is newer than OutFile
- file:copy(InFile, OutFile);
- true ->
- ok
- end
- end.
-
-%%--Resolve header data-------------------------------------------------
-
-extract_header_data(Key, {header, [], List}) ->
- case lists:keyfind(Key, 1, List) of
- {Key, [], []} ->
- "";
- {Key, [], [{pcdata, [], Value}]} ->
- pcdata_to_html(Value);
- false ->
- ""
- end.
-
-all_header_data(Header) ->
- all_header_data(Header,
- [title, prepared, responsible, docno, approved,
- checked, date, rev, file]).
-
-all_header_data(_Header, []) ->
- [];
-all_header_data(Header, [Key| Rest]) ->
- [extract_header_data(Key, Header) | all_header_data(Header, Rest)].
-
-%%--Resolve hypertext references----------------------------------------
-
-%% URI regular expression (RFC 2396):
-%% "^(([^:/?#]+):)?(//([^/?#]*))?([^?#]*)(\\?([^#]*))?(#(.*))?"
-%% We split it in five parts:
-%% scheme: "^(([^:/?#]+):)?" (includes trailing `:')
-%% authority: "^(//([^/?#]*))?" (includes leading `//')
-%% path: "^([^?#]*)"
-%% query: "^(\\?([^#]*))?" (includes leading `?')
-%% fragment: "^(#(.*))?" (includes leading `#')
-
-make_uri(Cs) ->
- lists:flatmap(
- fun({path, S}) ->
- case regexp:match(S, "\.xml?\$") of
- {match, _, _} ->
- {ok, NS, _} = regexp:sub(S, "\.xml?\$", ".html"),
- NS;
- _ ->
- S
- end;
- ({_, S}) ->
- S
- end,
- split_uri(Cs)).
-
-split_uri(URI) ->
- split_uri(URI, [{scheme, "^(([^:/?#]+):)?"},
- {authority, "^(//([^/?#]*))?"},
- {path, "^([^?#]*)"},
- {'query', "^(\\?([^#]*))?"},
- {fragment, "^(#(.*))?"}]).
-
-split_uri("", [{Tag, _R}| T]) ->
- [{Tag, ""}| split_uri("", T)];
-split_uri(Cs0, [{Tag, R}| T]) ->
- {match, 1, N} = regexp:match(Cs0, R),
- Cs1 = string:substr(Cs0, 1, N),
- Cs2 = strip_and_escape_uri_component(Tag, Cs1),
- [{Tag, Cs2}| split_uri(string:substr(Cs0, N+1), T)];
-split_uri(_, []) ->
- [].
-
-strip_and_escape_uri_component(authority, "//" ++ Cs) ->
- "//" ++ escape_uri(string:strip(Cs));
-strip_and_escape_uri_component(path, Cs) ->
- escape_uri(string:strip(Cs));
-strip_and_escape_uri_component('query', "?" ++ Cs) ->
- "?" ++ escape_uri(string:strip(Cs));
-strip_and_escape_uri_component(fragment, "#" ++ Cs) ->
- "#" ++ escape_uri(string:strip(Cs));
-strip_and_escape_uri_component(_, "") ->
- "";
-strip_and_escape_uri_component(_, Cs) ->
- escape_uri(string:strip(Cs)).
-
-escape_uri([C|Cs]) when C =< 32;
- C == $<; C == $<; C == $#; C == $%; C == $";
- C == $?;
- C == ${; C == $}; C ==$|; C == $\\; C == $^;
- C == $[; C == $]; C ==$';
- C >= 127 ->
- [$%, mk_hex(C div 16), mk_hex(C rem 16)| escape_uri(Cs)];
-escape_uri([C|Cs]) ->
- [C|escape_uri(Cs)];
-escape_uri([]) ->
- [].
-
-mk_hex(C) when C<10 ->
- C + $0;
-mk_hex(C) ->
- C - 10 + $a.
-
-make_anchor_href(HRef) ->
- case regexp:split(HRef, "#") of
- {ok, [HRef]} ->
- %% No `#' in HRef, i.e. only path
- make_anchor_href(HRef, "");
- {ok, [Path, Fragment]} ->
- make_anchor_href(Path, Fragment)
- end.
