diff --git a/src/lib/xml.go b/src/lib/xml.go
new file mode 100644
index 00000000000..3b4277878b5
--- /dev/null
+++ b/src/lib/xml.go
@@ -0,0 +1,426 @@
+// Copyright 2009 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// NOTE(rsc): Actually, this package is just a description
+// of an implementation that hasn't been written yet.
+
+// This package implements an XML parser but relies on
+// clients to implement the parsing actions.
+
+// An XML document is a single XML element.
+//
+// An XML element is either a start tag and an end tag,
+// like ..., or a combined start/end tag .
+// The latter is identical in semantics to ,
+// and this parser does not distinguish them.
+//
+// The start (or combined start/end) tag can have
+// name="value" attributes inside the angle brackets after
+// the tag name, as in 
.
+// Names are drawn from a fixed set of alphabetic letters;
+// Values are strings quoted with single or double quotes.
+//
+// An element made up of distinct start and end tags can
+// contain free-form text and other elements inside it,
+// as in Google
+// or Google.
+// The former is an element with the text "Google" inside it.
+// The latter is a element with that element inside it.
+// In general, an element can contain a sequence of elements
+// and text inside it. In XML, white space inside an element is
+// always counted as text--it is never discarded by the parser.
+// XML parsers do translate \r and \r\n into \n in text.
+//
+// This parser reads an XML document and calls methods on a
+// Builder interface object in response to the text.
+// It calls the builder's StartElement, Text, and EndElement
+// methods, mimicking the structure of the text.
+// For example, the simple XML document:
+//
+//
+// 
+//
+//
+// results in the following sequence of builder calls:
+//
+// StartElement("a", []Attr(Attr("href", "http://www.google.com")));
+// Text("\n\t");
+// StartElement("img", []Attr(Attr("src", "http://www.google.com/icon.png"),
+// Attr("alt", "Google")));
+// EndElement("img");
+// Text("\n");
+// StartElement("br", []Attr());
+// EndElement("br");
+// EndElement("a");
+//
+// There are, of course, a few more details, but the story so far
+// should be enough for the majority of uses. The details are:
+//
+// * XML documents typically begin with an XML declaration line like
+// .
+// This line is strongly recommended, but not strictly required.
+// It introduces the XML version and text encoding for the rest
+// of the file. XML parsers are required to recognize UTF-8 and
+// UTF-16. This parser only recognizes UTF-8 (for now?).
+//
+// * After the XML declaration comes an optional doctype declaration like
+//
+// The parser should pass this information on to the client in some
+// form, but does not. It discards such lines.
+//
+// * The XML declaration line is an instance of a more general tag
+// called a processing instruction, XML's #pragma. The general form is
+// , where target is a name (like "xml") specifying
+// the intended recipient of the instruction, and text is the
+// instruction itself. This XML parser keeps the declaration
+// to itself but passes along other processing instructions using
+// the ProcInst method. Processing instructions can appear anywhere
+// in an XML document. Most clients will simply ignore them.
+//
+// * An XML comment can appear anywhere in an XML document.
+// Comments have the form . The XML parser passes
+// them along by calling the Comment method. Again, most clients
+// will simply ignore them.
+//
+// * Text inside an XML element must be escaped to avoid looking like
+// a start/end tag. Specifically, the characters < and & must be
+// written as < and &. An alternate quoting mechanism is to
+// use the construct . The quoted text ... can contain
+// < characters, but not the sequence ]]>. Ampersands must still be
+// escaped. For some reason, the existence of the CDATA quoting mechanism
+// infects the processing of ordinary unquoted text, which is not allowed
+// to contain the literal sequence ]]>. Instead, it would be written
+// escaped, as in ]]>. The parser hides all these considerations
+// from the library client -- it reports all text, regardless of original
+// form and already unescaped, using the Text method.
+//
+// * A revision to XML 1.0 introduced the concept of name spaces
+// for attribute and tag names. A start tag with an attribute
+// xmlns:prefix="URL" introduces `prefix' as a shorthand
+// for the name space whose identifier is URL. Inside the element
+// with that start tag, an element name or attribute prefix:foo
+// (as in ) is understood to refer
+// to name `foo' in the name space denoted by `URL'. Although
+// this is a shorthand, there is no canonical expansion. Thus:
+//
+//
+// text1
+// text2
+//
+//
+// and
+//
+//
+// text1
+// text2
+//
+//
+// are equivalent XML documents, and there is no canonical form.
