[openstack-dev] [Horizon] Update on Angular Identity work
Tripp, Travis S
travis.tripp at hp.com
Fri Aug 21 21:36:25 UTC 2015
I definitely think the two panel tables in tree now (images and users) should be reduced down in the number of html partials. On initial glance, it seemed pretty easy to just look at the files and know where things are. However, in practice, it makes it error prone and harder to see everything when the header file is separated from the row file. I will put up a patch that at least collapses the HTML fragments down on images, so that can be seen.
I also think that more sections of the html should be reduced to additional fine grained directives, such as the table footer directive Cindy has nearly ready. And these finer grained sections could be combined into a template used by a *basic* full table directive as mentioned below in option 1.
Option 1’s primary strength actually is that it is more rigid and more contract like. The data and the html are mostly separated, meaning you change the data inputs and can centrally control the template. I think for very simple tables, option 1 is probably better. However, that also becomes a firewall of sorts for customizability and ease of making changes.
When it comes to composability, reusability, extensibility, customizability, and readability I don’t think option 1 handles all those aspects better. To achieve simple results, I think it actually can later lead to a lot of complexity in JS that could be solved directly by modifying html.
When I looked at the option 1 below, and thought about how we have different representations of data that should go in table cells that would need to be handled. I also did not see any mention of how the collapse / expand details drawer would be handled and I actually have a lot of concerns on how that would really work or be simple to use or achieve, because the collapse expand details may differ quite a bit on what we want to display.
We’ve already used the table drawer in a number of ways based on the data we want to show (horizontal property listings, vertical property listings, qutoa charts, nested tables, metadata display widgets) and it was easy to do for each case because we had direct control of the HTML and could directly pass the data needed for the various widgets to them. So, with option 1, we’d have to figure out how to make that really easy to do.
Below is something I just sketched out in the last few minutes to maybe help explain it. I’ve also put in a reference to a number of the existing details drawers following that:
ctrl.columns = [
{
"name" : '1’,
“displayName": gettext('column 1’),
“headerClasses": "extraClassesHere to add”,
"rowClasses": "extraClassesHere to add",
"permissions": "blah, blah"
"data": "What goes here? How should this be formatted? Is it HTML? Is it a list?”,
"dataType": "I guess we need a field to specify the format of the data”,
"template": "Or maybe this column needs it own custom formatting, so we have to pass in the template here.”,
"responsivePriority": “5”,
"responsiveHandling": "Need some strategy for what to do when columns appear / disappear. - Do they go in the collapsible table drawer? Where should they be placed? What format should they be placed in? How should it interact with what else is in the detail drawer?",
},
{"etc" : "etc"}
]
"details-drawer": "How is the detail drawer handled? This can be everything from quota charts, to metadata display widgets, to raw properties, to an inner table (security group detail drawer is actually in inner table of the security group rules). How do we have control over what is in here and what the data looks like in here? And should all the data be preloaded or fetched upon expansion?",
<div ng-controller="table.controller.js as ctrl">
<horizon-table
columns ="ctrl.columns"
batch-actions="ctrl.batchActionList"
row-actions=“ctrl.rowActionList”
details=“Ummm, how do I pass through all the various detail formats?">
</horizon-table>
</div>
A few detail table drawer examples:
Flavors: Quota charts and Metadata Display
https://github.com/openstack/horizon/blob/master/openstack_dashboard/dashboards/project/static/dashboard/project/workflow/launch-instance/flavor/select-flavor-table.html#L101-L120
Security Groups: Nested security groups
https://github.com/openstack/horizon/blob/master/openstack_dashboard/dashboards/project/static/dashboard/project/workflow/launch-instance/security-groups/security-group-details.html
Keypairs: Raw output
https://github.com/openstack/horizon/blob/master/openstack_dashboard/dashboards/project/static/dashboard/project/workflow/launch-instance/keypair/keypair-details.html
NG Images: Responsive columns mixed with additional data
https://github.com/openstack/horizon/blob/master/openstack_dashboard/static/app/core/images/table/images-table-row-details.html
From: Thai Q Tran
Reply-To: OpenStack List
Date: Friday, August 21, 2015 at 1:38 PM
To: OpenStack List
Subject: Re: [openstack-dev] [Horizon] Update on Angular Identity work
Hi Doug,
I think your point is valid, but it would basically move the point of conflict from the HTML page to the controller. You could alleviate that problem by having services, aka service for headers, service for table batch action, etc.... that could then follow similar to the angular workflow plugin pattern we discussed at the midcycle https://blueprints.launchpad.net/horizon/+spec/angular-workflow-plugin (Lin, this is how angular does inheritance).
