W3C home > Mailing lists > Public > w3c-dist-auth@w3.org > January to March 2000

Why IE's Web Folders are accessed through File/Open

From: Yaron Goland <yarong@Exchange.Microsoft.com>
Date: Wed, 2 Feb 2000 17:07:58 -0800
Message-ID: <7DE119D3D0E15543874F7561EECBDBED02619E3F@BEG.platinum.corp.microsoft.com>
To: "'Joe Orton'" <joe@orton.demon.co.uk>, w3c-dist-auth@w3.org
O.k., I admit it, I am the person responsible for the awful file/open check
box UI in Web Folders in IE. It is all my fault. Throw rotten tomatoes my
way.

All of us at MS tried to come up with a better solution. The ways to improve
Web Folders in IE fell into two types - 
1) Enable people, only by typing in a URL to tell the system that they want
a PROPFIND view
2) Put in a switch to allow people, once they are at a site, to flip between
GET/PROPFIND

We were, unfortunately, not able to achieve either of these goals for the
reasons I will outline below. This is why we ended up with the file/open UI.
It was the best thing we could come up with. 

There are however, two features that we did put in. The first makes it
possible in IE 5.0 to markup a HTML anchor tag so that when clicked it will
do a PROPFIND rather than a GET. The beauty of this feature is that if the
HTML is displayed on a non-IE 5.0 system then the extra text will be ignored
and the anchor will work as a normal GET anchor. Please see
http://msdn.microsoft.com/workshop/author/behaviors/overview/webfolder.asp
for details.
	The second is that when you have navigated to a Web Folder you can
use the favorites menu to create a favorite to it. That way, in the future,
you can get back to the site through favorites rather than file/open. Note
that this favorite doesn't work across machines so if you send it to a
friend it probably won't work. This feature is actually a side effect of how
Web Folders is implemented.

Our first goal, to enable people to just type in a URL and have that URL
navigate to a web folder, failed because we couldn't find a reliable way to
tell if a URL should be resolved as a PROPFIND or a GET. 
	The first trick we tried was to say that if a URL ended in a "/"
then we would try to do a PROPFIND. If the PROPFIND failed then we would do
a GET. The problem is that http://foo.com/bar/, for example, is a legal name
and something a user would very likely want to do a GET on. Can you imagine
their confusion when they get some weird icon filled page rather than their
normal homepage? We could hear the support phone lines ringing. 
	The next suggestion was to use a "." at the end of a HTTP URL as an
indicator that one should navigate to a PROPFIND. Everyone liked this one a
bit better because even though a HTTP URL could technically end in a "." it
is sort of traditional to treat "." as meaning "this directory". Note, this
is not the standard. As defined in RFC 2396 the character "." only has
special meaning in the context of a relative URI. Of course it didn't hurt
that IIS doesn't allow resources to have the name ".". Unfortunately we had
to abandon this idea as well because it turns out that Office strips "." at
the end of HTTP URLs thinking that they are periods at the end of a
sentence. This was necessary to prevent confusion when users typed a HTTP
URL at the end of a sentence.
	There is, btw, a way to get around this parsing. If you enclose your
HTTP URL in "<" ">" then Office will still recognize the HTTP URL and thus
highlight it but won't strip the "." at the end. However this still
presented the problem that a user could type a HTTP URL into IE such as
http://foo.com/bar/. and then copy/paste it into an e-mail message and their
friend at the other end would click on it and not get the same experience
because Outlook would strip the ".". Again, we could hear the support phone
lines ringing.
	We gave some thought to perhaps using an illegal HTTP URL character
to indicate that one should do a PROPFIND instead of a GET but it seemed
like a truly awful idea to depend on illegal behavior. This is exactly the
sort of thing that destroys interoperability and annihilates processing
paths. So this idea was rejected.
	The last idea to be considered, which I admit responsibility for,
was to introduce a new type of URL, dav:. Any time we saw a dav: URL we
would do a PROPFIND. Ignoring the obvious conflict this would introduce with
WebDAV's use of the dav: URL (we could have used webdav:) everyone really
hated the idea of having two different schemes for the same protocol. Nobody
liked https, for example. While I wasn't enamored with this idea I felt it
was practical. However people were very concerned about what would happen
when someone sent a friend a webdav: URL and nothing happened when the
friend (on a non-MS system or using older MS software) clicked on it. For
example, what was liked about the first two proposals was that on down level
systems one still got the GET behavior. This was considered a critical
feature in order to enable a smooth transition between the WebDAV and
non-WebDAV worlds. Also people really hated the idea of introducing yet
another URL type. Users barely are able to deal with HTTP URLs. Asking them
to figure out when to use HTTP and when to use DAV just seemed to be too
much. Also the IE UI guys went absolutely insane at the mere suggestion that
a DAV: URL could appear in the address bar. They too could hear the support
phone lines ringing as people demanded to know what the hell a DAV URL was
and where the HTTP URL had gone to. So this idea was also killed off.
	The real issue is that a URL points at something but tells you
nothing about how to interact with it. It is a noun without a verb. The
reason we were running into so much trouble was that we were trying to
verbify the noun. What we really need is a two-part system. XML could do a
good job here. One could use a piece of XML to say things like "Go to this
URL and perform this Action". Of course having users type in XML was too
much to ask. We could, of course, try to put in a helper UI (say a wizard)
but even that was a bit much to ask. The general feeling was that we should
be trying to have users type in less URLs rather than making entering URLs
even more complex. 
	I talked with our usability people about this problem and they told
me that according to their lab tests the majority of users could not type in
a URL at all. They couldn't use the address bar. They couldn't see an Ad
with a URL then type that URL into their browser. All these users can do is
click on URLs, not type them in. This has lead to situations where
sysadmins, if they want to tell a user their file share, will just send an
e-mail with a link to the file share. The user will then live out of that
e-mail. Every time they want to get to their files they will find that
e-mail, open it and click on the link. This is why it is so critical to have
home pages with lots of links, up front and easily available. Users won't
type in links. So once someone sends them a homepage in e-mail they will
continue to go back to that homepage to find everything. So clearly trying
to solve the problem by inventing URL tricks wasn't going to buy us much in
the case of the majority of our users. I still wanted to at least do the
DAV: URL, if only for power users and other early adopters, but I didn't
have the time or resources left to deliver the feature.

