Hi Andy,
Apologies for the week-long delay in responding. I've replied inline
below:
On Mon, Jan 30, 2012 at 03:51:07PM +1000, Andy Fitzsimon wrote:
Hi all,
I've spent quite some time thinking about global UI improvements we can make to
conductor for a future release.
Now it's time for me to share a small summary of those ideas and hopefully get some
feedback.
I'll be updating
http://ndftz.com/playground/ with much the stuff mentioned below but
as of right now it is not a representation of the functionality described; so if you have
seen it before, please disregard everything aside from the responsiveness and the vector
icons aspects. I'll now switch gears to get the mock interface up to speed with all
the client side features described.
I'd chatted with you a bit on IRC about this in the past, but to
reiterate: this looks *awesome*!
(attached is a formatted copy in-case you don't like reading markdown)
.. so here we go..
======================
#Goals for a better UI
> This is a plan to evolve the conductor UI into a lean, adaptive, beautiful and
considered experience.
### Specifically, we hope to address:
* Navigating and issuing tasks quickly
* Without unnecessary http-roundtrips just to check on a view.
* Without having to learn new components in every view
* Without having to refer to documentation to be assisted in core tasks.
* Keeping the ‘big 6’ consistent:
* Consistent aesthetic
* Consistent behaviour
* Consistent notification
* Consistent messaging
* Consistent help
* Consistent referencing
* Address as many media types as we can get *for free*
* __Mobile__ - 480px and up via minor accommodations in our UI
* __Workstation__ 640px-1280px - our default view
* __Section 508__ compliance including screen readers
* __Printable__ - As a read-only inventory manifest of a users entire access.
I've sort of been watching the "responsive design" movement from the
sidelines and am rather intrigued. The ability to have one layout that
works across a wide variety of devices would be fantastic.
* Feel like a 2012 client app
* Cutting edge UI conventions
* Swiss style visual metaphors & elements of typographic style
* Baseline grid aligned typography consisting of 2 weights with single leading.
* Use only vector graphics
This one gets a "Whoa, _that_ is nifty" reaction from me. :)
* Blazingly fast client side render times
# Glossary of terms / explanation of features.
## Views
One distinct category of content.
### explained
“the users view” or “the providers view”.
A views name can even be as specific as “userXX-provider-permissions view” or
“admin-provider-permissions-userXX view”
Does this imply that we are moving away from the "pretty view" vs.
"detail view" system? (I would not be averse to this at all, but wanted
to be clear.)
For more information on how we implement views, see 'View anchors' below
******
## Classes
Think HTML/CSS/JS classes
Classes track status or content type but not display.
<a href="#provider23" class="provider23 enabling ec2"
>online</a>`
<div id="provider23" class="provider23 enabling
ec2">Ec2…</div>
### Classing convention
#### Reserved classes
.active
is reserved for navigating views
.enabling .disabling .verifying .deleting .requesting
are action classes reserved for client side action confirmation
.enabled .disabled .verified .complete .building
are action classes reserved for server events
.warning .note .info .tip .caution
are reserved for assistive content
.static-reference .assist-reference
reserved for anchored clones of content via static and assistive referencing (see
section)
.progress .loading .building .compiling .altering
are progress classed to be used in conjunction with a numerical class
.50 .10 .20 .25 .75 .100 .30 etc…
are percentage classes to be used in conjunction with a value
e.g.: `<span class="75 usage">75% of 300gb </...`
or better: `<span class="75 usage">225gb of 300gb </...`
.usage .progress .load .capacity
are **status** classes reserved for identifying the **type** of measurement
.instance .provider .user .pool .group
are object classes to be used on any object
I think I need a bit longer for all of this to sink in.
### Lack of classes for layout
We will __NOT__ use classes for defining grid layout e.g. `.span-12` or `.left-120/`
All layouts will be defined in CSS via a combination of their parent view and the
relevant media query.
for example we will use the CSS
#admin #providers .account .summary{width:33.333%}
rather than relying on injecting style logic into the html like this
`<class="summary span-4">`
This feels much cleaner, though I fear the old pattern will be a hard
habit to break. (For me, at least.)
