aboutsummaryrefslogtreecommitdiffstats
path: root/lib/docbuilder/src/docb_edoc_xml_cb.erl
diff options
context:
space:
mode:
Diffstat (limited to 'lib/docbuilder/src/docb_edoc_xml_cb.erl')
-rw-r--r--lib/docbuilder/src/docb_edoc_xml_cb.erl1154
1 files changed, 0 insertions, 1154 deletions
diff --git a/lib/docbuilder/src/docb_edoc_xml_cb.erl b/lib/docbuilder/src/docb_edoc_xml_cb.erl
deleted file mode 100644
index 90491bc007..0000000000
--- a/lib/docbuilder/src/docb_edoc_xml_cb.erl
+++ /dev/null
@@ -1,1154 +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 Licence for the specific language governing rights and limitations
-%% under the License.
-%%
-%% The Initial Developer of the Original Code is Ericsson AB.
-%% Portions created by Ericsson are Copyright 1999-2006, Ericsson AB.
-%% All Rights Reserved.��
-%%
-%% $Id$
-%%
--module(docb_edoc_xml_cb).
-
-%% This is the EDoc callback module for creating DocBuilder erlref
-%% documents (man pages) in XML format, and also a DocBuilder chapter
-%% document based on "overview.edoc".
-%%
-%% Usage examples:
-%% docb_gen File
-%% docb_gen -chapter overview.edoc
-%% or (from an Erlang shell)
-%% edoc:file(File, [{layout,docb_edoc_xml_cb},{file_suffix,".xml"},
-%% {preprocess,true}]).
-%%
-%% The origin of this file is the edoc module otpsgml_layout.erl
-%% written by Richard Carlsson.
-
--export([module/2, overview/2]).
-
--include("xmerl.hrl").
-
--define(NL, "\n").
-
-%%-User interface-------------------------------------------------------
-
-%% ERLREF
-module(Element, Opts) ->
- SortP = proplists:get_value(sort_functions, Opts, true),
- XML = layout_module(Element, SortP),
- xmerl:export_simple([XML], docb_xmerl_xml_cb, []).
-
-%% CHAPTER
-overview(Element, _Opts) ->
- XML = layout_chapter(Element),
- xmerl:export_simple([XML], docb_xmerl_xml_cb, []).
-
-%%--Internal functions--------------------------------------------------
-
-layout_module(#xmlElement{name = module, content = Es}=E, SortP) ->
- Name = get_attrval(name, E),
- Desc = get_content(description, Es),
- ShortDesc = text_only(get_content(briefDescription, Desc)),
- FullDesc = otp_xmlify(get_content(fullDescription, Desc)),
- Types0 = get_content(typedecls, Es),
- Types1 = lists:sort([{type_name(Et), Et} || Et <- Types0]),
- Functions =
- case SortP of
- true ->
- lists:sort([{function_name(Ef), Ef} ||
- Ef <- get_content(functions, Es)]);
- false ->
- [{function_name(Ef), Ef} ||
- Ef <- get_content(functions, Es)]
- end,
- Header = {header, [
- ?NL,{title, [Name]},
- ?NL,{prepared, [""]},
- ?NL,{responsible, [""]},
- ?NL,{docno, ["1"]},
- ?NL,{approved, [""]},
- ?NL,{checked, [""]},
- ?NL,{date, [""]},
- ?NL,{rev, ["A"]},
- ?NL,{file, [Name++".xml"]}
- ]},
- Module = {module, [Name]},
- ModuleSummary = {modulesummary, ShortDesc},
- Description = {description, [?NL|FullDesc]},
- Types = case Types1 of
- [] -> [];
- _ ->
- [?NL, {section,[{title,["DATA TYPES"]},
- {marker,[{id,"types"}],[]},
- ?NL|types(Types1)]}]
- end,
- Funcs = functions(Functions),
- See = seealso_module(Es),
- Authors = {authors, authors(Es)},
- {erlref,
- [?NL,Header,
- ?NL,Module,
- ?NL,ModuleSummary,
- ?NL,Description]
- ++ Types ++
- [?NL,Funcs,
- ?NL,See,
- ?NL,Authors]
- }.
-
-layout_chapter(#xmlElement{name=overview, content=Es}) ->
- Title = get_text(title, Es),
- Header = {header, [
- ?NL,{title,[Title]},
- ?NL,{prepared,[""]},
- ?NL,{docno,[""]},
- ?NL,{date,[""]},
- ?NL,{rev,[""]},
- ?NL,{file, ["chapter.xml"]}
- ]},
- DescEs = get_content(description, Es),
- FullDescEs = get_content(fullDescription, DescEs),
- Sections = chapter_ify(FullDescEs, first),
- {chapter, [?NL, Header, ?NL | Sections]}.
-
-chapter_ify([], _) ->
- [];
-chapter_ify(Es, first) ->
- %% Everything up to the first section should be made into
- %% plain paragraphs -- or if no first section is found, everything
- %% should be made into one
- case find_next(h3, Es) of
- {Es, []} ->
- SubSections = subchapter_ify(Es, first),
- [{section, [?NL,{title,["Overview"]},
- ?NL | SubSections]}];
- {FirstEs, RestEs} ->
- otp_xmlify(FirstEs) ++ chapter_ify(RestEs, next)
- end;
-chapter_ify([#xmlElement{name=h3} = E | Es], next) ->
- {SectionEs, RestEs} = find_next(h3, Es),
- SubSections = subchapter_ify(SectionEs, first),
- {Marker, Title} = chapter_title(E),
- [{section, [?NL,{marker,[{id,Marker}],[]},
- ?NL,{title,[Title]},
- ?NL | SubSections]} | chapter_ify(RestEs, next)].