One idea that was considered was enhancing the right click menu on HTTP URLs
to allow the user to select GET or PROPFIND behavior. The idea is that you
could type in a HTTP URL, right click on it and then select how you wanted
the HTTP URL resolved. However adding this to the right click menu was
non-trivial. Still, it does set up the stage for the second set of proposals
- a switch.

The original argument was that we should add an option to IE's view menu to
allow the user to switch to a PROPFIND view. The problem with this proposal
is that 99.9% of the time for the lifespan of IE 5.0 the resource would not
be a WebDAV resource. This meant that we would have a feature in the menu
that didn't work most of the time. Anyone but me hearing the support phone
lines ringing? (Yes, I know, I should have a check up for these ringing
sounds =) So the idea of just putting a "directory" view choice on the view
menu was ruled out. 
	However we decided to be smart. What if we knew that the resource
that was currently displayed was a WebDAV resource? That is, what if the
user typed in http://www.webdav.org and we already knew that
http://www.webdav.org was a WebDAV resource? Everyone agreed that if we knew
that a resource was a webdav resource then we could make the menu item
appear. We figured what we could do was to let the user type in the URL,
press return/go and navigate to the resource. Then, in the background, after
the file had completely loaded (so there would be no performance penalty) we
would do a background OPTIONS request to see if we were dealing with a
WebDAV resource. If we were then we would do a PROPFIND on resourcetype and
see if it was a collection. If it was then we would make the menu item
appear.
	The theory was o.k. but the implementation would be awful. First of
all, this means that every time a user was just randomly browsing around we
would issue at least one request (OPTIONS) and sometimes two (PROPFIND).
Even assuming we figured out a caching scheme (anyone know how long to cache
an OPTIONS output? How about how long to cache that a resource is a
collection?) to reduce the load the implications were still frightening. I
could see the headlines "IE 5.0 single handedly takes out the Internet, news
at 11." However, believe it or not, there was an even worse problem, from
our point of view.
	Pretend our normal user Missy is sitting in front of her computer.
Her admin has repeated to her, three times, that she is supposed to type in
http://company/files/missy into the address bar. Then she is supposed to go
to view and select "Directory View".
	So, Missy, types in http://company/files/missy, presses return, got
to the view menu and... well... nothing. There is no "Directory View"
option. First the page has to finish displaying before we will try OPTIONS.
Then the amount of time it takes for the OPTIONS message to get there and
back is variable. Then, assuming http://company/files/missy is a WebDAV
resource, we still have to make the PROPFIND request. This means that for
some indeterminate and largely inconsistent period of time the "Directory
View" menu option wouldn't appear. The user experience was just too awful to
contemplate. We couldn't implement a randomly appearing menu item.
	Another suggestion was to try and make some use of the largely
useless go button. The reason the go button was introduced was that users
did not seem to understand the idea that one typed a URL into the address
bar and then pressed return. It was the "then press return" part that got
them. However other browsers had put in some sort of button that users could
press to start the navigation. We got so much feedback from system
administrators and others that their users were having this problem that we
felt we had to put in the go button. Nobody liked it but oh well. So we
figured at least we could use the stupid thing. We thought we could put in a
drop down menu off that button that the user could use to select GET or
PROPFIND behavior.
	However this suggestion suffered from the same problem that the
previous suggestion suffered from, 99.9% of the time the PROPFIND selection
wouldn't do anything useful.
	There was thought given to having a switch somewhere in advanced
that would just add a "directory" switch to the view menu or add the drop
down menu to the GO button but then the feature would only be useful to
super power users. While we have found that it is often useful to enable
super power users the IE WebDAV project at the time was incredibly resource
constrained. I couldn't justify the effort for a power user feature. 