******
## Graceful degradation
Support for all core functionality in as many scenarios as possible
* New browser features (subtle animation, gradients etc)
* Older browsers lacking features
* Browsers with javascript turned off
+1000 for doing this. Right now we badly misbehave in some sections if
JavaScript is disabled.
* Screen readers (section 508, we’re going there)
******
## Responsive design
Support for viewing from various media eg cellphones, touchscreen devices, kiosks, tv’s,
even print.
### Explained
1. We collapse tables into record views for narrow screens.
2. We Grow fonts and icons for tremendously high resolutions or touchable large
screens.
3. We use only vector graphics (css /webfonts) for adaptive scaling and themeing.
Anchor referencing with js inlining means repeated content is printed only once. a print
version would read like an infrastructure status report from the users perspective.
******
## Visual Metaphors & Vector Graphics
Visual metaphors are graphics symbolic of the function they perform.
Vector graphics are a resolution independent implementation.
### Explained
All icons, visual indicators and ornamental graphics are implemented like this:
* Vector glyphs are encoded into a woff webfont
* The woff font is base64 encoded in the stylesheet to minimise http requests (the
current set of 30 icons is ~5k) so this is very acceptable.
This is my first real exposure to this concept, but this sounds really
novel.
* We fallback to url linked images supplies for legacy browsers (no
.WOFF support) by using [
modernizr](http://www.modernizr.com/) classes `.no-webfonts` and
`.webfonts` to discriminate our implementation through css
******
## View anchors - or 'single page app'
View anchors are portions of our design implemented within the one DOM using basic
anchoring functionality.
### Explained
In the proposed change, the entire conductor is one page for all default views,
What I still don't quite understand here is whether this means that,
when we render this page, we need to render *all* views in advance, or
if they're lazy-loaded via AJAX. The former sounds like it will be
rather slow (on the back-end), since we have a few pages that make
expensive calls to external services.
1. Link anchors will activate the specific view or sub-view.
2. Anchors that do not resolve will initiate a "cannot find" message which will
most commonly occur when a user does not have access to an object and has manually entered
a custom # URL. the two exceptions to this will be ~/# and ~/#top which will initiate the
default app view.
### In action
For example:
> The URL below should open the admin > providers > ec2eastcoast >
connectivity view.
https://conductor/#ec2eastcoast-connectivity
^ ^ ^
Server ObjectID ChildView
#### What the html does
e.g. the `<div>` representing admin > providers > ec2eastcoast would look
like
<div id="ec2eastcoast">.. ec2 east coast information goes here
and the connectivity tab which exists inside the ec2eastcoast provider would look like
<div id="ec2eastcoast-connectivity">
^ ^
objectID ChildViewname
Why?
Because the provider _ec2eastcoast_ lives natively within the _admin > provider_
section, it does not need to be referenced in the ID.. it simply has to be unique. But
because the connectivity tab is only unique because of the provider it is associated with,
it needs to be prefixed with the objectID
All direct links to the provider view would look like `<a href=#ec2eastcoast>EC2
East Coast</a> which would activate the provider tab.
#### what JS does
1. JS hides all non active views except for the default view (classed as active by the
js)
How does this behave in a browser without JavaScript / with JS disabled?
2.
3. when an ID is requested the object and all parent views get classed `.active`
4. [hashchange for old
browsers](http://benalman.com/projects/jquery-hashchange-plugin/)
to preserve browsing history
As Scott brought up on IRC this morning, there appear to be places where
we currently break the "Back" button. If we're able to become consistent
in preserving history, that would be fantastic.
