Xanadu Server Top Ten Issues

[Marc Stiegler <marcs>]

Herewith is a new, improved list of server issue highlights.  This list 
should not be compared too closely with the last list of Top 
Ten Trauma, because they have significantly different roots: 
the old Top Ten was a list abstracted from all issues, whereas 
this list focuses 
solely on backend issues. So they really are apples and oranges, 
though they all come from the same database.

The total list has grown a little as I have collected more issues, 
both from old issue lists and new issues as they appeared. There 
used to be 274;, and now there are 284.

However, many of those 284 are now in the "resolved" category. 
For those who remember from the last discussion how I keep score 
(with an urgency rating composed of importance, troublesomeness, 
difficulty, and timing), we now have a batting average of .488 
on the most critical part of the list: of the 43 original non-display 
pre-beta issues of urgency 12 or stronger, we have resolved 21. 
So we're doing well, I think, by just doing.

Indeed, when seeking really nasty issues to fill out a Top Ten 
list, it turns out we're doing surprisingly well: I had to scrounge 
to find worrisome problems, and even after scrounging, the problems 
I came up with just weren't all that fierce.  The most worrisome 
issue here, archiving, is worrisome because we haven't thought 
about it as much as we need to (something which we've known all 
along :-)


The Current Server Top Ten:

-- Archiving: Yes, we are still postponing the heavy-duty thinking 
about all the issues associated with archiving. 

-- Deleting: There are 2 kinds of issues with deletion, frontend 
issues and backend issues. The frontend issues are much scarier 
than the backend issues; since this list is focused on backend 
stuff, deletion really doesn't qualify as a major problem: the 
way Xanadu works, this falls into the same box as Archiving, 
and solving archiving with deletion in mind will solve deletion. 
But I thought I'd list it for joel's piece of mind, so he'd know 
we haven't forgotten :-) 

-- Mapping document types onto applications that can read/write 
them: Suppose I'm sitting in Tapestry, and I traverse a link 
to a document that I don't understand. Do I call up AutoCAD to 
display it? Or Montage? Would both of them be able to work with 
it--and if so, which do I have, and if I have both, which do 
I prefer? 

Actually, this is going to be pretty solvable, but it needs to 
be solved, and we haven't thought it through yet. This is both 
harder and more important for Xanadu to solve than it has ever 
been for Apple or Microsoft, because we hope to work in heterogeneous 
networks, where the reader can't possibly have the same application 
that the author used (unless the author used AutoCAD, of course, 
which runs on all the interesting platforms :-)

-- If we put a sensor on just the bert, we need to be able to 
detect a link attached to one of the document's paragraphs even 
though the link is attached in the context of a different bert. 
Markm and Dean once thought they had the answer to this, but 
not any more (guys, let us get together to discuss this, shall 
we? :-)

-- If we change the specification for a filter, all the sensors 
already using that filter should find themselves using the new 
specification. From the user's perspective, this would mean that 
if he changed his "Refutations Only" filter so that John's refutations 
were ignored (because John complains about everything, I now 
realize after having installed a thousand sensors using this 
filter), all the sensors using the "Refutations Only" filter 
would also now ignore John.

Alas, as it stands now, we would have to destroy all the old 
sensors and create new ones to support this capability for the 
user. 

This is actually tolerable for Release 1.0. It may even be tolerable 
forever, if patterns of use follow the model I currently expect, 
based on my ignorance. But it deserves a moment's thought to 
see if we know how to upgrade for Release 2 or 3. The current 
situation puts a significant premium on getting the filter right 
the first time--potentially unpleasant in an evolving information 
pool.

-----------

As you can tell, I still can't count: the last Top Ten list had 
12 items. This  Top Ten list has 5.

 So, hey, count yourselves lucky :-) 

--marcs

P.S. This list is already obsolete: ravi and markm and dean have 
solved the problem with sensors attached to berts.