-
-make_anchor_href(Path0, Frag0) ->
- Frag1 = string:strip(Frag0),
- Path1 = case Path0 of
- "" ->
- "";
- _ ->
- case regexp:match(Path0, "\.xml?\$") of
- nomatch ->
- Path0 ++ ".html";
- _ ->
- {ok, NewPath, _} = regexp:sub(Path0,
- "\.xml?\$",
- ".html"),
- NewPath
- end
- end,
- case Frag1 of
- "" ->
- attribute_cdata_to_html(Path1);
- _ ->
- attribute_cdata_to_html(Path1) ++
- "#" ++
- attribute_cdata_to_html([case Ch of $/ -> $-; _ -> Ch end||
- Ch <-Frag1])
- end.
-
-make_anchor_href_short(Path, Frag, RefType) ->
- ShortFrag = make_funcdef_short(Frag, RefType,"-"),
- make_anchor_href(Path, ShortFrag).
-
-make_anchor_name_short(FuncName0, RefType) ->
- FuncName1 = make_funcdef_short(FuncName0, RefType,"-"),
- attribute_cdata_to_html(FuncName1).
-
-make_funcdef_short(FuncDef0, RefType) ->
- make_funcdef_short(FuncDef0, RefType, "/").
-
-make_funcdef_short(FuncDef0, RefType,Delimiter) ->
- FuncDef1 = docb_util:trim(FuncDef0),
- Any0 = case lists:member(RefType, [cref, erlref]) of
- true ->
- case catch docb_util:fknidx(FuncDef1, Delimiter) of
- {'EXIT', _} ->
- false;
- Any1 ->
- Any1
- end;
- false ->
- false
- end,
- case Any0 of
- false ->
- case string:tokens(FuncDef1, " ") of
- [Any2| _] ->
- Any2;
- _ ->
- FuncDef1
- end;
- _ ->
- Any0
- end.
-
-%%--Include tags--------------------------------------------------------
-
-%% Only used in report DTD
-erl_include(File, Tag) ->
- case docb_main:include_file(File, Tag) of
- {ok, Cs} ->
- {drop, "\n<pre>\n" ++ text_to_html(Cs) ++ "\n</pre>\n"};
- error ->
- {drop, ""}
- end.
-
-code_include(File, Tag) ->
- case docb_main:include(File, Tag, Tag) of
- {ok, Cs} ->
- {ok,text_to_html(Cs)};
- error ->
- {error, {codeinclude,File}}
- end.
-
-erl_eval(Expr) ->
- Cs = docb_main:eval_str(Expr),
- {drop, "\n<pre>\n" ++ text_to_html(Cs) ++ "\n</pre>\n"}.
-
-%% Only replaces certain characters. Spaces and new lines etc are kept.
-%% Used for plain text (e.g. inclusions of code).
-text_to_html([$>| Cs]) ->
- [$&, $#, $6, $2, $;| text_to_html(Cs)];
-text_to_html([$<| Cs]) ->
- [$&, $#, $6, $0, $;| text_to_html(Cs)];
-text_to_html([$&| Cs]) ->
- [$&, $#, $3, $8, $;| text_to_html(Cs)];
-text_to_html([$\"| Cs]) ->
- [$&, $#, $3, $4, $;| text_to_html(Cs)];
-text_to_html([C| Cs]) ->
- [C| text_to_html(Cs)];
-text_to_html([]) ->
- [].
-
-%%--Number sections-----------------------------------------------------
-
-number({Tag, Attrs, More}, none, File) ->
- {Tag, Attrs, do_number(More, [1], File)};
-number({Tag, Attrs, More}, Prefix, File) ->
- {Tag, Attrs, do_number(More, [list_to_integer(Prefix)], File)}.
