Fork me on GitHub
isolate.js via AST analysis
interactivate
Recent changes in SDK
nodeconf 2012
Write logic, not mechanics
protocol based polymorphism
(clojurescripting :intro)
namespaces
JS Guards
Packageless modules
Addons in multi process future
Yet another take on inheritance
Shareable private properties
Evolving VS Adjusting
oh my zsh
Git status in bash prompt
CommonJS based Github library
Taskhub
Gist plugin for Bespin
Reboot
narwzilla
JSDocs
bespin - JavaScript Server
bespin chromend
Bespin to Helma
bespin multibackend mockup
Adjectives | Ubiquity + Bugzilla love
Some Mock-up around Ubiquity

In this post I am planing to talk about new Ubiquity command I’m working on currently & share some of my thought about:

  • Ubiquity in general
  • Problems I’ve faced during development
  • Improvement ideas

ubiquitylightbox

Not a long time ago I have realized that I’ve posted about several ubiquity commands I developed, but I have never tried to share my opinions regarding project itself. So I have decided to do so: I think that Ubiquity is just incredible way to increase usability of a browser on one hand and usability of web itself on another. One more reason why I think Ubiquity is a really cool project is it’s unique ‘command’ development API that allows writing very complex ‘commands’ while preserving simplicity. I’m pretty sure that one hour with ubiquity will be enough for any web developer to get started. Wiki has a nice tutorial to get you up to speed.

Before I will start talking about difficulties I’ve faced, I will describe briefly the command itself:

Actually it’s not a command, it’s a set of commands for working with Bugzilla. As you might know Bugzilla has a standard API for external client applications. Methods can be accessed via XML-RPC. So I have decided to write Bugzilla client on top of Ubiquity API. Currently commands are under the development so there’s not much functionality yet, but fell free to play with it or [contribute][bugzilla command sourece].

You might could’ve guess that the command is quite complex. Actually that’s one of the issues I’ve faced. While it’s really simple to develop complex but ‘targeted commands’, it almost exponentially hard to develop generic ones.By ‘targeted commands’ I mean commands that do one or maybe two tasks and accept only one or two modifiers. To give a better overview I will try to describe users requirements (at least the way I see them).

  • As a user I would like to use one command during the flow of a task. Technically speaking command should have kind of namespace? / topic used for communication with Ubiquity while performing a task. It should be sort of a keyword describing target against which I’m planning to take some actions.
  • As a user I want to have a set of verbs to describe what actions I am able to perform against target.
  • As a user I want to have a set of modifiers to describe what exactly actions have to do with a target.
  • To preserve simplicity, I would like to have more humanized interface where ubiquity remembers whatever I’m discussing with it. I mean that I don’t have to describe all the details (modifiers) every time I execute command.
    Example:
    > When I am discussing some firefox bug with a real > person which is only reproducible in Linux, I don’t repeat > details of a bug over and over again.
    > I would never say:
    > > “I have commented firefox bug, which occurs only on > > Linux, that I was not able to reproduce it”.
    > I would rather say:
    > > “I have commented a bug that I was not able to reproduce it” > cause I would expect that person I’m talking with still remembers topic of our discussion.

Here is another example with an actual thing I would like to have for Bugzilla ubiquity commands.

  • I’d like to get comments about bug from bugzilla. (There can be several bugzilla systems of my interest. One from mozilla and another from KDE …)
  • I’d like to be able to store my sessions the way I’ll be able to simply identify the origin of the bug I’m talking (mozilla or KDE).
  • comment a bug with some useful information (again in particular bugzilla session)
  • I’d like to get information about user *(can get user by username / email) again in particular bugzilla session)

In the example above we have:

  • actions : get, comment.
  • data: information.
  • modifiers: bug, session, user, username, email.

but the problem is that action get can handle two different modifiers bug and user. Lets assume that in case we mean bug its just a number / id, while in case of user it’s bit more complicated, since it can take username or email as a value, besides it just needs one at a time. One way to solve this is to add all of the modifiers to the get command. Problem is that about it is that then you would have to calculate which modifiers have been passed and based on that figure out weather user meant to get info on a user or a bug. We should keep in mind that we can confuse users that way as well as they will see all this possible modifiers. One more problem is that confused user may type something like:

get 156

Did user meant to find info on a bug or a user? Well obviously he’s confused and typed something even more confusing, but we have to find out that. So we have to add some extra logic to parse user input more precisely.

Another example:

get bug 156 username me

Is user looking for a bug or user info? Should we suggest bugs with numbers starting with 156 or users with the names me or both ?

Think it’s clear now that adding modifiers is a bad solution, besides in a real life it’s way more complex, since you can find bug by: assignees, reporters, resolution, by report date, by target os and many other modifiers. It’s a same with users, even worse, cause some of them may have modifiers like user_id so if you’ll search by assignees you’re passing the user_id and you can do the same if you’re looking for the users with the user_id. Sure it’s possible to identify them with different names but for a complex commands it still a nightmare, and not a simple to use at all.

Well there’s another solution to use different names for actions, actually that what usually ubiquity command developers do (From what I noticed), but unfortunately here is a limitation as we can’t use space separated names for actions. So the solution is to have two commands get-user and get-bug in this case most of the mentioned problems are solved, but it’s not really a natural language. Besides names are quite generic and a lot of commands might introduce them so in order to avoid conflicts we have to use bugzilla-get-user, bugzilla-get-bug. Tendency is visible commands become less and less natural :(. I have to mention that in case of Bugzilla command[s] we would have to complicate command even more.