We would also need to follow up this work by enhancing some of the existing directives. Let's take the action-list directive as an example. Currently, you will have to list those actions out manually, so we would have to enhance it by allowing users to add in their own JSON list and have the directive render the full content. Theoretically, that should get us to a point where you can extend actions, workflows, and possibly even columns.
Hi Lin,
Basically, the problem that I am seeing is: we are trading semantic readability, customizability, and ease of LEARNING vs extensibility, complexity, and ease of USE. I agree that we should set a solid example before we let the flood gate open. This is a great discussion, now I'm more resolved to find a pattern that could give us more without the tradeoffs. We have a great community with many smart folks, I'm sure we'll figure something out.
-----Lin Hua Cheng <os.lcheng at gmail.com<mailto:os.lcheng at gmail.com>> wrote: -----
To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
From: Lin Hua Cheng <os.lcheng at gmail.com<mailto:os.lcheng at gmail.com>>
Date: 08/20/2015 09:45PM
Subject: Re: [openstack-dev] [Horizon] Update on Angular Identity work
Hi Thai,
From your example, option 1 seems closer to the *current pattern* not option 2. :) Where the user define a list of action separately from the table presentation (HTML template) rather than embedding it in the HTML. And if the user wants to extend it, they just add it to the list of columns/actions on the table class.
Option 1 seems better to me due to I find it closer to the current pattern. As long as we can reduce the duplicate code (not having to write 9 files to create one table), I'm good with that. :)
My main concern is really to polish first the initial table implementation before folks jump into implementing the tables in all other panels. So we can avoid re-work, don't want another cycle of clean-up/refactor. :)
I think we already have 2 angular tables out, should be enough data to figure out what duplicate code can be abstracted out based from those two implementation.
-Lin
On Thu, Aug 20, 2015 at 4:36 PM, Doug Fish <the.doug.fish at gmail.com<mailto:the.doug.fish at gmail.com>> wrote:
It appears to me that option 1 would be better prepared to be extensible ... That is if a plugin needed to add an action or a column, we could make that happen with pattern 1 (possibly after adding in a service) I'm not sure how plugins ever add these things with pattern 2.
On Aug 20, 2015, at 1:41 PM, "Thai Q Tran" <tqtran at us.ibm.com<mailto:tqtran at us.ibm.com>> wrote:
Hi Lin,
Let me draw on some examples to help clarify what I mean.
Option 1:
table.controller.js
--------------------
ctrl.headers = {
gettext('column 1'),
gettext('column 2')
};
ctrl.noItemMessage = gettext('You no longer have any items in your table. You either lack the sufficient priveledges or your search criteria is not valid');
ctrl.batchActionList = [
{ name: 'create', onclick: somefunction, etc.... }
{ name: 'delete', onclick: somefunction, etc.... }
];
ctrl.rowActionList = [
{ name: 'edit', onclick: somefunction, etc.... }
{ name: 'delete', onclick: somefunction, etc.... }
];
table.html
-----------
<div ng-controller="table.controller.js as ctrl">
<horizon-table
headers="ctrl.headers"
batch-actions="ctrl.batchActionList"
row-actions="ctrl.rowActionList">
</horizon-table>
</div>
So now your controller is polluted with presentation and translation logic. In addition, we will have to live with long gettext messages and add eslint ignore rules just to pass it. The flip side is that you do have a simple directive that points to a common template sitting somewhere. It is not that much "easier" to the example below. What we're really doing is defining the same presentation logic, but in the HTML instead. Lastly, I'll bring up the customization again because many products are going to want to customize their tables. They maybe the minority but that doesn't mean we shouldn't support them.
