svn commit: r1396329 - /subversion/branches/http-dynamic-prop-namespaces/BRANCH-README

[cm...@apache.org]

Author: cmpilato
Date: Tue Oct  9 21:12:27 2012
New Revision: 1396329

URL: http://svn.apache.org/viewvc?rev=1396329&view=rev
Log:
On the 'http-dynamic-prop-namespaces' branch, add BRANCH-README.
  

Added:
    subversion/branches/http-dynamic-prop-namespaces/BRANCH-README

Added: subversion/branches/http-dynamic-prop-namespaces/BRANCH-README
URL: http://svn.apache.org/viewvc/subversion/branches/http-dynamic-prop-namespaces/BRANCH-README?rev=1396329&view=auto
==============================================================================
--- subversion/branches/http-dynamic-prop-namespaces/BRANCH-README (added)
+++ subversion/branches/http-dynamic-prop-namespaces/BRANCH-README Tue Oct  9 21:12:27 2012
@@ -0,0 +1,108 @@
+This branch is a playground for one proposed solution to Subversion's
+currently inability to support property names which have colon
+characters in them without generating invalid XML in its WebDAV
+PROPFIND and PROPPATCH requests and responses.  See
+http://subversion.tigris.org/issues/show_bug.cgi?id=1971 for more
+information.
+
+
+BACKGROUND INFORMATION
+======================
+
+The WebDAV specification dictates that in PROPFIND and PROPPATCH
+requests and responses, property names are carried as XML tags.  XML
+tag names have requirements which limit the valid characters used in
+the name.  Subversion tries to ensure that its property names employ
+only character whichs are valid in XML "Names", but there's one area in
+particular where this protection isn't enough to ensure that valid XML
+is generated in these requests and responses: colon characters in
+property names.
+
+Subversion makes use of colon-carrying property names itself
+("svn:log", "svn:date", "svn:ignore", etc.).  A single colon character
+in an XML tag is permissible, but the meaning differs from what
+Subversion requires.  In a PROPFIND XML response, "<svn:date>" doesn't
+mean "the 'svn:date' property", it means "the property 'date' in the
+XML namespace whose prefix is "svn".  But if no XML namespace has been
+defined to have such a prefix, that XML is invalid.
+
+Even a property name which doesn't have a colon can present a problem,
+as a user could choose a property name that, when represented as an
+XML tag, has some other meaning.  And certainly, properties with
+multiple colons are problematic, as "<svn:foo:bar>" is invalid.
+
+Subversion attempts to compensate for some of these problems.  It
+defines namespaces for Subversion properties ("svn:*") and
+user-created properties (which are assumed to have no colons in their
+names).  Subversion then uses those defined namespaces and prefixes in
+the PROPFIND and PROPPATCH requests and responses.  Thus, the property
+"svn:log" is represented as "<S:log>", and that's okay because
+elsewhere in the request/response there is XML which maps the prefix
+"S" to a static namespace URI which Subversion clients all agree means
+"the 'svn:' namespace".  A property such as "foo" is carried as
+"<C:foo>", with a mapping of the prefix "C" to another static
+namespace URI which means (by Subversion convention) "a bare,
+user-generated property".
+
+But this compensation only works to an extent.  The property
+"svn:foo:bar" is carried as "<S:foo:bar>"; the property "custom:prop"
+is carried as "<C:custom:prop>".  Both are invalid XML.
+
+
+PROPOSED SOLUTION
+=================
+
+The solution proposed which this branch attempts to implement is to
+dynamically pre-examine all the properties which need to be referred
+to in a given PROPFIND or PROPPATCH request/response, and generate
+namespace mappings for every unique property prefix using a namespace
+URI which extends from a fixed base namespace URI.
+
+Here's an example.  Say the server is answering a request for all the
+properties on a given resource.  And say those properties have names
+as follows:
+
+   svn:ignore
+   svn:config:auto-props
+   svn:config:ignore
+   last-tested
+   cvs2svn:orig-rev
+
+Subversion can determine that there are three unique prefixes:
+
+   svn:
+   svn:config:
+   cvs2svn:
+
+For historical reasons, "svn:" properties will continue to use their
+existing mapping.  The bare property "last-tested" will do so, too.
+So the list of required XML namespace mappings is:
+
+   S = $(SVN_DAV_PROP_NS_SVN)
+   C = $(SVN_DAV_PROP_NS_CUSTOM)
+   svn0 = $(SVN_DAV_PROP_NS_EXTENSIBLE)/svn:config
+   svn1 = $(SVN_DAV_PROP_NS_EXTENSIBLE)/cvs2svn
+
+(where SVN_DAV_PROP_NS_EXTENSIBLE is a new URI).  Given those
+mappings, the properies are transmitted like so:
+
+   <S:ignore>
+   <svn0:auto-props>
+   <svn0:ignore>
+   <C:last-tested>
+   <svn1:orig-rev>
+
+On the receiving end, Subversion's XML parsers will match the tag
+namespace prefixes to the namespace URIs.  Property tags which make
+use of the existing "Subversion" and "custom" namespace URIs will be
+treated as they are today, with "svn:" prepended to the tag names of
+the former and the latter taken as bare property names.  However, for
+property tags which have a namespace URI that is a child of the fixed
+extensible base URI, Subversion prepends the URI-decoded relative path
+that remains when you skip past that base URI in the namespace URI.
+
+
+OUTLOOK
+=======
+
+This whole thing could be doomed to utter failure.  --cmpilato
\ No newline at end of file

Re: svn commit: r1396329 - /subversion/branches/http-dynamic-prop-namespaces/BRANCH-README

[Daniel Shahaf <d....@daniel.shahaf.name>]

cmpilato@apache.org wrote on Tue, Oct 09, 2012 at 21:12:27 -0000:
> +the name.  Subversion tries to ensure that its property names employ
> +only character whichs are valid in XML "Names", but there's one area in
...
> +   svn0 = $(SVN_DAV_PROP_NS_EXTENSIBLE)/svn:config

It seems that the obvious next step, if this approach works out, would
be to relax the requirement on property names from "the ASCII subset of
XML Name" towards "Anything that's valid after the slash in
$(SVN_DAV_PROP_NS_EXTENSIBLE)/".

(Maybe not _anything_, but a subset of that which is still a superset of
"ASCII XML Name".)

Re: svn commit: r1396329 - /subversion/branches/http-dynamic-prop-namespaces/BRANCH-README

[Daniel Shahaf <d....@daniel.shahaf.name>]

cmpilato@apache.org wrote on Tue, Oct 09, 2012 at 21:12:27 -0000:
> +the name.  Subversion tries to ensure that its property names employ
> +only character whichs are valid in XML "Names", but there's one area in
...
> +   svn0 = $(SVN_DAV_PROP_NS_EXTENSIBLE)/svn:config

It seems that the obvious next step, if this approach works out, would
be to relax the requirement on property names from "the ASCII subset of
XML Name" towards "Anything that's valid after the slash in
$(SVN_DAV_PROP_NS_EXTENSIBLE)/".

(Maybe not _anything_, but a subset of that which is still a superset of
"ASCII XML Name".)