Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
RE: [eclipse-incubator-e4-dev] [resources] EFS IFileTree

The advantage of the Directory Stream approach is that clients can
see / work on intermediate results before the traversal is complete,
and they can cancel the Stream retrieval if the intermediate
results don't match their expectations.
I agree that programmatic approaches like visitors or arbitrary
coded filters are problematic when the receiver of the query
is remote (Java does support things like remote object execution,
but not all server-sides support Java...and local evaluation of
the filter might not always be a good fallback). LDAP filters
might be an interesting thing to explore, I haven't heard about
these before but anything Standard seems like a good thing
to look at.
Here is a different idea that might also work:
* Initiate
Martin Oberhuber, Senior Member of Technical Staff, Wind River
Target Management Project Lead, DSDP PMC Member

From: eclipse-incubator-e4-dev-bounces@xxxxxxxxxxx [mailto:eclipse-incubator-e4-dev-bounces@xxxxxxxxxxx] On Behalf Of John Arthorne
Sent: Monday, October 20, 2008 9:12 PM
To: E4 developer list
Subject: RE: [eclipse-incubator-e4-dev] [resources] EFS IFileTree

IFileTree doesn't have a depth because it was really intended to optimize the DEPTH_INFINITE case. For DEPTH_ZERO or DEPTH_ONE, it doesn't provide much advantage. I agree though that it could be interesting to support different scoping to pick up smaller subtrees, or to have some termination condition. The purpose was really to cut down on round-trips, so I'm not sure you'd get the same effect with a visitor or directory stream approach. Perhaps filters could be passed in a serialized form (like an LDAP filter), so that it can be interpreted on the machine where the file system lives, to avoid passing unnecessary data over the wire.


"Oberhuber, Martin" <Martin.Oberhuber@xxxxxxxxxxxxx>
Sent by: eclipse-incubator-e4-dev-bounces@xxxxxxxxxxx

10/20/2008 12:49 PM

Please respond to
E4 developer list <eclipse-incubator-e4-dev@xxxxxxxxxxx>

"E4 developer list" <eclipse-incubator-e4-dev@xxxxxxxxxxx>
RE: [eclipse-incubator-e4-dev] [resources] EFS IFileTree

Hi John,
One disadvantage of the IFileTree today is that the file tree to be
returned must terminate at some point, and not the slightest
amount of information is returned if it does not terminate.
This makes it unusable for file systems that can be endless
(like the Internet), or even very very large with the user only
being interested in parts of the file tree. Even refreshLocal
performs a breadth-first-search and can be limited by the
depth to visit. fetchFileTree() does not support that.
One option for improving this situation could be if the
fetchFileTree() method would not return a snapshot like
the IFileTree, but rather return a Stream of IFileStore / IFileInfo
objects that the client can cancel at any time. A
breadth-first-search algorithm for returning these nodes might
be preferred, though I'd also see some potential for allowing
arbitrary ordering of returned nodes with some concept of
allowing a node to be "incomplete" e.g. a folder node with
not all children returned yet.
In JSR 203, the Path#newDirectoryStream() [1] and
Files#walkFileTree() [2] methods go in this direction.
Martin Oberhuber, Senior Member of Technical Staff, Wind River
Target Management Project Lead, DSDP PMC Member

From: eclipse-incubator-e4-dev-bounces@xxxxxxxxxxx [mailto:eclipse-incubator-e4-dev-bounces@xxxxxxxxxxx] On Behalf Of John Arthorne
Friday, October 17, 2008 7:40 PM
E4 developer list
[eclipse-incubator-e4-dev] [resources] EFS IFileTree

I forgot to mention this during the meeting, so I just wanted to mention a little know EFS interface called IFileTree. The idea of this interface was to allow for batched interaction with an entire file sub-tree. This prevents the large number of round-trips needed when a client needs to walk over an entire subtree of a slow/remote file system. It allows you to get a snapshot of an entire remote tree state in a single round-trip. In some experiments we did back in the day, this dramatically sped up certain operations like refreshLocal over a high latency remote tree. I mention it only because it's probably an under-exploited concept that could perhaps be expanded upon to improve performance in remote resource scenarios. I could imagine expanding the idea to allow a client to queue up a whole batch of file changes, that could be fired off in a single round-trip to the remote system for processing.

eclipse-incubator-e4-dev mailing list

Back to the top