Wednesday, 17 August 2016

React-ing to the need for a modern MapGuide viewer (Part 4): A necessary detour

Continuing where we left off from last time, we had something that was closely approaching our existing AJAX viewer.

Sadly, I've hit an impasse.

This viewer has reached a point where I must take a momentary and necessary detour to accommodate a change of architecture to make further development work more manageable. At this point the viewer was a series of components connected together through various event handlers. The encapsulating application component was turning into a big ball of event handler spaghetti trying to get various components to talk to each other.

Any React application that grows in complexity will eventually encounter such problems, and when such problems arise, it necessitates the introduction of Flux architecture to centralize and streamline application state. So if we're going to introduce Flux into this viewer, we might as well go with the best: Redux.

What sells Redux for me as the top-tier Flux library of choice for React applications is its jaw-dropping developer tools support. Just check out this video for a taste of the developer experience we get with a React application using Redux.

Or for something more relatable, here's our current viewer redux-ified

All our viewer actions are now dispatched to a central redux store where various reducers are in place to update/replace various parts of the now centralized application state, and in combination with the Redux DevTools, we get a live audit trail of all our application actions and their respective data payloads and state transitions. Our various redux-ified components then respond to various changes in this redux store to update/render themselves.

You may also notice that we've got our iconic zoom navigator on our map now. Notice how it's properly flashing the loading animation as the map is loading/refreshing? This is because it's connected to a branch of the redux store where the map viewer component is dispatching Map/SET_BUSY_COUNT actions to whenever our MapGuide image sources are about to start/finish a map image request. A non-zero value means show the animation (as one or more rendering operations are in progress), a zero value means to hide it.

That is the extent of the "wiring up" that our components need. Dispatch to update state, connect to listen and re-render against updated state. Redux handles and coordinates the rest.

The Redux DevTools also has a nice graph view that allows us to see our centralized application state in a nice animated graph (which also live updates with each action and change of state)

Can you imagine this kind of developer experience with our current viewer offerings? Not a chance!

This tree (and the values within) is what entirely drives and describes the state of the whole viewer application. This is what a Flux architecture gives us. A single and centralized source of truth about our application.

Can you imagine as well, the ease of reproducing an issue in this redux-ified viewer? You just export the state tree with the Dev Tools, and send it to the developer. He/she can load import that state tree and replay the whole series of dispatched actions up to the point of failure.

The process of redux-ifying the whole map viewer is basically splitting our various components into "smart" and "dumb" versions. The dumb components simply takes the props given to it and renders itself from that, whereas the smart components are redux-aware, wrap the respective dumb component and passes new props to it whenever their connected state branches are updated.

This redux-ification is still in progress, meaning things like the Task Pane (and its AJAX viewer API emulation) are currently broken, but once it's done we can continue onwards with a solid, robust and more maintainable foundation in place.

Monday, 8 August 2016

React-ing to the need for a modern MapGuide viewer (Part 3): We almost have an AJAX viewer replacement

WARNING: This post is GIFs galore. You have been warned :)

When I last wrote about this viewer, we had the following:

  • A functional map viewer component based on OpenLayers
  • A working legend component
  • A mock Task Pane that will eventually function like its AJAX viewer and Fusion counterparts
So where are we at since that post? I think we have something that can almost replace the AJAX viewer.

The above screenshot may not fully cover the extent of changes, so let's cover the changes bit by bit.

External Base Layers

The map viewer can now have external layers incorporated. Currently, this can be OpenStreetMap or any XYZ-based tile set, like Stamen tiles as seen from the above screenshot.

One thing that will surprise you (as it did for me when I got this first working), is that there is no longer a hard EPSG:3857 requirement in order for your MapGuide Map Definitions to properly line up against OpenStreetMap and friends. The above screenshot shows our venerable Sheboygan sample map in its original EPSG:4326 coordinates lined up against EPSG:3857 stamen tiles.

This is possible due to client-side raster re-projection capabilities present in OpenLayers 3 itself. What this means is that as long as your Map Definition is in a coordinate system that has a corresponding EPSG code, OpenLayers will happily re-project other layers to line up with your Map Definition instead of vice versa. In the above screenshot, the Stamen tiles are being re-projected to line up with the Map Definition.

