[DAS] LDAS/DasClient problems

[Lincoln Stein]

Hi Thomas,

Now it makes sense.  I thought you were asking for the annotation servers to 
distinguish between SEGMENTERROR and LANDMARKNOTANNOTATED.

How do you feel about calling the latter tag <SEGMENTUNKNOWN> (to be issued 
by annotation servers only), since it is completely noncommittal about 
whether the segment is an error or just unannotated.

So I'm updating the LDAS server to do the following:

	1) <SEGMENTERROR> for reference servers
	2) <SEGMENTUNKNOWN> for annotation servers
	3) pass the "id" attribute in the features request 
		to look up a feature by its ID
		- can return several segments if the feature spans
			them
	4) the HTTP header to contain some capabilities information:
		x-das-capabilities:  multiple-segments; segment-status; xff-format; 
				feature-by-id

How do you feel about modifying XFF slightly in order to normalize groups?  I 
would love to be able to give exons unique IDs and then to explicitly assign 
them to different splice patterns, rather than duplicating them within the 
document.

With regards to Dmitry's original problem, I can either produce an 
intermediate version of LDAS that will give empty segments when a landmark is 
not recognized so that it will work with Dazzle right now, or I can release a 
version with the above modifications.  This will require coordinate changes 
in Dazzle, Bio::DAS, and any other client libs floating around.  Or you can 
patch Dazzle to accept LDAS's silent omission of missing landmarks.  Which do 
you prefer?

Lincoln

On Tuesday 05 February 2002 03:52, Thomas Down wrote:
> On Mon, Feb 04, 2002 at 03:11:50PM -0500, Lincoln Stein wrote:
> > If the lightweight server doesn't know about a particular landmark, how
> > can it tell the difference between an incorrect landmark and an
> > unannotated one?  Or are you thinking that the lightweight server will
> > know enough about the assembly to have the names of all the landmarks,
> > but not necessarily their lengths and versions?
>
> No -- I think I must have missed something out of the argument.
> Reference servers should never be returning LANDMARKNOTANNOTATED,
> since they (by definition) know at least basic information about
> every landmark in the reference sequence set.  If they get a
> request for an unknown segment, that is, indeed, a SEGMENTERROR,
> since it shows that a data integrity bug has occurred somewhere.
>
> Annotation servers are in a rather different position.  They
> may only know about a relatively small proportion of the
> landmarks in the full sequence set -- some people want
> annotation servers which just cover a few clones.  The point
> of LANDMARKNOTANNOTATED is that it allows annotation servers
> to simply deny all knowledge of a given landmark ID, while
> not actually raising an error.
>
> I guess it's possibly to make do with just a SEGMENTERROR
> state, but for the sake of robustness I'd much prefer to
> keep error conditions separate from conditions which occur
> during normal operation.
>
> Does this make any kind of sense?
>
> > I've already bumped LDAS up to version 1.5 and the spec will follow as
> > soon as I'm sure it's all implementable.
>
> Okay, that's fine.
>
>    Thomas.

-- 
========================================================================
Lincoln D. Stein                           Cold Spring Harbor Laboratory
lstein@cshl.org			                  Cold Spring Harbor, NY
========================================================================