+//
+// The special attribute xmlns="URL" sets the default name space
+// for unprefixed tags (but not attribute names) to URL.
+// Thus:
+//
+//
+// text1
+// text2
+//
+//
+// is another XML document equivalent to the first two, and
+//
+//
+// text1
+// text2
+//
+//
+// would be equivalent, except that `attr' in attr="value" has no
+// associated name space, in contrast to the previous three where it
+// is in the http://google.com/bar name space.
+//
+// The XML parser hides these details from the client by passing
+// a Name struct (ns + name pair) for tag and attribute names.
+// Tags and attributes without a name space have ns == "".
+//
+// References:
+// Annotated XML spec: http://www.xml.com/axml/testaxml.htm
+// XML name spaces: http://www.w3.org/TR/REC-xml-names/
+
+package xml
+
+import (
+ "io";
+ "os";
+)
+
+// XML name, annotated with name space URL
+type Name struct {
+ ns, name string;
+}
+
+// XML attribute (name=value).
+type Attr struct {
+ name Name;
+ value string;
+}
+
+// XML Builder - methods client provides to Parser.
+// Parser calls methods on builder as it reads and parses XML.
+// If a builder method returns an error, the parse stops.
+type Builder interface {
+ // Called when an element starts.
+ // Attr is list of attributes given in the tag.
+ //
+ //
+ // xmlns and xmlns:foo attributes are handled internally
+ // and not passed through to StartElement.
+ StartElement(name Name, attr []Attr) *os.Error;
+
+ // Called when an element ends.
+ //
+ //
+ EndElement(name Name) *os.Error;
+
+ // Called for non-empty character data string inside element.
+ // Can be called multiple times between elements.
+ // text
+ //
+ Text(text []byte) *os.Error;
+
+ // Called when a comment is found in the XML.
+ //
+ Comment(text []byte) *os.Error;
+
+ // Called for a processing instruction
+ //
+ ProcInst(target string, text []byte) *os.Error;
+}
+
+// Default builder. Implements no-op Builder methods.
+// Embed this in your own Builders to handle the calls
+// you don't care about (e.g., Comment, ProcInst).
+type BaseBuilder struct {
+}
+
+func (b *BaseBuilder) StartElement(name Name, attr []Attr) *os.Error {
+ return nil;
+}
+
+func (b *BaseBuilder) EndElement(name Name) *os.Error {
+ return nil;
+}
+
+func (b *BaseBuilder) Text(text []byte) *os.Error {
+ return nil;
+}
+
+func (b *BaseBuilder) Comment(text []byte) *os.Error {
+ return nil;
+}
+
+func (b *BaseBuilder) ProcInst(target string, text []byte) *os.Error {
+ return nil;
+}
+
+// XML Parser. Calls Builder methods as it parses.
+func Parse(r io.Read, b Builder) *os.Error {
+ return os.NewError("unimplemented");
+}
+
+// Channel interface to XML parser: create a new channel,
+// go ParseTokens(r, c), and then read from the channel
+// until TokenEnd. This variant has the benefit that
+// the process reading the channel can be a recursive
+// function instead of a set of callbacks, but it has the
+// drawback that the channel interface cannot signal an
+// error to cause the parser to stop early.
+
+// An XML parsing token.
+const (
+ TokenStartElement = 1 + iota;
+ TokenEndElement;
+ TokenText;
+ TokenComment;
+ TokenProcInst;
+ TokenEnd;
+)
+
+type Token struct {
+ Kind int; // TokenStartElement, TokenEndElement, etc.
+ Name Name; // name (TokenStartElement, TokenEndElement)
+ Attr []Attr; // attributes (TokenStartElement)
+ Target string; // target (TokenProcessingInstruction)
+ Text []byte; // text (TokenCharData, TokenComment, etc.)