-
-subchapter_ify([], _) ->
- [];
-subchapter_ify(Es, first) ->
- %% Everything up to the (possible) first subsection should be
- %% made into plain paragraphs
- {FirstEs, RestEs} = find_next(h4, Es),
- otp_xmlify(FirstEs) ++ subchapter_ify(RestEs, next);
-subchapter_ify([#xmlElement{name=h4} = E | Es], next) ->
- {SectionEs, RestEs} = find_next(h4, Es),
- Elements = otp_xmlify(SectionEs),
- {Marker, Title} = chapter_title(E),
- [{section, [?NL,{marker,[{id,Marker}],[]},
- ?NL,{title,[Title]},
- ?NL | Elements]} | subchapter_ify(RestEs, next)].
-
-chapter_title(#xmlElement{content=Es}) -> % name = h3 | h4
- case Es of
- [#xmlElement{name=a} = E] ->
- {get_attrval(name, E), get_text(E)}
- end.
-
-%%--XHTML->XML transformation-------------------------------------------
-
-%% otp_xmlify(Es1) -> Es2
-%% Es1 = Es2 = [#xmlElement{} | #xmlText{}]
-%% Fix things that are allowed in XHTML but not in chapter/erlref DTDs.
-%% 1) lists (<ul>, <ol>, <dl>) and code snippets (<pre>) can not occur
-%% within a <p>, such a <p> must be splitted into a sequence of <p>,
-%% <ul>, <ol>, <dl> and <pre>.
-%% 2) <a> must only have either a href attribute (corresponds to a
-%% <seealso> or <url> in the XML code) in which case its content
-%% must be plain text; or a name attribute (<marker>).
-%% 3a) <b> content must be plain text.
-%% b) <em> content must be plain text (or actually a <code> element
-%% as well, but a simplification is used here).
-%% c) <pre> content must be plain text (or could actually contain
-%% <input> as well, but a simplification is used here).
-%% 4) <code> content must be plain text, or the element must be split
-%% into a list of elements.
-%% 5a) <h1>, <h2> etc is not allowed - replaced by its content within
-%% a <b> tag.
-%% b) <center> is not allowed - replaced by its content.
-%% c) <font> -"-
-%% 6) <table> is not allowed in erlref, translated to text instead.
-%% Also a <table> in chapter without a border is translated to text.
-%% A <table> in chapter with a border must contain a <tcaption>.
-%% 7) <sup> is not allowed - is replaced with its text content
-%% within parenthesis.
-%% 8) <blockquote> contents may need to be made into paragraphs
-%% 9) <th> (table header) is not allowed - is replaced by
-%% <td><em>...</em></td>.
-otp_xmlify([]) ->
- [];
-otp_xmlify(Es0) ->
- Es = case is_paragraph(hd(Es0)) of
-
- %% If the first element is a <p> xmlElement, then
- %% the entire element list is ready to be otp_xmlified.
- true ->
- Es0;
-
- %% If the first element is not a <p> xmlElement, then all
- %% elements up to the first <p> (or end of list) must be
- %% made into a paragraph (using the {p, Es} construction)
- %% before the list is otp_xmlified.
- false ->
- case find_next(p, Es0, []) of
- {[#xmlText{value=Str}] = First, Rest} ->
- %% Special case: Making a paragraph out of a
- %% blank line isn't a great idea.
- case is_empty(Str) of
- true ->
- Rest;
- false ->
- [{p,First}|Rest]
- end;
- {First, Rest} ->
- [{p,First}|Rest]
- end
- end,
-
- %% Fix paragraph breaks not needed in XHTML but in XML
- EsFixed = otp_xmlify_fix(Es),
-
- otp_xmlify_es(EsFixed).
-
-%% EDoc does not always translate empty lines (with leading "%%")
-%% as paragraph break, this is the fix
-otp_xmlify_fix(Es) ->
- otp_xmlify_fix(Es, []).
-otp_xmlify_fix([#xmlText{value="\n \n"++_} = E1, E2 | Es], Res) ->
- %% This is how it looks when generating an erlref from a .erl file
- case is_paragraph(E2) of
- false ->
- {P, After} = find_p_ending(Es, []),
- otp_xmlify_fix(After, [{p, [E2|P]}, E1 | Res]);
- true ->
- otp_xmlify_fix([E2|Es], [E1|Res])
- end;
-otp_xmlify_fix([#xmlText{value="\n\n"} = E1, E2 | Es], Res) ->
- %% This is how it looks when generating a chapter from overview.edoc
- case is_paragraph(E2) of
- false ->
- {P, After} = find_p_ending(Es, []),
- otp_xmlify_fix(After, [{p, [E2|P]}, E1 | Res]);
- true ->
- otp_xmlify_fix([E2|Es], [E1|Res])
- end;
-otp_xmlify_fix([E|Es], Res) ->
- otp_xmlify_fix(Es, [E|Res]);
-otp_xmlify_fix([], Res) ->
- lists:reverse(Res).
-
-otp_xmlify_es([E | Es]) ->
- case is_paragraph(E) of
- true ->
- case otp_xmlify_psplit(E) of
-
- %% paragraph contained inline tags and text only
- nosplit ->
- otp_xmlify_e(E) ++ otp_xmlify_es(Es);
-
- %% paragraph contained dl, ul and/or pre and has been
- %% splitted
- SubEs ->
- lists:flatmap(fun otp_xmlify_e/1, SubEs) ++
- otp_xmlify_es(Es)
- end;
- false ->
- otp_xmlify_e(E) ++ otp_xmlify_es(Es)
- end;
-otp_xmlify_es([]) ->
- [].
-
-%% otp_xmlify_psplit(P) -> nosplit | [E]
-%% Handles case 1) above.
-%% Uses the {p, Es} construct, thus replaces an p xmlElement with one
-%% or more {p, Es} tuples if splitting the paraghrap is necessary.
-otp_xmlify_psplit(P) ->
- otp_xmlify_psplit(p_content(P), [], []).
