[Date Prev][Date Next][Thread Prev][Thread Next][Author Index][Date Index][Thread Index]
First Class Links Make Second Strike
- To: <xanatech>
- Subject: First Class Links Make Second Strike
- From: Marc Stiegler <marcs>
- Date: Thu, 19 Oct 89 15:23:58 PDT
So much for my hope that a fact forum could lead to a truce on
first class and embedded links.
I have ordered the bombers into the air, SAC Posture 7. The missiles
are standing by on DefCon 1 :-)
But before launching my final terrible attack on embedded links,
I would like to try one more time to find the heart of our disagreement,
and reach reconciliation. I am suspicious that there are two
goals, dean, that you and I share. What we disagree on is the
utility of embedded links for achieving those goals.
Terminology: I use the term sensor here. I believe I use it correctly--there
is an alert sensor attached to a recorder, and we all recognize
that the recorder is the thing actually attached to the document
(and the sensor is the thing actually attached to the user :-)
Goal 1: We must find a really powerful way of displaying, selecting,
and going to the "right" bert context when the user traverses
a link. Here, the term "right" means, "the context the user wanted
to go to, whether he knew it or not." Embedded links, since they
are copied along with the rest of the document, give you a visible
choice of bert...sometimes.
Unfortunately, whether the link is embedded or not, FINDING THE
RIGHT CONTEXT IS STILL AN UNSOLVED PROBLEM. In the scenario I
created to ask questions earlier, we had 4 berts, A1-A4, and
3 links, to A1, A3, and A4. A2 didn't have a link embedded in
it because it had branched off before we made the link.
Question for the reader: Was A2 left without an embedded link
because of the careful, methodical planning of the creators of
A1 and A2 and A3? Or was it left without a link because the author
didn't know about B until after A2 had spun off, or because the
author forgot that B was there until A2 spun off, or because
B was created after A2 spun off? If the cause were any of the
reasons except Cause Number 1 above (Number 1 being the careful
intentional planning of the author, the cause which I honestly
find the least likely), A2 is just as interesting as an endset
as A3 and A4.
We have to find a solution to the fundamental problem of picking
the right context for the user. I am suspicious that this is
a frontend function, one way or the other. The best solution
I have yet heard is dean's idea for going to the most recently
used context, i.e., if the user already has a window open on
one of the berts at the end of the link, go to that one. This
is not a complete solution, though I believe it is a part of
a complete solution.
But embedding the link is not the answer to finding the right
context--it depends too much on successful preplanning by the
Goal 2: Get alerted when someone improves on a document that
we have reason to believe will be interesting when improved.
Since duplicating a document with embedded links causes new links
to be created, sensors will ring, and readers will get alerted
to the presence of a new version.
Unfortunately, if a document has 5 links embedded in it, and
it is in my area of interest, I will probably attach recorders
to every document it is linked to. If a new version of the document
is created, I don't want 5 sensors to go off--I want one, and
I want exactly one.
The correct answer to this desire is a recorder that will tell
me if the bert hops, and a recorder that will tell me if a new
bert acquires a part of the document. We already have recorders
for those, right? This is a solved problem, is it not?
(below are a couple of accidental, low yield nuke bursts; this
is what happens when you go to DefCon 1, accidents happen :-)
Even if recording on vcopying is not currently solved, 5 sensors
ringing is not the right answer. Rather than face that kind of
noise (which you, dean, claim I would have earned for having
a promiscuous sensor), you can be sure that I would make my sensor
less promiscuous--I would say, "hey, filter it out and don't
bother me if that guy makes a vcopy of an existing link". At
that point, I get NO rings on a new version. This is in no one's
interest, though as a reader, it's better than the alternatives
Please note also that the filter criteria I described above,
to filter out the duplicates, is the only one that will work:
by definition, the vcopy of a link that passes my filters will
itself pass my filters. It has a type that is interesting, and
it has an author whom I used to think was interesting, at least
until he started reusing his documents, creating slightly customized
alternates of them for slightly different situations (the fiend!
How dare he? :-). Since the "new" link passes all my filters,
it has perfect "camouflage".
But because the "new" link is attached to material that I have
already read (albeit in a new context), it is of lower quality
and lower value to me than a really new link, to really new material
(after all, if I wanted to know about changes in the context
of the material surrounding the link, I would have attached a
version-stimulated recorder and sensor to the document). I want
to be alerted to really new things with a level of intensity
far beyond my interest in being alerted to slightly different
Dean, you say "We invented embedded links because first class
links broke in so many unpredictable and surprising ways." To
quote you out of context :-),
"Got a good example?" When I look at embedded links, I see something
that breaks every time someone embeds a link after the original
document has started spinning off variants.
Speaking of examples, here is an example of a situation requiring
a first class link, i.e., a link that follows the permissions
of neither the beginning nor ending document: Suppose I want
to mail to you a link to my mailbox, in effect sending you my
mailbox address. The mailboxes are private. So a link embedded
in my mailbox would be invisible to you, and a link embedded
in your mailbox cannot be created by me.
Second example: How can I find your mailbox, when neither of
us has the other's mailbox? This is even more amusing than how
I send you my mailbox once I've got yours. We would probably
go to a public document with a list of mailboxes. And therein
lies a tale:
You don't want this mail list to be publicly writeable,
you just want everyone to attach links to it. So to get my mailbox
listed, I create an intermediary document with embedded links
to both the mailbox and the public mail list--in other words,
I create a first class link from the mailbox to the mail list
by hand. Except it's not as good as a real first class link for
the reader--because, of course, the link in the public mail list
is not actually the link that mail senders want to use. Oh no,
the link in the mail list is only a link to the document that
contains the mail link!
As an aside, it took me some considerable thought to figure out
how to trade mailboxes at all with embedded links--the average
manager and secretary wouldn't have a chance. There are subtleties
here that will require system administrators, just to trade mail
On the other hand, if we give those managers and secretaries
first class links, they just run the link and set the permission
(permit themselves as authors, everyone as readers, the intuitive
thing to do); we do the rest for them.
(end of mild nuke bursts)
Is part of the difference a difference in the viewpoint between
the reader and the writer? Dean, when you said, "I want the link
to version with my comment" you were thinking from a writer's
point of view, wanting to make sure the readers all got the message
that something new had happened. Looking at it from the reader's
point of view, I say, well I SOMETIMES want to know something
new has happened, but I want to know the nature of the newness--and
since the link is a duplicate, it is not a good indicator of
what is new. The newness indicator should indicate what is new,
not what is duplicated.
I suspect we may be looking at different size documents, too.
I worry about large documents with lots of links and many versions.
If you think primarily about small documents with few links and
few versions, the issues and choices all get gentler.
Getting very, very philosophical, I go back to an underpinning
of the Xanadu religion: the reader must have control over what
he sees. Only this way can one have a pool that is 99% sludge
while every reader sees a view of a pool that is 99% gold. I
see the automatic creation of camouflaged links via document
vcopying as taking that control out of the reader's hands and
giving it to the author. To paraphrase dean again, no, no, no.
Dean, if you aren't overwhelmed by my reasoning here, we probably
need higher bandwidth, i.e., let's nuke it out in the office.