+ Err *os.Error; // error (TokenEnd)
+}
+
+type ChanBuilder chan Token;
+
+func (c ChanBuilder) StartElement(name Name, attr []Attr) *os.Error {
+ var t Token;
+ t.Kind = TokenStartElement;
+ t.Name = name;
+ t.Attr = attr;
+ c <- t;
+ return nil;
+}
+
+func (c ChanBuilder) EndElement(name Name) *os.Error {
+ var t Token;
+ t.Kind = TokenEndElement;
+ t.Name = name;
+ c <- t;
+ return nil;
+}
+
+func (c ChanBuilder) Text(text []byte) *os.Error {
+ var t Token;
+ t.Kind = TokenText;
+ t.Text = text;
+ c <- t;
+ return nil;
+}
+
+func (c ChanBuilder) Comment(text []byte) *os.Error {
+ var t Token;
+ t.Kind = TokenComment;
+ t.Text = text;
+ c <- t;
+ return nil;
+}
+
+func (c ChanBuilder) ProcInst(target string, text []byte) *os.Error {
+ var t Token;
+ t.Kind = TokenProcInst;
+ t.Target = target;
+ t.Text = text;
+ c <- t;
+ return nil;
+}
+
+func ParseToChan(r io.Read, c chan Token) {
+ var t Token;
+ t.Kind = TokenEnd;
+ t.Err = Parse(r, ChanBuilder(c));
+ c <- t;
+}
+
+
+// scribbled notes based on XML spec.
+
+// document is
+// xml decl?
+// doctype decl?
+// element
+//
+// if xml decl is present, must be first. after that,
+// can have comments and procinsts scattered throughout,
+// even after the element is done.
+//
+// xml decl is:
+//
+// <\?xml version='[a-zA-Z0-9_.:\-]+'( encoding='[A-Za-z][A-Za-z0-9._\-]*')?
+// ( standalone='(yes|no)')? ?\?>
+//
+// spaces denote [ \r\t\n]+.
+// written with '' above but can use "" too.
+//
+// doctype decl might as well be ]*>
+//
+// procinst is <\?name( .*?)\?>. name cannot be [Xx][Mm][Ll].
+//
+// comment is .
+//
+// tags are:
+// start tag
+// combined start/end tag
+// end tag
+// (the " ?" is an optional space, not a literal question mark.)
+//
+// plain text is [^<&]* except cannot contain "]]>".
+// can also have escaped characters:
+// &#[0-9]+;
+// &#x[0-9A-Fa-f]+;
+// &name;
+//
+// can use to avoid escaping < characters.
+//
+// must rewrite \r and \r\n into \n in text.
+//
+// names are Unicode. valid chars listed below.
+//
+// attrib is name="value" or name='value'.
+// can have spaces around =.
+// attribute value text is [^<&"]* for appropriate ".
+// can also use the &...; escape sequences above.
+// cannot use .
+//
+// xmlns attributes are name=value where name has form xmlns:name
+// (i.e., xmlns:123 is not okay, because 123 is not a name; xmlns:a123 is ok).
+// sub-name must not start with : either.
+//
+// name is first(second)*.
+//
+// first is
+//
+// 003A 04D0-04EB 0A59-0A5C 0C35-0C39 0F49-0F69 1E00-1E9B
+// 0041-005A 04EE-04F5 0A5E 0C60-0C61 10A0-10C5 1EA0-1EF9
+// 005F 04F8-04F9 0A72-0A74 0C85-0C8C 10D0-10F6 1F00-1F15
+// 0061-007A 0531-0556 0A85-0A8B 0C8E-0C90 1100 1F18-1F1D
+// 00C0-00D6 0559 0A8D 0C92-0CA8 1102-1103 1F20-1F45