I was thinking about alternative solution, in order to keep things more natural and I think there is good potential in ‘dynamic modifiers’ - context aware modifiers that are capable of introducing additional modifiers. Lets look at the flow below:

me: Can you please comment on a bug ?
ubiquity: What should be the comment ? (requires additional input)
me: I’m not able to reproduce it
ubiquity: I suppose you mean bug 146 or might be one of 846, 400, 8942 (assumes that I’m talking about a bug we discussed previously, but it also reminds me some other bugs we’ve discussed recently)
me: bug 846
ubiquity: So you mean bug 846 in mozilla (verifies that I am talking about mozilla bugzilla cause we talked about it last time)
me: and attach screenshot please
ubiquity: Do you like this one ??
me: perfect!

No, no you don’t have to type all this!! It’s just a model of a conversation we’re gonna create flow out of:

  1. bugzilla co[nnect][# “suggested complition”] as there are two actions (1. connect 2. comment) targeting bugzilla command ubiquity completes input with “bugzilla connect” because connect is most recently used action with bugzilla command. Ubiquity preselects auto-completed part so that user can modify it while just keep typing. Ubiquity also adds comment action to a suggestions. In case if I’ll press button:

    • right arrow / enter / tab
      ubiquity will request / verify next modifier (more details will follow in a next section)
    • down arrow in this case we’ll see a picture described in the 2nd list item.
  2. bugzilla comment (text)
    ubiquity will auto-complete word “comment” and will show that it expects text input by adding and selecting “(text)” in case of active selection selection will be used instead. If clipboard contains data it will be show up in suggestions.
    In case if I’ll:

    • stop typing / will make a delay
      I’ll get a picture described in the 3nd list item.
    • press right arrow / enter / tab
      I’ll get a picture described in the 3nd list item.
    • press left arrow
      action comment will be selected and the other bugzilla associated actions will be suggested. In the preview help for the comment action will appear. In case of selecting other action all of the following modifiers will disappear.
    • will press down arrow
      In case of suggestions next one will get activated and all the input data will be adjusted accordingly.
  3. bugzilla commentI’m not able to reproduce it” on a bug 146
    Ubiquity auto-completes with “on a bug 146” (“on a bug” is a prefix for modifier bug). Selecting “146” as it’s a modifier bug accessed most recently, ubiquity verifies that I’m still interacting with it, at the same time it still leaves ability to type the id manually. It adds other bugs into suggestions 846, 400, 8942 as they’ve been recently used. Ubiquity suggests everything it has in history as the whole modifier is selected in this case. In the preview information for the currently selected bug (146) is shown.
    In case if I press button:

    • 8
      Ubiquity will auto-complete it with “46” (as it’s was accessed more recently then 8946) and will select auto-completed part (46). In the suggestions only 8942 will show up as it’s an only matching bug id from the history. In the preview information for the currently selected bug (846) is shown.
    • right arrow / enter / tab
      In this case ubiquity will request / verify next modifier
    • down arrow In this case we’ll have a picture described in the 4th list item (as the first suggestion is 846).
    • left arrow input “I’m not able to reproduce it” will be selected in case of having selection, clipboard data they will be suggested.
  4. bugzilla comment I’m not able to reproduce it on a bug 846 in session mozilla.
    Ubiquity auto-completes with “in the session mozilla” (“in the session” is a prefix for modifier “session”). Ubiquity verifies that it needs to use “mozilla” as modifier session value as it was used most recently. It also adds all the available sessions to suggestions, cause whole modifier is selected. (kde, eclipse) At the same time in preview it displays info for “mozilla” session. (user, url).
    In case if I press button:

    • e
      Ubiquity will auto-complete it with “eclipse” (it’s was used earlier then “kde”, but it allows to keep typing unlike “kde”) and will select auto-completed part (“clipse”). In the suggestions only “kde” will show up as it’s an only matching session left. In the preview information for the currently selected session “eclipse” will be displayed.
    • right arrow / tab
      in this case ubiquity will place cursor at the end of the line and will show all possible (optional) modifiers in the suggestion.
    • down arrow
      will select “kde” as a session and will select only matching part of it “e”
    • left arrow
      modifier-bug “846″ will be selected in all possible suggestions will be suggested.
    • enter
      Will add a comment to the bug
  5. bugzilla comment I’m not able to reproduce it on a bug 846 and attach (data)
    I have typed at as I wanted to attach file with a comment. Ubiquity did not suggested attach-modifier as it’s an optional. but when I type “at”, it auto-completes it with “and attach (data)” where “and attach “ is a prefix for the attach-modifier. “(data)” got selected. Ubiquity suggests clipboard, screen-shot in the suggestions as that are the magic words which attach-modifier can take as an input.

What are the advantages of this kind of architecture:

  • All the commands performing the actions against one target can be scoped by the namespace. Actually it can be a one command.
  • Command Modifiers can have their own modifiers, that makes it possible to create better abstraction layer of the services on top of the Ubiquity.
  • Modifiers have are ordered in a logical way, and some of them can be optional, value of one modifier can automatically make useless use of another modifier.
  • Makes commands more closer to the natural language, cause in this model modifier is modeled as a sentence. Use of prefixes, suffixes makes it much easier to represent in more human way (it’s even more important in some languages where suffix / prefix can change whole meaning of the sentence)
  • Modifier is having step by step dialog with a user, keeping user focused more on the step rather on the end result.
  • Asks user for input only in case if there is no other way
  • User types only the data no modifier names.
  • Ubiquity uses history to make discussion more intuitive. (Actually it’s possible to do it even now through commands but think it might have a sense to expose this functionality some in some standard manner to the developers)
Mozshell
Ubiquity command Say
ubiquity command dictionary
Picasa Photo Viewer (Linux port) - Updated
Ubiquity command for JIRA & Crucible
Picasa Photo Viewer (Linux port)
VirtualBox
KeyZilla 0.1
XUL Development