-otp_xmlify_psplit([#xmlElement{name=Name}=E | Es], Content, Res) ->
- if
- Name==blockquote; Name==ul; Name==ol; Name==dl; Name==pre;
- Name==table ->
- case Content of
- [] ->
- otp_xmlify_psplit(Es, [], [E|Res]);
- [#xmlText{value=Str}] ->
- %% Special case: Making a paragraph out of a blank
- %% line isn't a great idea. Instead, this can be
- %% viewed as the case above, where there is no
- %% content to make a paragraph out of
- case is_empty(Str) of
- true ->
- otp_xmlify_psplit(Es, [], [E|Res]);
- false ->
- Pnew = {p, lists:reverse(Content)},
- otp_xmlify_psplit(Es, [], [E,Pnew|Res])
- end;
- _ ->
- Pnew = {p, lists:reverse(Content)},
- otp_xmlify_psplit(Es, [], [E,Pnew|Res])
- end;
-
- true ->
- otp_xmlify_psplit(Es, [E|Content], Res)
- end;
-otp_xmlify_psplit([E | Es], Content, Res) ->
- otp_xmlify_psplit(Es, [E|Content], Res);
-otp_xmlify_psplit([], _Content, []) ->
- nosplit;
-otp_xmlify_psplit([], [], Res) ->
- lists:reverse(Res);
-otp_xmlify_psplit([], [#xmlText{value="\n\n"}], Res) ->
- lists:reverse(Res);
-otp_xmlify_psplit([], Content, Res) ->
- Pnew = {p, lists:reverse(Content)},
- lists:reverse([Pnew|Res]).
-
-%% otp_xmlify_e(E) -> [E]
-%% This is the function which does the actual transformation of
-%% single elements, normally by making sure the content corresponds
-%% to what is allowed by the OTP DTDs.
-%% Returns a list of elements as the xmlification in some cases
-%% returns no element or more than one element (although in most cases
-%% exactly one element).
-otp_xmlify_e(#xmlElement{name=a} = E) -> % 2) above
- otp_xmlify_a(E);
-otp_xmlify_e(#xmlElement{name=Tag} = E) % 3a-c)
- when Tag==b; Tag==em; Tag==pre ->
- Content = text_only(E#xmlElement.content),
- [E#xmlElement{content=Content}];
-otp_xmlify_e(#xmlElement{name=code} = E) -> % 4)
- case catch text_only(E#xmlElement.content) of
- {'EXIT', _Error} ->
- otp_xmlify_code(E);
- Content ->
- [E#xmlElement{content=Content}]
- end;
-otp_xmlify_e(#xmlElement{name=Tag} = E) % 5a
- when Tag==h1; Tag==h2; Tag==h3; Tag==h4; Tag==h5 ->
- Content = text_only(E#xmlElement.content),
- [E#xmlElement{name=b, content=Content}];
-otp_xmlify_e(#xmlElement{name=Tag} = E) % 5b-c)
- when Tag==center;
- Tag==font ->
- otp_xmlify_e(E#xmlElement.content);
-otp_xmlify_e(#xmlElement{name=table} = E) -> % 6)
- case parent(E) of
- module ->
- otp_xmlify_table(E#xmlElement.content);
- overview ->
- Content0 = otp_xmlify_e(E#xmlElement.content),
- Summary = #xmlText{value=get_attrval(summary, E)},
- TCaption = E#xmlElement{name=tcaption,
- attributes=[],
- content=[Summary]},
- Content = Content0 ++ [TCaption],
- [E#xmlElement{attributes=[], content=Content}]
- end;
-otp_xmlify_e(#xmlElement{name=tbody} = E) ->
- otp_xmlify_e(E#xmlElement.content);
-otp_xmlify_e(#xmlElement{name=sup} = E) -> % 7)
- Text = get_text(E),
- [#xmlText{parents = E#xmlElement.parents,
- pos = E#xmlElement.pos,
- language = E#xmlElement.language,
- value = "(" ++ Text ++ ")"}];
-otp_xmlify_e(#xmlElement{name=blockquote} = E) -> % 8)
- Content = otp_xmlify_blockquote(E#xmlElement.content),
- [E#xmlElement{content=Content}];
-otp_xmlify_e(#xmlElement{name=th} = E) -> % 9)
- Content = otp_xmlify_e(E#xmlElement.content),
- EmE = E#xmlElement{name=em, content=Content},
- [E#xmlElement{name=td, content=[EmE]}];
-otp_xmlify_e(#xmlElement{name=p} = E) -> % recurse
- Content = otp_xmlify_e(E#xmlElement.content),
- [E#xmlElement{content=Content}];
-otp_xmlify_e({p, Content1}) ->
- Content2 = otp_xmlify_e(Content1),
- [{p, Content2}];
-otp_xmlify_e(#xmlElement{name=ul} = E) ->
- Content = otp_xmlify_e(E#xmlElement.content),
- [E#xmlElement{content=Content}];
-otp_xmlify_e(#xmlElement{name=li} = E) ->
- %% Content may need to be made into <p>s etc.
- Content = otp_xmlify(E#xmlElement.content),
- [E#xmlElement{content=Content}];
-otp_xmlify_e(#xmlElement{name=dl} = E) ->
- Content0 = otp_xmlify_e(E#xmlElement.content),
- Content = otp_xmlify_dl(Content0),
- [E#xmlElement{content=Content}];
-otp_xmlify_e(#xmlElement{name=dt} = E) ->
- %% Special fix: Markers in <taglist> <tag>s are not allowed,
- %% save it using 'put' and place the marker first in the <item>
- %% instead
- Content = case E#xmlElement.content of
- [#xmlElement{name=a} = A] ->
- put(dt_marker, otp_xmlify_e(A)),
- otp_xmlify_e(A#xmlElement.content);
- _ ->
- otp_xmlify_e(E#xmlElement.content)
- end,
- [E#xmlElement{content=Content}];
-otp_xmlify_e(#xmlElement{name=dd} = E) ->
- %% Content may need to be made into <p>s etc.