+// 00D8-00F6 0561-0586 0A8F-0A91 0CAA-0CB3 1105-1107 1F48-1F4D
+// 00F8-00FF 05D0-05EA 0A93-0AA8 0CB5-0CB9 1109 1F50-1F57
+// 0100-0131 05F0-05F2 0AAA-0AB0 0CDE 110B-110C 1F59
+// 0134-013E 0621-063A 0AB2-0AB3 0CE0-0CE1 110E-1112 1F5B
+// 0141-0148 0641-064A 0AB5-0AB9 0D05-0D0C 113C 1F5D
+// 014A-017E 0671-06B7 0ABD 0D0E-0D10 113E 1F5F-1F7D
+// 0180-01C3 06BA-06BE 0AE0 0D12-0D28 1140 1F80-1FB4
+// 01CD-01F0 06C0-06CE 0B05-0B0C 0D2A-0D39 114C 1FB6-1FBC
+// 01F4-01F5 06D0-06D3 0B0F-0B10 0D60-0D61 114E 1FBE
+// 01FA-0217 06D5 0B13-0B28 0E01-0E2E 1150 1FC2-1FC4
+// 0250-02A8 06E5-06E6 0B2A-0B30 0E30 1154-1155 1FC6-1FCC
+// 02BB-02C1 0905-0939 0B32-0B33 0E32-0E33 1159 1FD0-1FD3
+// 0386 093D 0B36-0B39 0E40-0E45 115F-1161 1FD6-1FDB
+// 0388-038A 0958-0961 0B3D 0E81-0E82 1163 1FE0-1FEC
+// 038C 0985-098C 0B5C-0B5D 0E84 1165 1FF2-1FF4
+// 038E-03A1 098F-0990 0B5F-0B61 0E87-0E88 1167 1FF6-1FFC
+// 03A3-03CE 0993-09A8 0B85-0B8A 0E8A 1169 2126
+// 03D0-03D6 09AA-09B0 0B8E-0B90 0E8D 116D-116E 212A-212B
+// 03DA 09B2 0B92-0B95 0E94-0E97 1172-1173 212E
+// 03DC 09B6-09B9 0B99-0B9A 0E99-0E9F 1175 2180-2182
+// 03DE 09DC-09DD 0B9C 0EA1-0EA3 119E 3007
+// 03E0 09DF-09E1 0B9E-0B9F 0EA5 11A8 3021-3029
+// 03E2-03F3 09F0-09F1 0BA3-0BA4 0EA7 11AB 3041-3094
+// 0401-040C 0A05-0A0A 0BA8-0BAA 0EAA-0EAB 11AE-11AF 30A1-30FA
+// 040E-044F 0A0F-0A10 0BAE-0BB5 0EAD-0EAE 11B7-11B8 3105-312C
+// 0451-045C 0A13-0A28 0BB7-0BB9 0EB0 11BA 4E00-9FA5
+// 045E-0481 0A2A-0A30 0C05-0C0C 0EB2-0EB3 11BC-11C2 AC00-D7A3
+// 0490-04C4 0A32-0A33 0C0E-0C10 0EBD 11EB
+// 04C7-04C8 0A35-0A36 0C12-0C28 0EC0-0EC4 11F0
+// 04CB-04CC 0A38-0A39 0C2A-0C33 0F40-0F47 11F9
+//
+// second is first plus
+//
+// 002D 06DD-06DF 09E6-09EF 0B56-0B57 0D3E-0D43 0F3E
+// 002E 06E0-06E4 0A02 0B66-0B6F 0D46-0D48 0F3F
+// 0030-0039 06E7-06E8 0A3C 0B82-0B83 0D4A-0D4D 0F71-0F84
+// 00B7 06EA-06ED 0A3E 0BBE-0BC2 0D57 0F86-0F8B
+// 02D0 06F0-06F9 0A3F 0BC6-0BC8 0D66-0D6F 0F90-0F95
+// 02D1 0901-0903 0A40-0A42 0BCA-0BCD 0E31 0F97
+// 0300-0345 093C 0A47-0A48 0BD7 0E34-0E3A 0F99-0FAD
+// 0360-0361 093E-094C 0A4B-0A4D 0BE7-0BEF 0E46 0FB1-0FB7
+// 0387 094D 0A66-0A6F 0C01-0C03 0E47-0E4E 0FB9
+// 0483-0486 0951-0954 0A70-0A71 0C3E-0C44 0E50-0E59 20D0-20DC
+// 0591-05A1 0962-0963 0A81-0A83 0C46-0C48 0EB1 20E1
+// 05A3-05B9 0966-096F 0ABC 0C4A-0C4D 0EB4-0EB9 3005
+// 05BB-05BD 0981-0983 0ABE-0AC5 0C55-0C56 0EBB-0EBC 302A-302F
+// 05BF 09BC 0AC7-0AC9 0C66-0C6F 0EC6 3031-3035
+// 05C1-05C2 09BE 0ACB-0ACD 0C82-0C83 0EC8-0ECD 3099
+// 05C4 09BF 0AE6-0AEF 0CBE-0CC4 0ED0-0ED9 309A
+// 0640 09C0-09C4 0B01-0B03 0CC6-0CC8 0F18-0F19 309D-309E
+// 064B-0652 09C7-09C8 0B3C 0CCA-0CCD 0F20-0F29 30FC-30FE
+// 0660-0669 09CB-09CD 0B3E-0B43 0CD5-0CD6 0F35
+// 0670 09D7 0B47-0B48 0CE6-0CEF 0F37
+// 06D6-06DC 09E2-09E3 0B4B-0B4D 0D02-0D03 0F39
+