Reusable CSS and components for the Untangled Web Framework. These components are usable from stock Om, but their mutations
assume you at least are willing to use the Untangled mutate
multimethod for your application and are using
the standard Om database format (e.g. you don't have to use anything else from Untangled, just the
default Om database format and one defmulti
).
The CSS is similar to other frameworks like Bootstrap. It can be styled, is prefixed to prevent CSS name collisions with other CSS (so it can be used with others), and is responsive. We chose to make our own CSS because:
- So that the component library is not a moving target subject to breakage from outside (CSS and components evolve in lock step with visual regression tests).
- We needed features not provided by stock libraries.
The CSS can be used in a standalone fashion, but it's primary purpose is to allow us to make a complete set of active reusable Untangled components for rapid application development.
See the README-css.md for more details on working with the CSS itself.
├── dev
│ ├── clj
│ └── cljs
├── resources
│ ├── public
│ │ ├── css ------------- The output folder for the generated stock CSS
├── script ------------------ Location of figwheel convenience script
├── src
│ ├── css ----------------- The CSS source files (which generate the CSS)
│ │ ├── components
│ │ ├── elements
│ │ ├── settings
│ │ └── utilities
│ ├── css-guide ----------- An Om-based UI describing the CSS available
│ │ ├── guideui
│ │ └── styles
│ ├── guide --------------- A devcards-based doc for the React/Om-based elements and components.
│ ├── main ---------------- The Library source (exported as the library untangled-ui. Server and client)
│ ├── test ---------------- Untangled specs for anything that merits BDD logic tests (server and client)
│ └── visuals ------------- A set of devcards for visual regression testing.
NOTE: Make sure you run the gulp command in the README.css to generate the CSS.
Adding something to this library should be quick and easy. The DOM/CSS for a ton of these is already worked out. What we need here are active defui versions of those that can be used easily without copying/pasting DOM.
To Help:
- Pick a component that has CSS/layout, but is not yet here. some of the stuff has gotten out of sync. So, only implement things that are currently rendering correctly.
- Ask "is anyone working on/can I work on X?" and create an Issue on GitHub
- Write the defui/model/mutations, probably all together in a single new namespace. Components that interact full-stack (like image library) need more thought/discussion.
- Make devcards for the rendering of that component in various states (for
visual regression testing) in the
visuals
source folder. - Make a devcard that wraps the component in an untangled-app (see examples)
to show the callback/interactions working. Include devcard documentation that
describes the use of the component. These go in the
guide
source folder.
Please contact us on the #untangled
channel of clojurians.slack.com if you
want to help.
More guidelines below:
To support server-side rendering all code should be written in CLJC files.
- Rendering must be pure! Don't use
js/setTimout
or other js-only things in your UI. That stuff belongs in mutations. If your lifecycle (e.g. componentWillMount) or something needs js things, make a clj/cljs function where the clj version is a no-op, and call that from the UI. - Untangled mutations (via
defmutation
) can be put in cljs-only blocks. They'll never be attempted on server-side code anyhow.
All components must have one or more devcards in the visuals
source directory that show their various states
statically (for visual regression testing). Use the same package structure, and name that namespace with a -cards suffix.
E.g. src/untangled.ui.calendar.cljs -> visuals/untangled.ui.calendar-cards.cljs
All components must have a devcard in the guide
source directory that documents/demonstrates
your live working component with callbacks, etc. Use MockNetwork or real server-side code for simulating full stack
interactions (examples coming soon).
- Things that generate DOM should be prefixed with
ui-
, e.g.ui-button
. This will help with naming conflicts and helps the user understand what generates UI and what does not. - Mutations should be written with
defmutation
so that their symbols end up in the namespace of the components. - Mutations should be written via helper functions of
(f state-map args)
that can be composed into other people's mutations, if necessary. The name of these helpers should be suffixed withimpl
(e.g.close-all-impl
) - Idents for components should use Om table names prefixed to their namespace (e.g.
:untangled.ui.menu/by-id
)- Make this DRY. Define a table-name symbol as ::id. One stateful component per namespace.
SEE COMMENTS IN menu.cljc
These are thoughts on how this should work...evolving.
First you should understand how Untangled does i18n:
We use GNU gettext to extract strings from the source by compiling the source to js, and using xgettext
. So, any
calls to, say, (tr "Hello")
turn into tr("Hello")
in Javascript, which the predefined tools for gettext can work with
(extraction, message merging, translator tools like POEdit). If you call tr on a variable, this does not work, since the
js ends up as tr(v)
. We have a function called tr-unsafe
that allows you to use a variable (tr
will intentionally
crash the compile if you use anything but a literal string for the reason), but since tr-unsafe
will result in a js call that
cannot be extracted statically from source you must somehow tell gettext about the possible values that variable can have,
or your i18n will fail.