- Content0 = otp_xmlify(E#xmlElement.content),
- Content = case get(dt_marker) of
- undefined -> Content0;
- [Marker] ->
- put(dt_marker, undefined),
- [Marker#xmlElement{content=[]}|Content0]
- end,
- [E#xmlElement{content=Content}];
-otp_xmlify_e(#xmlElement{name=tr} = E) ->
- Content = otp_xmlify_e(E#xmlElement.content),
- [E#xmlElement{content=Content}];
-otp_xmlify_e(#xmlElement{name=td} = E) ->
- Content = otp_xmlify_e(E#xmlElement.content),
- [E#xmlElement{content=Content}];
-otp_xmlify_e([E | Es]) ->
- otp_xmlify_e(E) ++ otp_xmlify_e(Es);
-otp_xmlify_e([]) ->
- [];
-otp_xmlify_e(E) ->
- [E].
-
-%%--Tags with special handling------------------------------------------
-
-%% otp_xmlify_a(A1) -> [A2]
-%% Takes an <a> element and filters the attributes to decide wheather
-%% its a seealso/url or a marker.
-%% In the case of a seealso/url, the href part is checked, making
-%% sure a .xml/.html file extension is removed (as DocBuilder inserts
-%% .html extension when resolving cross references).
-%% Also, references to other applications //App has a href attribute
-%% value "OTPROOT/..." (due to app_default being set to "OTPROOT" in
-%% docb_gen.erl), in this case both href attribute and content must be
-%% formatted correctly according to DocBuilder requirements.
-otp_xmlify_a(A) ->
- [Attr0] = filter_a_attrs(A#xmlElement.attributes),
- case Attr0 of
- #xmlAttribute{name=href, value=Href0} -> % seealso | url
- Content0 = text_only(A#xmlElement.content),
- {Href, Content} = otp_xmlify_a_href(Href0, Content0),
- [A#xmlElement{attributes=[Attr0#xmlAttribute{value=Href}],
- content=Content}];
- #xmlAttribute{name=name} -> % marker
- Content = otp_xmlify_e(A#xmlElement.content),
- [A#xmlElement{attributes=[Attr0], content=Content}]
- end.
-
-%% filter_a_attrs(Attrs) -> [Attr]
-%% Removes all attributes from a <a> element except the href or
-%% name attribute.
-filter_a_attrs([#xmlAttribute{name=href} = Attr | _Attrs]) ->
- [Attr];
-filter_a_attrs([#xmlAttribute{name=name} = Attr | _Attrs]) ->
- [Attr];
-filter_a_attrs([_Attr|Attrs]) ->
- filter_a_attrs(Attrs);
-filter_a_attrs([]) ->
- [].
-
-%% otp_xmlify_a_href(Href0, Es0) -> {Href1, Es1}
-%% Href = string()
-otp_xmlify_a_href("#"++_ = Marker, Es0) -> % <seealso marker="#what">
- {Marker, Es0};
-otp_xmlify_a_href("http:"++_ = URL, Es0) -> % external URL
- {URL, Es0};
-otp_xmlify_a_href("OTPROOT"++AppRef, Es0) -> % <.. marker="App:FileRef
- [AppS, "doc", FileRef1] = split(AppRef, "/"),
- FileRef = AppS++":"++otp_xmlify_a_fileref(FileRef1, AppS),
- [#xmlText{value=Str0} = T] = Es0,
- Str = case split(Str0, "/") of
- %% //Application
- [AppS2] ->
- %% AppS2 can differ from AppS
- %% Example: xmerl/XMerL
- AppS2;
- [_AppS,ModRef] ->
- case split(ModRef, ":") of
- %% //Application/Module
- [Module] ->
- Module++"(3)";
- %% //Application/Module:Type()
- [_Module,_Type] ->
- ModRef
- end;
- %% //Application/Module:Function/Arity
- [_AppS,ModFunc,Arity] ->
- ModFunc++"/"++Arity
- end,
- {FileRef, [T#xmlText{value=Str}]};
-otp_xmlify_a_href("../"++File, Es0) ->
- %% Special case: This kind of relative path is used on some
- %% places within i.e. EDoc and refers to a file within the same
- %% application tree.
- %% Correct the path according to the OTP directory structure
- {"../../"++File, Es0};
-otp_xmlify_a_href(FileRef1, Es0) -> % File within the same application
- FileRef2 = otp_xmlify_a_fileref(FileRef1, this),
- {FileRef2, Es0}.
-
-%% otp_xmlify_a_fileref(FileRef1, AppS|this) -> FileRef2
-%% AppS = FileRef = string()
-otp_xmlify_a_fileref(FileRef1, AppS) ->
- case split(FileRef1, ".#") of
-
- %% EDoc default name is "overview-summary.html,
- %% name of OTP User's Guide chapter is "chapter.xml"
- ["overview-summary", _Ext] ->
- "chapter";
- ["overview-summary", _Ext, Marker] ->
- "chapter#"++Marker;
-
- [File, Ext] when Ext=="xml";
- Ext=="html", AppS/=this ->
- File;
- [File, Ext, Marker0] ->
- %% Here is an awkward solution to an awkward problem
- %% The marker automatically inserted by DocBuilder at
- %% each function does not seem to work for EDoc generated
- %% ERLREFs.
- %% So if the referenced marker is in an ERLREF generated
- %% by EDoc, keep it "as is", ie "function-arity".
- %% If the referenced marker is NOT in an ERLREF generated
- %% by EDoc, the marker should be on the format
- %% "function/arity".