#### what the CSS does
The body class `.nojs` is removed once Javascript is loaded so the CSS specifies that
only views be visible when they are classed as `.active`
.js div{display:none} // Hiding all views
.js div.active{display:block} // Showing active views
.nojs div{display:block} // All views are shown when tabs don't work.
******
## Referenced content
> Referenced content is a live-copy of any existing part of the application. It can be
implemented with static references or assistive references
> See [this JS Fiddle
demo](http://jsfiddle.net/andyfitz/hRqGa/191/)
#### example
1. Global help content can be referenced from many places at once so as to avoid
redundancy and annoying ajax requests.. if the section the help comes from is loaded, so
is the help.
Semi-related, the concept of "global help" at all would be fantastic! ;)
Right now we have some confusing content with no help on-page.
2. Inspecting an object via a link can call the objects entire
content into view.
******
### Static-references
Where an anchor link to another section has been replaced with a live copy of the section
it links to.
.static-reference
> An example would be any monitoring view that requires various metrics already
rendered elsewhere be included.
******
### Assist-references
like a static reference but with the link-text preserved and the section only visible
when the link-text has focus (has been clicked).
.assist-reference
> An example would be a permissions editor referencing the objects that a user has
permissions to by name and allowing the user to inspect that object in-place with a
single click, without having to visit the object in questions default view.
******
## Help & content
Written content to help describe a feature or scenario.
Combined with admonition classes
.warning .note .info .tip .caution
Often implemented via one of the following methods
* Included in the view relevant to its description
* As static-references where the help content can be dismissed (via an X in the top
right corner)
* As assist-references where the help content can be revealed upon clicking a referring
link
Practically speaking this is text that can be placed inline with a view that it
represents or hidden in a non displayed useful content area to be static or
assist-referenced by many places.
### Explained
For example: _“don’t be silly“_ is a generic tip that may apply to an enormous variety
of contexts. This is useful to be placed in the "generic tips" area and linked
to via `<a href=”#dont be-silly” class=”assist-reference”>now please
remember..</a>` which means it wil be rendered inline of where it is referenced.
******
##Client-side status polling
All objects relevant to a view are classed with the viewname
On any action inside the view, a status classname is added to the view as well as any
objects that reference it regardless of where they are displayed
<a href=#user class=”userXX enabling”>link</a>
<div id=#”userXX” class=”userXX enabling” >UserXX View</div>.
We watch the referenced views classes and mimic them on all referencing links always as
they can contain or be updated with new event styles (for example: failed, warning,
upgraded, admin-only etc)
******
## Server status events *investigating*
Right now we can use ajax calls to update each view but in future it would make sense to
use eventSource for more advanced monitoring: [eventSource: server sent
events](http://html5doctor.com/server-sent-events/)
******
## Client verification / fast feedback
Interactive assurance that an action has been requested without the verification that the
server has performed the action.
For example a user clicks the enable button.. we immediately apply “enabling” status to
the button but wait for the server to return with the “enabled” confirmation...
### How it works
> **Assumption** intermittent states like verifying, enabling and disabling states are
safe for client side validation.
> They confirm an action has been requested but not applied.
As a rule, all elements are given the status of the objects they represent as well as
classed with the object ID.. e.g.: all buttons with `class="providerid12451-*"`
get the additional class of their intermittent state
``"enabling/disabling/verifying/whatever"`` When a functional object related to
them has been actioned.
1. On the client javascript side, we set the actioning classname to all elements with
the same object objectid class.
2. Once the action has been done an ajax call confirm the change and refreshes the class
to "enabled/disabled/done/whatever" and removes the intermediary state
It's not clear to me from this, but does this actually require a second
call? It seems like, on clicking 'submit' or whatnot, we could change
the status like you describe, and then, when the request completes
naturally, apply the update. That saves us from an extra AJAX query to
check the status.