So, in the end, the best idea we could come up with was file/open. Of course
the file/open switch suffers from the same sort of problems we identified
previously, 99.9% of the time the damn thing wouldn't work right. A small
improvement is that if the attempt to navigate to the resource as a
FrontPage/WebDAV resource fails then we will put up an error and ask the
user if they would like to see the page as a normal web page (GET). However,
under normal circumstances, this wouldn't be considered acceptable. Having a
feature that 99.9% of the time produces an error isn't an acceptable user
experience.
	The reason we were allowed to use file/open is that absolutely
nobody ever uses file/open. So the UI folks weren't worried about having
this funky "Open as Web Folder" switch in file/open. In fact the UI folk
wanted to remove file/open all together from IE but the MS UI regulations
require it to be there. So they said we could throw in the switch there. In
other words, we were allowed to produce a bad user experience because we
were counting on the majority of users never, ever, finding the switch
without being explicitly told it was there. Don't you wish I was making this
all up?

In the future I hope we are able to throw in a WebDAV switch as a power user
feature. I normally would have driven this into IE 5.5 but I left IE a year
or so ago and my new job took my attention away from WebDAV.

Obviously the considerations that went into IE's design don't necessarily
apply to all software products. Microsoft makes its money by selling
software to the average user. The needs of the average user are very
different from high end users. If you are making a high end user product
then obviously your considerations will be very different. I often find it
difficult to explain to IETFers why MS does what it does because people at
the IETF are most definitely NOT our target audience. Normal users think
very differently from technical users. Things that seem painfully obvious or
easy to technical users present insurmountable barriers to normal users.
Normal users do not want to understand how these systems work. They just
want the damn product to work. That means that our first priority has to be
to avoid making things complex. If a UI element isn't painfully obvious then
it is cut or hidden away for power users. That's the theory, anyway.

			Yaron

> -----Original Message-----
> From: Joe Orton [mailto:joe@orton.demon.co.uk]
> Sent: Tue, January 25, 2000 3:20 AM
> To: w3c-dist-auth@w3.org
> Subject: Re: Comments please!
> 
> 
> > Hmm. But this doesn't solve my problem. Let's say that my 
> website is called
> > 'http://www.foo.com'. If I do a
> > get on this url, I'll get 'http://www.foo.com/index.htm' by 
> default. But
> > 'http://www.foo.com' is also a directory at the to level. 
> So if a user types
> > 'http://www.foo.com' I don't know if I should perform a Get 
> or a Propfind,
> > both are valid. The user types the same thing for directory 
> browsing in
> > 'http://www.foo.com' and for getting the default page.
> 
> I think this mostly a UI problem. From the user's point of 
> view, have two "views" 
> on a collection resource; it's GET response, and the listing 
> of collection members
> from a PROPFIND response. These correspond to the traditional 
> "web browser" mode of
> operation, and the traditional "file browser/explorer" mode 
> of operation.
> 
> Like you say, in Web Folders you have the "Open as Web 
> folder" checkbox on open
> dialog, which selects which of these views should be used for 
> the given URI. 
> Another way to do this (maybe Web Folders does this too, I've 
> not been that side of
> the fence for a while?), is to have some way of switching 
> between modes from within
> the browser; a toolbar button, or a drop-down box, or whatever.
> 
> So then, the user can browse about their web site, happily 
> viewing their pages,
> then maybe they want to move some files about, so they hit 
> the "view collection
> contents" button, and they can do that, then switch back 
> again when they've
> finished. That would be a nice interface, IMO.
> 
> joe
> 
Received on Wednesday, 2 February 2000 20:08:24 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 2 June 2009 18:43:53 GMT