- %% The awkward part of the solution is to decide wheather
- %% the ERLREF is generated by EDoc or not: Here we make
- %% the decision based on which application the module
- %% belongs to -- which is ok when the module was written
- %% but probably not in the future...
- EDocApps = ["edoc","hipe","syntax_tools","xmerl"],
- IsEDocGenerated = lists:member(AppS, EDocApps),
- Marker = if
- %% The marker is in a file in *this*
- %% application (which documentation obviously
- %% is generated by EDoc), or it is in a file
- %% in an application which documentation
- %% is assumed to be generated by EDoc
- AppS==this; IsEDocGenerated ->
- Marker0;
-
- %% The marker is in a file in an application
- %% which documentation is assumed NOT to be
- %% generated by EDoc
- true ->
- case split(Marker0, "-") of
- [Func,Arity] ->
- Func++"/"++Arity;
- _ ->
- Marker0
- end
- end,
- if
- %% Ignore file extension in file reference if it either
- %% is ".xml" or if it is ".html" but AppS/=this, that
- %% is, we're resolving an OTPROOT file reference
- Ext=="xml";
- Ext=="html", AppS/=this ->
- File++"#"++Marker;
- true ->
- File++"."++Ext++"#"++Marker
- end;
-
- %% References to other files than XML files are kept as-is
- _ ->
- FileRef1
- end.
-
-%% otp_xmlify_blockquote(Es1) -> Es2
-%% Ensures that the content of a <blockquote> is divided into
-%% <p>s using the {p, Es} construct.
-otp_xmlify_blockquote([#xmlElement{name=p} = E|Es]) ->
- [E | otp_xmlify_blockquote(Es)];
-otp_xmlify_blockquote([#xmlText{} = E|Es]) ->
- {P, After} = find_p_ending(Es, []),
- [{p, [E|P]} | otp_xmlify_blockquote(After)];
-otp_xmlify_blockquote([]) ->
- [].
-
-%% otp_xmlify_code(E) -> Es
-%% Takes a <code> xmlElement and split it into a list of <code> and
-%% other xmlElements. Necessary when it contains more than a single
-%% xmlText element.
-%% Example:
-%% #xmlElement{name=code,
-%% content=[#xmlText{}, #xmlElement{name=br}, #xmlText{}]}
-%% =>
-%% [#xmlElement{name=code, content=[#xmlText{}]},
-%% #xmlElement{name=br},
-%% #xmlElement{name=code, content=[#xmlText{}]}]
-otp_xmlify_code(E) ->
- otp_xmlify_code(E, E#xmlElement.content, []).
-otp_xmlify_code(Code, [#xmlText{} = E|Es], Acc) ->
- otp_xmlify_code(Code, Es, [Code#xmlElement{content=[E]}|Acc]);
-otp_xmlify_code(Code, [#xmlElement{} = E|Es], Acc) ->
- otp_xmlify_code(Code, Es, [E|Acc]);
-otp_xmlify_code(_Code, [], Acc) ->
- lists:reverse(Acc).
-
-%% otp_xmlify_dl(Es1) -> Es2
-%% Insert empty <dd> elements if necessary.
-%% OTP DTDs does not allow <taglist>s with <tag>s but no <item>s.
-otp_xmlify_dl([#xmlElement{name=dt} = E|Es]) ->
- [E|otp_xmlify_dl(Es, E)];
-otp_xmlify_dl([E|Es]) ->
- [E|otp_xmlify_dl(Es)];
-otp_xmlify_dl([]) ->
- [].
-
-otp_xmlify_dl([#xmlElement{name=dd} = E|Es], _DT) ->
- [E|otp_xmlify_dl(Es)];
-otp_xmlify_dl([#xmlElement{name=dt} = E|Es], DT) ->
- DD = DT#xmlElement{name=dd, attributes=[], content=[]},
- [DD,E|otp_xmlify_dl(Es, E)];
-otp_xmlify_dl([E|Es], DT) ->
- [E|otp_xmlify_dl(Es, DT)];
-otp_xmlify_dl([], DT) ->
- DD = DT#xmlElement{name=dd, attributes=[], content=[]},
- [DD].
-
-%% otp_xmlify_table(Es1) -> Es2
-%% Transform <table> contents into "text", that is, inline elements.
-otp_xmlify_table([#xmlText{} = E|Es]) ->
- [E | otp_xmlify_table(Es)];
-otp_xmlify_table([#xmlElement{name=tbody} = E|Es]) ->
- otp_xmlify_table(E#xmlElement.content)++otp_xmlify_table(Es);
-otp_xmlify_table([#xmlElement{name=tr, content=Content}|Es]) ->
- %% Insert newlines between table rows
- otp_xmlify_table(Content)++[{br,[]}]++otp_xmlify_table(Es);
-otp_xmlify_table([#xmlElement{name=th, content=Content}|Es]) ->
- [{em, Content} | otp_xmlify_table(Es)];
-otp_xmlify_table([#xmlElement{name=td, content=Content}|Es]) ->
- otp_xmlify_e(Content) ++ otp_xmlify_table(Es);
-otp_xmlify_table([]) ->
- [].
-
-%%--Misc help functions used by otp_xmlify/1 et al---------------------
-
-%% find_next(Tag, Es) -> {Es1, Es2}
-%% Returns {Es1, Es2} where Es1 is the list of all elements up to (but
-%% not including) the first element with tag Tag in Es, and Es2
-%% is the remaining elements of Es.
-find_next(Tag, Es) ->
- find_next(Tag, Es, []).
-find_next(Tag, [#xmlElement{name=Tag} = E | Es], AccEs) ->
- {lists:reverse(AccEs), [E|Es]};
-find_next(Tag, [E|Es], AccEs) ->
- find_next(Tag, Es, [E|AccEs]);
-find_next(_Tag, [], AccEs) ->
- {lists:reverse(AccEs), []}.