******
## Important action certification
[JS Fiddle
demo](http://jsfiddle.net/andyfitz/mvdST/)
Requiring the user to certify that they understand the action before allowing it to take
place.
for example:
> “Do you understand what you are doing?[yes] okay, here’s the button [click]”
###explained
> Enabling and disabling providers, deleting users etc
We will use three methods to verify that a user has accepted the severity of the task
they are about to perform:
1. CSS physical obfuscation.
Where the confirmation toggle sits physically on top of the action toggle so that an
action can not be clicked without clicking the confirmation toggle first. this helps
simplify the component to one physical space part of the same consideration set.
2. javascript validation.
The action button beneath is disabled until the confirmation toggle has been clicked
3. final server verification.
the submitted action is not accepted by the system unless the post contains both the
verification enabled and the action. the server then pushes the update
Right now we're just using JS verification in a handful of places,
because Rails includes an easy helper for this. Is there much of an
advantage to involving CSS and server-side verification as well?
******
##Global status and system notification stack.
Important information about the entire system or a current process, displayed
prominently.
###explained
Using the referenced content heavily, we throw system, warning and notification content
to the top of the page in a collapsing, growl-style notification system”
<div id="notifications">
<h1>System Messages</h1>
<p class=”system”>you should to install aeolus-all to properly use
conductor</p>
<p class=”status provider123 build”>
<a
href=”provider123-build-status”
class=”reflink 50”>
building provider123 <span>50% Complete</span>
</a>
</p>
</div>
From talking to a few people who don't use Conductor every day, it
sounds like our current flash messages are easy to miss / overlook. I'd
love to make them much more prominent.
******
## Printable manifest
Because our content is not replicated, and all enabled views are included in the DOM;
our app can actually print as if it were a manifest for the authenticated user.
We get this feature for free by proxy of our method for referencing views and content.
Having the ability to produce a hard copy infrastructure report may be a gimmick, but
it's an easy way to show how conductor can help respond an IT audit situation.
### example print stylesheet changes
@media print {
a:after {
content: " ("attr(href)") ";}
div:before{
content: " ("attr(id)") ";}}
}
This will preserve all references and links throughout the UI.
Also, we can manually define sections to be printed.
@media print {
div{display:none}
#providers, #instances, #users, #global-tips {display:block !important}
I will definitely yield to you here because you are astronomically more
experienced with this stuff than I am, but I thought that using
!important was a bit of a hack to be avoided except where you needed to
override styles on a particular page. Is this not correct?
}
Which means only sections we desire to be included in the manifest get printed
also
@media print {
.tabs:before{
page-break-before: always;
content:"This section contains the following:";
.tabs li a:after{content:""}
.tabs { font-size: 2em;page-break-after: always;}
.tabs li{display:list}
}
Which will convert our tabs into whole page section summaries.
> remember, this feature is free because we are designing conductor to gracefully
degrade. Very little effort goes into preserving this feature so long as our standards are
maintained.
# Existing conductor paradigms
These changes don't create any changes to terminology, or have much impact for
pre-existing views for components already built into conductor. It is likely that the
majority of what has already been done will be preserved and only undergo minor changes to
the HAML/SASS so that features may be integrated into the global view.
So this brings me to one point that I do want to stress... I think this
outline and the demo you have are fantastic, and I hope we're able to
implement it. But at the same time, I've been here for just over a year
and have already seen 3 wholly different user interfaces implemented, to
the point that it's almost a running joke that we rip out and
reimplement our UI when the seasons change. I think our sanity would be
greatly preserved if we could approach this as a series of evolutionary
improvements, as opposed to a revolutionary, "reboot"-style
implementation of a fourth totally-new UI.
This has been written purely to start discussion on how we evolve
Conductor going forward and I'd love to hear your thoughts: [andy.fitzsimon(a)redhat.com
](mailto:andy.fitzsimon@redhat.com)
One other thing I've noticed is that both the current UI and your mockup
are pretty gray. I'm not necessarily lobbying for a technicolor UI, but
I find the monochromatic interface to be kind of bleak. Is it possible
to add some splashes of color?
-- Matt