Approaches to this are:
- Manually add the missing values to your translation files by hand
- Add a function to your code base that never gets called, and just calls
tr
with the missing ones to aid in extraction. This results in automatic extraction and serves as documentation you can put "near" the uses oftr-unsafe
. - Have labels be something that are not ever stored in state. This is ok (probably preferable) for things like Button, but more complex things like menu's are probably declared with labels when defined, not when used.
- Have the user pass you a lambda for doing the label generation. This approach is a problem with Om, because we cannot store a function in app state (it breaks serialization of history). So, we could, for example, allow them to drop in a unique keyword and define some multimethod for the component.
(defmulti render-button-label (fn [k] k))
(defmethod render-button-label :ok [k] (tr "OK"))
This is extensible, but I don't love it.
Comments/ideas welcome.
Some components have no real active nature (e.g. buttons): they just have things like properties and callbacks and no need for an Om query. For those it seems best to just do normal React kinds of things (pass parameters in a map, and children via varargs).
Other components like checkboxes need to manage state. For simple ones it may make sense to let the user completely manage the state (e.g. the boolean as to the checked state of a box).
In more complex components, like menus, there may be some significant advantage
to making them full-on Om components. For example, a close-all-menus
mutation is impossible to write unless the menus
are represented in the app state.
It is common for a parent component to want to look at the state of some child component. For example, what is the current
date selected in that calendar? If there is user-usable state within your component, you might choose to make simple helper
methods for asking for that data, or you can simply let them access your component's props by keyword. The former is better
for complex components (e.g. calendar) and the latter is probably fine for simple ones. In both cases these functions
should be (f component-props)
. E.g., they should be usable from the parent that has that component's properties locally. These
are also easy enough to use from custom mutations since the ident is supposed to be standardized and well-known for all
components (e.g. getting a component's state is just a (get-in state-map component-ident)
).
In utilities like layout helpers: These are probably plain React components (or even just functions).
We expect standards around this to resolve rapidly, but please participate in the discussion as we progress.
These are thoughts...interested in input at this early stage:
- See exploration_cards in the
guide
build. - See menu_visuals and menus in the
visuals
build. - Changes to appearance for stateful components should go through mutations
- Mutations should be written in an IDE-friendly way using the new 0.7.1-SNAPSHOT+ untangled-client
defmutation
. - Lean towards having all of the UI components interact nicely with the VCR support viewer, so avoid using component
local state for stateful components except in performance hotspots (e.g. animations like panning an image might not be fast if they use
transact!
). - Many simple components will be simple stateless functions and not even React components. These are easy, in that anything they do will be determined directly from arguments.
There are four builds: test
, visuals
, css-guide
, and guide
. Select them via JVM_OPTS
with -D
The visuals
build is for the visual regression cards that show each possible visible state of a component, and are
(TODO) run through a browser-based image capture diff to detect visual regressions due to code and CSS changes.
The guide
build is the development guide cards. These cards demonstrate live examples of the components and have full
markdown documentation. The hope is to evolve this to a better UI, but devcards allows for rapid development at the
moment.
The test
build is for untangled-specs. Those adding component that have any algorithms with complexity are encourage to
write specifications around those algorithms to help ensure correctness. Many of the components are quite simple, so
most components probably will not have specs; however, things like the forms support include a number of more interesting
behaviors that need full testing support.
The css-guide
build is for the raw CSS with examples of the DOM/CSS rules to do raw rendering. Please note you will
need to install npm in order to run this task. Learn more
JVM_OPTS="-Dguide -Dvisuals" lein run -m clojure.main script/figwheel.clj
Open a browser on (port settable :figwheel
section of project.clj
):
- localhost:8001 (Index of the pages below, with their JVM OPT for running)
- localhost:8001/guide.html
- localhost:8001/css-guide.html
- localhost:8001/visuals.html
- localhost:8001/test.html
Support for declarative forms that have full-stack integration, validation support, pluggable validation, pluggable rendering, pluggable form elements, nested to-one and to-many relations, and pluggable database integration for persistence.
Image library (upload/selection) and clipping tools. Includes pluggable image metadata/image storage (e.g. save name of image in postgresql, image data in S3). The image clipping returns clip bounds for an image for embedding arbitrary images CMS-style. Clipping/serving the resulting image is up to you.