-
-%% find_p_ending(Es, []) -> {Es1, Es2}
-%% Returns {Es1, Es2} where Es1 is the list of all elements up to (but
-%% not including) the first paragraph break in Es, and Es2 is
-%% the remaining elements of Es2.
-%% Paragraph break = <p> tag or empty line
-%% the next blank line, <p> or end-of-list as P, and the remaining
-%% elements of Es as After.
-find_p_ending([#xmlText{value="\n \n"++_} = E|Es], P) ->
- {lists:reverse(P), [E|Es]};
-find_p_ending([#xmlElement{name=p} = E|Es], P) ->
- {lists:reverse(P), [E|Es]};
-find_p_ending([E|Es], P) ->
- find_p_ending(Es, [E|P]);
-find_p_ending([], P) ->
- {lists:reverse(P), []}.
-
-%% is_paragraph(E | P) -> bool()
-%% P = {p, Es}
-is_paragraph(#xmlElement{name=p}) -> true;
-is_paragraph({p, _Es}) -> true;
-is_paragraph(_E) -> false.
-
-%% p_content(E | P) -> Es
-p_content(#xmlElement{content=Content}) -> Content;
-p_content({p, Content}) -> Content.
-
-%% is_empty(Str) -> bool()
-%% Str = string()
-%% Returns true if Str is empty in the sense that it contains nothing
-%% but spaces, tabs or newlines.
-is_empty("\n"++Str) ->
- is_empty(Str);
-is_empty(" "++Str) ->
- is_empty(Str);
-is_empty("\t"++Str) ->
- is_empty(Str);
-is_empty("") ->
- true;
-is_empty(_) ->
- false.
-
-%% split(Str, Seps) -> [Str]
-split(Str, Seps) ->
- split(Str, Seps, []).
-
-split([Ch|Str], Seps, Acc) ->
- case lists:member(Ch, Seps) of
- true -> split(Str, Seps, Acc);
- false -> split(Str, Seps, Acc, [Ch])
- end;
-split([], _Seps, Acc) ->
- lists:reverse(Acc).
-
-split([Ch|Str], Seps, Acc, Chs) ->
- case lists:member(Ch, Seps) of
- true -> split(Str, Seps, [lists:reverse(Chs)|Acc]);
- false -> split(Str, Seps, Acc, [Ch|Chs])
- end;
-split([], _Seps, Acc, Chs) ->
- lists:reverse([lists:reverse(Chs)|Acc]).
-
-%%--Functions for creating an erlref document---------------------------
-
-%% function_name(F) -> string()
-%% F = #xmlElement{name=function}
-%% Returns the name of a function as "name/arity".
-function_name(E) ->
- get_attrval(name, E) ++ "/" ++ get_attrval(arity, E).
-
-%% functions(Fs) -> Es
-%% Fs = [{Name, F}]
-%% Name = string() "name/arity"
-%% F = #xmlElement{name=function}
-functions(Fs) ->
- Es = lists:flatmap(fun ({Name, E}) -> function(Name, E) end, Fs),
- if
- Es==[] ->
- [];
- true ->
- {funcs, Es}
- end.
-
-function(_Name, E=#xmlElement{content = Es}) ->
- TypeSpec = get_content(typespec, Es),
- [?NL,{func, [ ?NL,
- {name,
- case funcheader(TypeSpec) of
- [] ->
- signature(get_content(args, Es),
- get_attrval(name, E));
- Spec -> Spec
- end
- },
- ?NL,{fsummary, fsummary(Es)},
- ?NL,local_types(TypeSpec),
- ?NL,{desc,
- label_anchor(E)++
- deprecated(Es)++
- fulldesc(Es)++
- seealso_function(Es)}
- ]}].
-
-fsummary([]) -> ["\s"];
-fsummary(Es) ->
- Desc = get_content(description, Es),
- case get_content(briefDescription, Desc) of
- [] ->
- fsummary_equiv(Es); % no description at all if no equiv
- ShortDesc ->
- text_only(ShortDesc)
- end.
-
-fsummary_equiv(Es) ->
- case get_content(equiv, Es) of
- [] -> ["\s"];
- Es1 ->
- case get_content(expr, Es1) of
- [] -> ["\s"];
- [Expr] ->
- ["Equivalent to ", Expr, ".",?NL]
- end
- end.
-
-label_anchor(E) ->
- case get_attrval(label, E) of
- "" -> [];
- Ref -> [{marker, [{id, Ref}],[]},?NL]
- end.
-
-label_anchor(Content, E) ->
- case get_attrval(label, E) of
- "" -> Content;
- Ref -> {p,[{marker, [{id, Ref}],[]},
- {em, Content}]}
- end.
-
-signature(Es, Name) ->
- [Name, "("] ++ seq(fun arg/1, Es) ++ [") -> term()", ?NL].
-
-arg(#xmlElement{content = Es}) ->
- [get_text(argName, Es)].
-
-funcheader([]) -> [];
-funcheader(Es) ->
- [t_name(get_elem(erlangName, Es))] ++ t_utype(get_elem(type, Es)).
-
-local_types([]) -> [];
-local_types(Es) ->
- local_defs2(get_elem(localdef, Es)).
-
-local_defs2([]) -> [];
-local_defs2(Es) ->
- {type,[?NL | [{v, localdef2(E)} || E <- Es]]}.
-
-%% Like localdef/1, but does not use label_anchor/2 -- we don't want any
-%% markers or em tags in <v> tag, plain text only!
-localdef2(#xmlElement{content = Es}) ->
- case get_elem(typevar, Es) of
- [] ->
- t_utype(get_elem(type, Es));
- [V] ->
- t_var(V) ++ [" = "] ++ t_utype(get_elem(type, Es))
- end.
-
-type_name(#xmlElement{content = Es}) ->
- t_name(get_elem(erlangName, get_content(typedef, Es))).