For maps that are not in EPSG:3857 or EPSG:4326, there is currently a small additional setup process to carry out.

For example if I load the map from the Melbourne sample dataset, the viewer throws this cryptic error message.

What this error means (and something I can hopefully clean up), is that OpenLayers has no idea about the projection EPSG:28355 (the projection of the map). Although the viewer includes proj4js and automatically wires it up to OpenLayers, it still only knows about two projections out of the box: 4326 and 3857.

Foreign projections have to be registered to proj4js first ( conveniently provides proj4js snippets to show you what to do) before mounting the map viewer application component. Once I registered the projection for EPSG:28355 to proj4js, the Melbourne map shows up properly, with OpenLayers happily re-projecting the EPSG:3857 stamen tiles to line up.


The viewer now contains a floating toolbar where we can dock our various commands to. Toolbars are completely data-driven and commands inside can be individually be selected/enabled/disabled based on various states of the map viewer.

The react-flyout component is used to provide support for flyout menus

Task Pane

In our previous post, the Task Pane was nothing more than a plain UI mockup. Now it is a mostly functional simulation of what is provided by our existing viewers.

The Task Pane functions as a generic content container where various viewer commands can show their UI in. In order to fully simulate the Task Pane, it needs to provide the JS functions that Task Pane content can call back into. In other words, our viewer needs to shim/emulate the AJAX viewer APIs to allow existing Task Pane content to work.

So to this end, I've been using the official PHP Developer Guide samples as a "reference implementation" for Task Pane content and the AJAX Viewer APIs they're trying to call. The page in the first screenshot is the landing page of the developer guide samples. So what this means now, is we have emulated AJAX viewer APIs for:

Zooming/Panning to a particular location

Setting selections

Refreshing the map in response to server-side layer/group changes

Digitizing geometry

And much more!

The AJAX viewer API emulation works so well, we can throw things like the existing AJAX Viewer buffer tool at it, and it will work just fine in our new viewer environment.


Finally, as can be inferred from previous screenshots, the Legend component is now aware of external base layers and provides a switcher UI if more than one external layer is present.

So when you combine all these changes together, I think I will get agreement here that we have something that nearly approaches (in some cases, exceed) the AJAX viewer in functionality. The only glaring omissions at this point are various display elements (like a status bar), reaching command parity with the AJAX viewer, and a layout/configuration abstraction akin to a WebLayout or ApplicationDefinition, which is something I intend to implement in some form.

Before I close out this post, here's the current weigh-in for this production viewer bundle

The weight increase was mainly due to adding proj4js, but compared to Fusion we're still way smaller. As long as things stay under 1MB, I'd be content.

Saturday, 6 August 2016

MapGuide tidbits: mapguide-rest and PHP 5.6

Since MapGuide Open Source 3.1 Beta 2 shipped with PHP 5.6, some might be wondering if my mapguide-rest extension will work on this version of PHP (from a compatibility standpoint, we mainly are referring to the MapGuide API itself, which is the same as before, and not the surrounding Web Tier environment like PHP, Tomcat, etc).

The good news is: Yes, it works for the most part

I say that because, if you see a message like this:

Automatically populating $HTTP_RAW_POST_DATA is deprecated and will be removed in a future version

It's because of some some deprecation warnings that are active in PHP 5.6. This message in particular, is due to the setting always_populate_raw_post_data not being disabled in php.ini.

And for another interesting fact? It's not set to 1, which should imply the setting is disabled right? Well you guessed wrong, it has to be explicitly set to -1 in php.ini for it to be disabled. Once disabled, these warning messages go away.

Another thing to look out for from the client-side is that PHP 5.6 may include charset in its Content-Type response headers. So if your client code was previously testing for application/json
exactly for a JSON response, that may fail now because PHP 5.6 will most likely send down application/json; charset=utf-8 instead. If you have such code, adjust accordingly.

But besides that, my mapguide-rest function test suite says things are A-OK with this release of PHP.

For a better out-of-the-box experience, we'll look to have this setting default to -1 in the php.ini we ship with MapGuide.