-
-do_number([], _, _) ->
- [];
-do_number([{header, Attrs, More}| Rest], NN, File) ->
- [{header, Attrs, More}| do_number(Rest, NN, File)];
-do_number([{section, Attrs, More}| Rest], [N| NN], File) ->
- [{section, Attrs, do_number(More, [1, N| NN], File)}|
- do_number(Rest, [N+1| NN], File)];
-do_number([{title, _, [{pcdata, _, Title}]}| More], [N| NN], File) ->
- Format = make_format(length(NN)),
- Number = lists:flatten(io_lib:format(Format, lists:reverse(NN))),
- [{marker, [{"ID", "CDATA", Number}], []},
- {title, [{"NUMBER", "CDATA", Number},
- {"FILE", "CDATA", File}],
- [{pcdata, [], Title}]}| do_number(More, [N| NN], File)];
-do_number([{pcdata, Attrs, More}| Rest], NN, File) ->
- [{pcdata, Attrs, More}| do_number(Rest, NN, File)];
-do_number([{Tag, Attrs, More}| Rest], NN, File) ->
- [{Tag, Attrs, do_number(More, NN, File)}|do_number(Rest, NN, File)].
-
-make_format(1) ->
- "~w";
-make_format(N) ->
- "~w." ++ make_format(N-1).
-
-count_sections([section| Rest]) ->
- 1 + count_sections(Rest);
-count_sections([_| Rest]) ->
- count_sections(Rest);
-count_sections([]) ->
- 0.
-
-%%--Make a ToC----------------------------------------------------------
-
-format_toc(Toc) ->
- [format_toc1(T) || T <- Toc].
-
-format_toc1({Number, Title}) ->
- [Number, " <a href = \"#", Number, "\">", Title, "</a><br/>\n"].
-
-%%--Convert HTML ISO Latin 1 characters to ordinary characters----------
-
-%% To be used for sorting. Cs must be flat.
-html_latin1_sort_order(Cs) ->
- hlso(Cs).
-
-hlso([]) ->
- [];
-hlso([$&, $#, C2, C1, C0, $;| Cs])
- when $0 =< C2, C2 =< $9, $0 =< C1, C1 =< $9, $0 =< C0, C0 =< $9 ->
- C = ((C2-$0)*10 + (C1-$0))*10 + C0-$0,
- hlso0(C, Cs);
-hlso([$&, $#, C1, C0, $;| Cs])
- when $0 =< C1, C1 =< $9, $0 =< C0, C0 =< $9 ->
- C = (C1-$0)*10 + C0-$0,
- hlso0(C, Cs);
-hlso([C| Cs]) ->
- [C| hlso(Cs)].
-
-hlso0(C, Cs) when 0 =< C, C =< 159 ->
- [C| hlso(Cs)];
-hlso0(160, Cs) -> %% no-break space
- hlso(Cs); % Remove it.
-hlso0(161, Cs) -> %% inverted exclamation mark
- [$? |hlso(Cs)];
-hlso0(162, Cs) -> %% cent sign
- [$$|hlso(Cs)];
-hlso0(163, Cs) -> %% pound sterling sign
- [$$|hlso(Cs)];
-hlso0(164, Cs) -> %% general currency sign
- [$$|hlso(Cs)];
-hlso0(165, Cs) -> %% yen sign
- [$$|hlso(Cs)];
-hlso0(166, Cs) -> %% broken (vertical) bar
- [$| |hlso(Cs)];
-hlso0(167, Cs) -> %% section sign
- [$$|hlso(Cs)];
-hlso0(168, Cs) -> %% umlaut (dieresis)
- [$: |hlso(Cs)];
-hlso0(169, Cs) -> %% copyright sign
- [$c |hlso(Cs)];
-hlso0(170, Cs) -> %% ordinal indicator, feminine
- [$f |hlso(Cs)];
-hlso0(171, Cs) -> %% angle quotation mark, left
- [$" |hlso(Cs)];
-hlso0(172, Cs) -> %% not sign
- [$- |hlso(Cs)];
-hlso0(173, Cs) -> %% soft hyphen
- [$- |hlso(Cs)];