-
-types(Ts) ->
- Es = lists:flatmap(fun ({Name, E}) -> typedecl(Name, E) end, Ts),
- [?NL, {taglist,[?NL|Es]}].
-
-typedecl(Name, #xmlElement{content = Es}) ->
- TypedefEs = get_content(typedef, Es),
- Id = "type-"++Name,
- [{tag, typedef(TypedefEs)},
- ?NL,
- {item, [{marker,[{id,Id}],[]} |
- local_defs(get_elem(localdef, TypedefEs)) ++ fulldesc(Es)]},
- ?NL].
-
-typedef(Es) ->
- Name = ([t_name(get_elem(erlangName, Es)), "("]
- ++ seq(fun t_utype_elem/1, get_content(argtypes, Es), [")"])),
- case get_elem(type, Es) of
- [] ->
- [{tt, Name}];
- Type ->
- [{tt, Name ++ [" = "] ++ t_utype(Type)}]
- end.
-
-local_defs([]) -> [];
-local_defs(Es) ->
- [?NL, {ul, [{li, [{tt, localdef(E)}]} || E <- Es]}].
-
-localdef(E = #xmlElement{content = Es}) ->
- Var = case get_elem(typevar, Es) of
- [] ->
- [label_anchor(t_abstype(get_content(abstype, Es)), E)];
- [V] ->
- t_var(V)
- end,
- Var ++ [" = "] ++ t_utype(get_elem(type, Es)).
-
-deprecated(Es) ->
- case get_content(deprecated, Es) of
- [] -> [];
- DeprEs ->
- Es2 = get_content(fullDescription,
- get_content(description, DeprEs)),
- Es3 = otp_xmlify_e(Es2),
- [{p, [{em, ["This function is deprecated: "]} |Es3]}, ?NL]
- end.
-
-fulldesc(Es) ->
- case get_content(fullDescription, get_content(description, Es)) of
- [] ->
- index_desc(Es);
- Desc ->
- [?NL|otp_xmlify(Desc)] ++ [?NL]
- end.
-
-index_desc(Es) ->
- Desc = get_content(description, Es),
- case get_content(briefDescription, Desc) of
- [] ->
- equiv(Es); % no description at all if no equiv
- ShortDesc ->
- ShortDesc
- end.
-
-seealso_module(Es) ->
- case get_elem(see, Es) of
- [] -> [];
- Es1 ->
- {section,[{title,["See also"]},{p,seq(fun see/1, Es1, [])}]}
- end.
-seealso_function(Es) ->
- case get_elem(see, Es) of
- [] -> [];
- Es1 ->
- [{p, [{em, ["See also:"]}, " "] ++ seq(fun see/1, Es1, ["."])},
- ?NL]
- end.
-
-%% ELEMENT see PCDATA
-%% ATTLIST name PCDATA
-%% href PCDATA
-see(#xmlElement{content=Es0} = E) ->
- Href0 = get_attrval(href, E),
- {Href, Es} = otp_xmlify_a_href(Href0, Es0),
- [{seealso, [{marker, Href}], Es}].
-
-equiv(Es) ->
- case get_content(equiv, Es) of
- [] -> ["\s"];
- Es1 ->
- case get_content(expr, Es1) of
- [] -> [];
- [Expr] ->
- Expr1 = [Expr],
- Expr2 = case get_elem(see, Es1) of
- [] ->
- {c,Expr1};
- [E=#xmlElement{}] ->
- case get_attrval(href, E) of
- "" ->
- {c,Expr1};
- Ref ->
- {seealso, [{marker, Ref}], Expr1}
- end
- end,
- [{p, ["Equivalent to ", Expr2, "."]}, ?NL]
- end
- end.
-
-authors(Es) ->
- case get_elem(author, Es) of
- [] ->
- [?NL,{aname,["\s"]},?NL,{email,["\s"]}];
- Es1 ->
- [?NL|seq(fun author/1, Es1, "", [])]
- end.
-
-author(E=#xmlElement{}) ->
- Name = case get_attrval(name, E) of
- [] -> "\s";
- N -> N
- end,
- Mail = case get_attrval(email, E) of
- [] -> "\s";
- M -> M
- end,
- [?NL,{aname,[Name]},?NL,{email,[Mail]}].
-
-t_name([E]) ->
- N = get_attrval(name, E),
- case get_attrval(module, E) of
- "" -> N;
- M ->
- S = M ++ ":" ++ N,
- case get_attrval(app, E) of
- "" -> S;
- A -> "//" ++ A ++ "/" ++ S
- end
- end.
-
-t_utype([E]) ->
- t_utype_elem(E).
-
-t_utype_elem(E=#xmlElement{content = Es}) ->
- case get_attrval(name, E) of
- "" -> t_type(Es);
- Name ->
- T = t_type(Es),
- case T of
- [Name] -> T; % avoid generating "Foo::Foo"
- T -> [Name] ++ ["::"] ++ T
- end
- end.
-
-t_type([E=#xmlElement{name = typevar}]) ->
- t_var(E);
-t_type([E=#xmlElement{name = atom}]) ->
- t_atom(E);
-t_type([E=#xmlElement{name = integer}]) ->
- t_integer(E);
-t_type([E=#xmlElement{name = float}]) ->
- t_float(E);
-t_type([#xmlElement{name = nil}]) ->
- t_nil();
-t_type([#xmlElement{name = list, content = Es}]) ->
- t_list(Es);
-t_type([#xmlElement{name = tuple, content = Es}]) ->
- t_tuple(Es);
-t_type([#xmlElement{name = 'fun', content = Es}]) ->
- t_fun(Es);
-t_type([#xmlElement{name = abstype, content = Es}]) ->
- t_abstype(Es);
-t_type([#xmlElement{name = union, content = Es}]) ->
- t_union(Es);
-t_type([#xmlElement{name = record, content = Es}]) ->
- t_record(Es).