Option 2:
table.html
------------
<table ng-controller="table.controller.js as ctrl">
<thead>
<tr>
<action-list>
<action callback="someFunc" translate>Create</action>
<action callback="someFunc" translate>Delete</action>
</action-list>
</tr>
<tr>
<th translate>Column 1</th>
<th translate>Column 2</th>
</tr>
</thead>
<tbody>
<tr ng-repeat="items in ctrl.items">
<td>....</td>
<td><action-list>
<action callback="someFunc" translate>Edit</action>
<action callback="someFunc" translate>Delete</action>
</action-list></td>
</tr>
</tbody>
</table>
Here, your table.controller.js worries ONLY about data and data manipulation. The presentation logic all resides in the HTML. If I want to add icons in the table header, I can do that easily. Remember that this is plain HTML, this is a lot easier for someone new to come in and learn this than our special horizon-table directive. It is definitely easier to USE, but I would argue that it is harder to learn.
--------------
If you compare the two options above, you'll see that all we've really done is move presentation logic from the controller into the HTML. You have to define that logic somewhere, why not in the HTML? This makes it easier to read and know what you're going to see in the browser (something HTML5 spec is evangelizing), and you get the bonus benefit of customization.
I'd like to point out that we aren't getting rid of directives, we're still using directives them (like <action-list>, <action>, <magic-search>, etc..) in our tables. The pattern is, you build your panels using smaller components instead of having one giant component that encapsulates everything. Of course, there isn't a right or wrong answer, in fact there are two very different implementations of a table directive out there right now:
http://ng-table.com (more inline with option 1)
http://lorenzofox3.github.io/smart-table-website/ (more inline with option 2)
Basically, what I'm trying to say is: let's build something simple and easy to understand first (small components that we work), then we can build something more complex on top of it so that it easier to use. I don't think there is a right or wrong answer, just two very different ways of thinking and implementation. But if we start with smaller components first, we get the goods of both world. The guys that want to customize will have a way to do it by bypassing the horizon-table directive, and the guys that just want a simple table can use the more complex directive.
-----Lin Hua Cheng <os.lcheng at gmail.com<mailto:os.lcheng at gmail.com>> wrote: -----
To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
From: Lin Hua Cheng <os.lcheng at gmail.com<mailto:os.lcheng at gmail.com>>
Date: 08/19/2015 05:15PM
Subject: Re: [openstack-dev] [Horizon] Update on Angular Identity work
Hi Thai,
Thanks for investigating the two options.
Option 2 might be better. Folks have to learn the new pattern of writing multiple files, so I think the learning curve for a new table directive is not that much of a difference.
I think option 2 is going to be easier to maintain, since we have a layer of abstraction. It might even also increase adoptability since it would be easier to use. It might be harder to customize, but that would probably not be done often. The table directive would be used as is most of the time.
My thought is design the code to be easy to use for the use case that will be used most of the time rather than the customization case which maybe harder to do. Which leads me to preferring option 2.
Thanks,
Lin
On Wed, Aug 19, 2015 at 12:16 PM, Thai Q Tran <tqtran at us.ibm.com<mailto:tqtran at us.ibm.com>> wrote:
Hi Lin,
I agree with you and Eric that we have a lot of HTML fragments. Some of them I think make sense as directives:
The table footer is a good example of something we can convert into a directive: https://review.openstack.org/#/c/207631/
The table header on the other hand is something more specific to your table: https://github.com/openstack/horizon/blob/master/openstack_dashboard/dashboards/identity/static/dashboard/identity/users/table/table-header.html
So there are two approaches we can take here:
1. Keep some of the presentation related data in the HTML: mainly things like table headers, column definitions, translated texts, etc... I like this approach a bit more because it allow us to read the HTML and know exactly what we are expecting to see. This table.html is compose of smaller directives like hz-table-footer and regular html tags like <th> and <td> etc... I think as we have more of these smaller directives available, we can combine the fragments into one file.
