Stephen McConnell wrote:
-----Original Message-----
From: Scott Stirling [mailto:[EMAIL PROTECTED]
[snip]
All of these protocols include support for the establishment of a
locally cached file that is a copy of the remote resource (partly
driven by the need to handle the relationship between models of
projects and Ant's build-time requirements). Typically a
link such as link:jar:acme/widget#1.2 (a symbolic link defining
a jar file under the group namespace acme, name widget, with a
published version of 1.2) could reference an artifact such as
'artifact:jar:net/osm/demo#1.2.9', which in turn maps to a range
of possible sources where each source defines its own scheme for
artifact to path transformation (i.e. the automation of the
transformation of the artifact specification to something like
file://d:/dpml/data/cache/net/osm/jars/demo-1.2.9.jar is
an example of one of potentially multiple source paths).
What you have in the [above] paragraph is very REST-ful in its
way of encoding location, version and name in URLs. I would
ask us to consider (Occam's Razor in hand) at what point one
actully needs to create a custom protocol (and code to
support it) versus using REST-ful URL conventions to achieve
much of the same thing with existing protocols and software.
The generalization here makes me feel a touch REST-less as the scope of
concerns are significantly different. The paper deals with the www as the
initial blackboard whereas the examples provided here focus primarily with
normalization within the confines of a single physical hardware device
within a context of a specific language API. In effect the Java language
via the protocol handler framework provides an important extension point.
DPML's collection of protocols are (in the context of this discussion)
nothing more than an example of the application of that API in dealing with
basic development and runtime issues in larger scale project scenarios.
This difference in scope an example that demonstrates an area where existing
protocols and software are insufficient.
If you look at 1060research.com's NetKernel (disclaimer; friends of
mine, its an ex-HPLabs project), you can see what loosely coupled REST
designs can do in-process. Netkernel is an XML processing
language/runtime that uses URIs extensively to refer to other things in
the app, be they database selects, remote web pages, invocations of
other functions, and then uses caching to decide which to re-evaluate
versus serve up.
to do this they've had to build everything from the ground up, and have
a nice (but complex) arch to do it.
If we retrofit stuff into Java, with Java URL handling, then some URL
like a reference to a file on the classpath, it would have to be turned
into a string that included a classloader reference:
classpathresource://classloader/03455he/resource/org/example/something
To turn it into a URL that represented the infinite random stream of a
seeded random number generator, loaded as an antlib, I would be scared
to think where to begin.
Yet I do know how to have an Ant resource to do this
<ex:random length="412" seed="28" />
I can also imagine Ivy references fitting in here
<ivy:artifact project="org.apache.log4j" artifact="log4j"
version="1.2.13" />
Or even the output of an pipe hooked up to the output of a running java
program
<pipe:output refid="running.junit" />
I dont see that URLs fit in well to all these things.
My plans @ work post-ant1.7 would be to resource-enable our ant tasks so
that we could feed the output of stuff into wherever resources are used,
e.g
<sf:resolved "HOST 192.168.2.1:console:main:iconized" />
to get back whatever resolve to the path console:main:iconized on a
remote host.
Seems to me that there is a lot of fun to be had with Resource without
worrying too much about REST-enabling ant's core.
Who else is doing stuff with resources?
The key thing is that these protocols enable the removal
of location concerns by shifting the focus to identity.
Another take on the [above] statement is that location is a
concern to make explicit to the reader of a build script,
rather than obscuring a thing's actual location
When we look at an individual codebase (and I'm referring here to a single
codebase delivering a small number of discrete deliverables) the usage of
location information (relative files locations and selection patterns) is
totally appropriate. However, when codebase A references content in codebase
B via location we are introducing a dependency not only on a foreign
codebase location, but more importantly we are adding information about a
foreign build strategy (via paths into a foreign structure). If you
consider a scenario where you are working with 10s or 100s of projects - the
usage of explicit location (and the implicit leaking of foreign build
strategies into the consuming project has (in my experience) been a major
source of failure and contributes dramatically to build system maintenance
overhead. At the end of the day the movement towards identity versus
locality is an example of the application of 'inversion of control'
principals in that an individual build (or runtime) solution should not need
to know where something comes from (which is a management concern) but
instead should deal with identifying the things it needs.
inside of a protocol handler black's box.
I would agree that protocol handling is an area of the JVM API a less well
known (but I would disagree with the black-box metaphor). In the context of
this thread a protocol handler is simply a standard factory for standard
connections. The clarity of the behaviour of the respective protocol
handler is directly related to its implementation and documentation.
Just some thoughts on this interesting topic. I don't buy the
need for all these protocols and handlers or the associated
flexibility in Ant (yet).
From my point of view the support for the association of a resource with a
configured URL opens up some significant opportunities for down-scaling
management applications (and here I am thinking specifically about products
such as the Depot build-system). The downscaling would be possible simply
as a result of the use of common data elements - specifically URL instances
that could flow in bother directions (from Depot to ant during project
configuration, and from Ant to Depot during project execution).
URL protocol handlers are simple standard structures and support within Ant
for these structures would have a significant positive impact on the forward
development of Depot.
Cheers, Steve.
--------------------------
Stephen McConnell
mailto:[EMAIL PROTECTED]
http://www.dpml.net
Scott Stirling
Framingham, MA
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED] For
additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