-
-t_var(E) ->
- [get_attrval(name, E)].
-
-t_atom(E) ->
- [get_attrval(value, E)].
-
-t_integer(E) ->
- [get_attrval(value, E)].
-
-t_float(E) ->
- [get_attrval(value, E)].
-
-t_nil() ->
- ["[]"].
-
-t_list(Es) ->
- ["["] ++ t_utype(get_elem(type, Es)) ++ ["]"].
-
-t_tuple(Es) ->
- ["{"] ++ seq(fun t_utype_elem/1, Es, ["}"]).
-
-t_fun(Es) ->
- ["("] ++ seq(fun t_utype_elem/1, get_content(argtypes, Es),
- [") -> "] ++ t_utype(get_elem(type, Es))).
-
-t_record([E|Es]) ->
- ["#", get_attrval(value, E), "{"++ seq(fun t_field/1, Es) ++"}"].
-t_field(#xmlElement{name=field, content=[Atom,Type]}) ->
- [get_attrval(value, Atom), "="] ++ t_utype_elem(Type).
-
-t_abstype(Es) ->
- case split_at_colon(t_name(get_elem(erlangName, Es)),[]) of
- {Mod,Type} ->
- [Type, "("] ++
- seq(fun t_utype_elem/1, get_elem(type, Es), [")"]) ++
- [" (see module ", Mod, ")"];
- Type ->
- [Type, "("] ++
- seq(fun t_utype_elem/1, get_elem(type, Es), [")"])
- end.
-
-%% Split at one colon, but not at two (or more)
-split_at_colon([$:,$:|_]=Rest,Acc) ->
- lists:reverse(Acc)++Rest;
-split_at_colon([$:|Type],Acc) ->
- {lists:reverse(Acc),Type};
-split_at_colon([Char|Rest],Acc) ->
- split_at_colon(Rest,[Char|Acc]);
-split_at_colon([],Acc) ->
- lists:reverse(Acc).
-
-t_union(Es) ->
- seq(fun t_utype_elem/1, Es, " | ", []).
-
-%% seq(Fun, Es)
-%% seq(Fun, Es, Tail)
-%% seq(Fun, Es, Sep, Tail) -> [string()]
-%% Fun = function(E) -> [string()]
-%% Sep = string()
-%% Tail = [string()]
-%% Applies Fun to each element E in Es and return the appended list of
-%% strings, separated by Sep which defaults to ", " and ended by Tail
-%% which defaults to [].
-seq(Fun, Es) ->
- seq(Fun, Es, []).
-seq(Fun, Es, Tail) ->
- seq(Fun, Es, ", ", Tail).
-seq(Fun, [E], _Sep, Tail) ->
- Fun(E) ++ Tail;
-seq(Fun, [E | Es], Sep, Tail) ->
- Fun(E) ++ [Sep] ++ seq(Fun, Es, Sep, Tail);
-seq(_Fun, [], _Sep, Tail) ->
- Tail.
-
-%%--Misc functions for accessing fields etc-----------------------------
-
-%% Type definitions used below:
-%% E = #xmlElement{} | #xmlText{}
-%% Es = [E]
-%% Tag = atom(), XHTML tag
-%% Name = atom(), XHTML attribute name
-%% Attrs = [#xmlAttribute{}]
-%% Ts = [#xmlText{}]
-
-%% parent(E) -> module | overview
-parent(E) ->
- Parents = E#xmlElement.parents,
- {Parent,_} = lists:last(Parents),
- Parent.
-
-%% get_elem(Tag, Es1) -> Es2
-%% Returns a list of all elements in Es which have the name Tag.
-get_elem(Name, [#xmlElement{name = Name} = E | Es]) ->
- [E | get_elem(Name, Es)];
-get_elem(Name, [_ | Es]) ->
- get_elem(Name, Es);
-get_elem(_, []) ->
- [].
-
-%% get_attr(Name, Attrs1) -> Attrs2
-%% Returns a list of all attributes in Attrs1 which have the name Name.
-get_attr(Name, [#xmlAttribute{name = Name} = A | As]) ->
- [A | get_attr(Name, As)];
-get_attr(Name, [_ | As]) ->
- get_attr(Name, As);
-get_attr(_, []) ->
- [].
-
-%% get_attrval(Name, E) -> string()
-%% If E has one attribute with name Name, return its value, otherwise ""
-get_attrval(Name, #xmlElement{attributes = As}) ->
- case get_attr(Name, As) of
- [#xmlAttribute{value = V}] ->
- V;
- [] -> ""
- end.
-
-%% get_content(Tag, Es1) -> Es2
-%% If there is one element in Es1 with name Tag, returns its contents,
-%% otherwise []
-get_content(Name, Es) ->
- case get_elem(Name, Es) of
- [#xmlElement{content = Es1}] ->
- Es1;
- [] -> []
- end.
-
-%% get_text(Tag, Es) -> string()
-%% If there is one element in Es with name Tag, and its content is
-%% a single xmlText, return the value of this xmlText.
-%% Otherwise return "".
-get_text(Name, Es) ->
- case get_content(Name, Es) of
- [#xmlText{value = Text}] ->
- Text;
- [] -> ""
- end.
-
-%% get_text(E) -> string()
-%% Return the value of an single xmlText which is the content of E,
-%% possibly recursively.
-get_text(#xmlElement{content=[#xmlText{value=Text}]}) ->
- Text;
-get_text(#xmlElement{content=[E]}) ->
- get_text(E).
-
-%% text_only(Es) -> Ts
-%% Takes a list of xmlElement and xmlText and return a lists of xmlText.
-text_only([#xmlElement{content = Content}|Es]) ->
- text_only(Content) ++ text_only(Es);
-text_only([#xmlText{} = E |Es]) ->
- [E | text_only(Es)];
-text_only([]) ->
- [].