2. We could create a more general table directive with a common template. This is more inline with what we have currently for legacy. BUT the presentation logic like translations, definitions would now have to reside in the table controller AND we lose the semantic readability part. Doing it this way could potentially introduce more complexity as it now requires people to learn the table directive, which could be very complex if it does not use smaller directives. Another common problem we encountered with this pattern was a lack of customization. In legacy, it was pretty hard to add an icon into a table cell. If we go down this route, I believe we might start to encounter the same issues.
In summary, we are working on addressing the HTML fragments, but I think we as a community should go with option 1 and stay away from option 2.
-----Lin Hua Cheng <os.lcheng at gmail.com<mailto:os.lcheng at gmail.com>> wrote: -----
To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
From: Lin Hua Cheng <os.lcheng at gmail.com<mailto:os.lcheng at gmail.com>>
Date: 08/18/2015 02:36PM
Cc: Vince Brunssen/Austin/IBM at IBMUS
Subject: Re: [openstack-dev] [Horizon] Update on Angular Identity work
I think the table setup pattern have some opportunity for reducing code duplication before it gets re-used by other panels..
We used to just need to write one file to define a table, now we have to write 9 files [1]. Can we have a table directive to reduce the duplicated code before moving forward to other panels?
-Lin
[1] https://github.com/openstack/horizon/tree/master/openstack_dashboard/dashboards/identity/static/dashboard/identity/users/table
On Tue, Aug 18, 2015 at 11:49 AM, Thai Q Tran <tqtran at us.ibm.com<mailto:tqtran at us.ibm.com>> wrote:
Hi everyone,
Just wanted to keep everyone up to date on the angular panels work. The goal was to set a pattern that others can follow, to that end, there were a few requirements:
1. reusable and possibly pluggable
2. easy to understand
3. reduce code duplication
These requirements don't always go hand-in-hand, and that is the primary reason why it is taking a bit longer. I believe we are nearing the end of it, here are some items remaining that I believe is crucial to finishing up this work.
a. i18n was completed, so we need help moving gettext blobs to HTML templates (example patch: https://review.openstack.org/#/c/210366/ ) volunteers are welcomed! We want others to use the translate directive as the main way to translate text blobs, this was why we went down this road using babel and angular_extractor plugin.
b. transfer table supports clone feature ( https://review.openstack.org/#/c/211345/ ). There were a lot of template duplications, this clone feature reduces the HTML by a considerable amount. Since this is something we use quite often, it made sense to invest time into improving it. We have had complaints that there was too much HTML fragments, this will address a bit of that. One of the challenge was to get it working with existing launch-instance, so I spent about 2 weeks making sure it worked well with the old code while allowing the new clone feature.
c. I believe we have a pretty good pattern setup for tables. The final piece of the puzzle was the patterns for various actions. We have wizard (create user), form (edit user), confirmation dialog (delete user), and actions with no modal dialog (enable user). We wanted a general pattern that would address the requirements mentioned above. There were some discussions around extensibility at the midcycle that I think will fit well here as well ( https://blueprints.launchpad.net/horizon/+spec/angular-workflow-plugin ). The actions can follow a similar pattern to workflow. I believe this pattern would address #1 and #3 but making it easy to understand is a bit challenging - I think this is where documentation could help.
https://review.openstack.org/#/c/202315/ and a few other patches are going to be ready for review soon (sometime end of this week)! Item #c is the most important piece, it is going to be the general pattern that people will use to build their angular panels with, so the more eyes we can get on it, the better. My aim is to get it in before the feature freeze and I think that is entirely possible with your help. So please help review even if you are not a core!
Thanks
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe<http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject<mailto:OpenStack-dev-request at lists.openstack.org?subject>:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe<http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject<mailto:OpenStack-dev-request at lists.openstack.org?subject>:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: OpenStack-dev-request at lists.openstack.org<mailto:OpenStack-dev-request at lists.openstack.org>?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe<http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
__________________________________________________________________________
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject<mailto:OpenStack-dev-request at lists.openstack.org?subject>:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
More information about the OpenStack-dev
mailing list