-hlso0(174, Cs) -> %% registered sign
- [$r |hlso(Cs)];
-hlso0(175, Cs) -> %% macron
- [$- |hlso(Cs)];
-hlso0(176, Cs) -> %% degree sign
- [$d |hlso(Cs)];
-hlso0(177, Cs) -> %% plus-or-minus sign
- [$+ |hlso(Cs)];
-hlso0(178, Cs) -> %% superscript two
- [$2 |hlso(Cs)];
-hlso0(179, Cs) -> %% superscript three
- [$3 |hlso(Cs)];
-hlso0(180, Cs) -> %% acute accent
- [$' |hlso(Cs)];
-hlso0(181, Cs) -> %% micro sign
- [$' |hlso(Cs)];
-hlso0(182, Cs) -> %% pilcrow (paragraph sign)
- [$$|hlso(Cs)];
-hlso0(183, Cs) -> %% middle dot
- [$. |hlso(Cs)];
-hlso0(184, Cs) -> %% cedilla
- [$c |hlso(Cs)];
-hlso0(185, Cs) -> %% superscript one
- [$1 |hlso(Cs)];
-hlso0(186, Cs) -> %% ordinal indicator, masculine
- [$m |hlso(Cs)];
-hlso0(187, Cs) -> %% angle quotation mark, right
- [$" |hlso(Cs)];
-hlso0(188, Cs) -> %% fraction one-quarter
- [$4 |hlso(Cs)];
-hlso0(189, Cs) -> %% fraction one-half
- [$2 |hlso(Cs)];
-hlso0(190, Cs) -> %% fraction three-quarters
- [$3 |hlso(Cs)];
-hlso0(191, Cs) -> %% inverted question mark
- [$? |hlso(Cs)];
-
-hlso0(C, Cs) when 192 =< C, C =< 198 -> %% capital A
- [$A |hlso(Cs)];
-hlso0(199, Cs) -> %% capital C, cedilla
- [$C |hlso(Cs)];
-hlso0(C, Cs) when 200 =< C, C =< 203 -> %% capital E
- [$E |hlso(Cs)];
-hlso0(C, Cs) when 204 =< C, C =< 207 -> %% capital I
- [$I |hlso(Cs)];
-hlso0(208, Cs) -> %% capital Eth, Icelandic
- [$D |hlso(Cs)];
-hlso0(209, Cs) -> %% capital N, tilde
- [$N |hlso(Cs)];
-hlso0(C, Cs) when 210 =< C, C =< 214 -> %% capital O
- [$O |hlso(Cs)];
-hlso0(215, Cs) -> %% multiply sign
- [$x |hlso(Cs)];
-hlso0(216, Cs) -> %% capital O, slash
- [$O |hlso(Cs)];
-hlso0(C, Cs) when 217 =< C, C =< 220 -> %% capital U
- [$U |hlso(Cs)];
-hlso0(221, Cs) -> %% capital Y, acute accent
- [$Y |hlso(Cs)];
-hlso0(222, Cs) -> %% capital THORN, Icelandic
- [$T |hlso(Cs)];
-hlso0(223, Cs) -> %% small sharp s, German (sz
- [$s |hlso(Cs)];
-hlso0(C, Cs) when 224 =< C, C =< 230-> %% small a
- [$a |hlso(Cs)];
-hlso0(231, Cs) -> %% small c, cedilla
- [$c |hlso(Cs)];
-hlso0(C, Cs) when 232 =< C, C =< 235 -> %% small e
- [$e |hlso(Cs)];
-hlso0(C, Cs) when 236 =< C, C =< 239 -> %% small i
- [$i |hlso(Cs)];
-hlso0(240, Cs) -> %% small eth, Icelandic
- [$d |hlso(Cs)];
-hlso0(241, Cs) -> %% small n, tilde
- [$n |hlso(Cs)];
-hlso0(C, Cs) when 242 =< C, C =< 246 -> %% small o
- [$o |hlso(Cs)];
-hlso0(247, Cs) -> %% divide sign
- [$/ |hlso(Cs)];
-hlso0(248, Cs) -> %% small o, slash
- [$o |hlso(Cs)];
-hlso0(C, Cs) when 249 =< C, C =< 252 -> %% small u
- [$u |hlso(Cs)];
-hlso0(253, Cs) -> %% small y, acute accent
- [$y |hlso(Cs)];
-hlso0(254, Cs) -> %% small thorn, Icelandic
- [$t |hlso(Cs)];
-hlso0(255, Cs) -> %% small y, dieresis or umlaut
- [$y |hlso(Cs)].