Tuesday, 27 September 2016

Introducing mapguide-react-layout. A modern viewer for MapGuide

Let's recap the story thus far:
Up until now, I have intentionally withheld the GitHub repo where all this development action was taking place as I wasn't going to reveal something that I didn't think was ready to be revealed. Well, it's now reached a point where I am comfortable with finally unveiling this project for general public consumption.

I'm proud to introduce mapguide-react-layout, the new modern map viewer for MapGuide Open Source and Autodesk Infrastructure Map Server.

mapguide-react-layout requires at a minimum MapGuide Open Source 3.0 or an equivalent version of Autodesk Infrastructure Map Server. As I've stated in part 2 of this series, this viewer will require a modern web browser, which practically speaking is any of the following:
  • Google Chrome (stable channel)
  • Mozilla Firefox (stable channel)
  • Internet Explorer 11. Older versions of IE will not be supported
  • Microsoft Edge
  • For iOS: Mobile Safari
  • For Android: Google Chrome or Mozilla Firefox
Overview

Conceptually speaking, you can think of mapguide-react-layout as the logical successor to Fusion and the AJAX viewer and it carries similar high-level concepts as well:
  • mapguide-react-layout has templates just like Fusion.
  • Your application consists of various components (as opposed to widgets in Fusion)
  • Components subscribe and dispatch actions to push data to a centralized redux store for application state. (as opposed to widgets having to explicitly subscribing to each other with event handler spaghetti). Application state is significantly easier to reason about in mapguide-react-layout as opposed to Fusion or the AJAX viewer.
  • mapguide-react-layout offers the same extension points for custom functionality for your MapGuide applications, such as InvokeURL commands.
  • Just like before, a mapguide-react-layout viewer is driven by a Web Layout or a Application Definition (this is not implemented at this stage, but is something I intend to complete).
It is the last point which is probably the most salient. Because a mapguide-react-layout viewer accepts a Web Layout or Application Definition, the authoring and development workflow for MapGuide applications using this viewer remains unchanged

Instead of pointing a Web Layout to an AJAX viewer URL or an Application Definition to a Fusion template URL, you just point either resource to your mapguide-react-layout viewer.

Templates

The mapguide-react-layout viewer includes the following layout templates out of the box:

1. AJAX Viewer



This template visually mimics the original MapGuide AJAX viewer (with much less frames!)

2. Sidebar


This template is a responsive layout based on the sidebar-v2 map viewer template. Well suited for phone/tablet sized displays.

In future releases (once I get the Application Definition support implemented), I'll look at bringing across the existing 5 Fusion templates as well.

Getting the viewer/code

The code to mapguide-react-layout can be found on my GitHub repo. There you will also find installation instructions and other various documentation/notes.

Feedback, questions and pull requests welcome.

Saturday, 24 September 2016

Announcing: MapGuide Open Source 3.1 RC1

Here's the first release candidate of MapGuide Open Source 3.1

Of the fixes made since the Beta 2 release, the most notable will probably be that the installer should now work properly on Windows 10. Didn't this work in the past? Yes it did, but when it did work, Windows 10 was still in preview and the version of IIS was most likely not the version it is now (10.0).

Why is this important? Because the windows installer does an IIS version check to determine what screens to present in the Windows Installer UI. Our version check was:

IISVERSIONMAJOR >= "#7"

Which basically says: show the IIS-specific UIs (and allow advancing through the IIS-specific screens) if the Windows Installer detected IIS (version 7 or higher) on the host machine. On Windows 10, the version reported is "#10".

And due to how lexicographical string comparisons generally work, the above check fails because "#10" is not greater than "#7". Ain't Microsoft versioning great? They skipped Windows 9 to avoid a whole class of version checking problems, but as we've found out some still slip through the cracks. The IIS installer check is now:

(NOT IISVERSIONMAJOR = "#0")

Which just means show the IIS-specific installer UI screens if IIS is present. We don't care what version it is, and "#0" signifies that IIS is not installed. 

And the reason we don't have to care about the particular version is because the versions of Windows that our installer supports will have IIS 7 or higher. So unless Microsoft decides to move away from an appcmd.exe-based approach to IIS configuration in the future, the Windows installer shouldn't have anymore IIS-related issues in the foreseeable future.

So with that interesting version story out of the way, here's the binaries at the usual place.

Wednesday, 7 September 2016

MapGuide tidbits: Turning selection XML into attributes or selection images via the mapagent

Since MapGuide Open Source 2.6, the mapagent has an improved QUERYMAPFEATURES, which allows for greater flexibility to handle the selection and tooltip requirements of our various client-side map viewers. Previously, you would've needed custom server-side code using the MapGuide Web API to achieve this, but now that it is part of the mapagent, all you need is a basic AJAX request.

The main purpose of this operation is to produce an encoded feature selection based on an input geometry (from a mouse click or a box drag) or depending on what you ask for, it can also include:

  • An inline base64 selection image (to avoid a subsequent selection image rendering request)
  • Tooltip/Hyperlink information
  • Attributes and bounding boxes of the selected features
Now what some of you may not know is that you can also use QUERYMAPFEATURES to pass in existing selection XML to get back the inline selection / tooltip / hyperlink / attributes. To do this, you have to omit the GEOMETRY parameter (which normally has the input WKT geometry from the mouse click or box drag) and specify a FEATUREFILTER parameter instead which contains your selection XML. Yes, it's a confusing name. You would've thought it would be named SELECTIONXML or something like that wouldn't you?

Now there is one important note about using QUERYMAPFEATURES in this manner, and was something that I originally thought was a critical bug, but is actually a false positive (hence my motivation for this post): Make sure to include a LAYERATTRIBUTEFILTER with a value of 0.

The LAYERATTRIBUTEFILTER parameter is a bitmask that applies layer-based restrictions on what part of the selection to render:
  • 1 = Include visible layers
  • 2 = Include selectable layers
  • 4 = Include layers with tooltips defined
When the LAYERATTRIBUTEFILTER is not specified, MapGuide will internally default to the value of 3 (include visible and selectable layers). When you combine this with passing up selection XML instead of an input geometry, can you see why it may raise a false alarm? Because your selection XML may be for selected features that are not visible in your current view, so when you pass the selection XML up to QUERYMAPFEATURES without setting LAYERATTRIBUTEFILTER=0, MapGuide will use default visibility/selectability criteria and most likely omit your selected features from the selection image / tooltip / hyperlink / attributes and as a result, you may get back nothing despite passing up valid selection XML and asking for selection image / tooltip / hyperlink / attributes.

So if in doubt, if you're passing up selection XML to QUERYMAPFEATURES, be sure to pass up LAYERATTRIBUTEFILTER=0 as well to avoid any nasty surprises.

Saturday, 3 September 2016

React-ing to the need for a modern MapGuide viewer (Part 5): Back on track and racing

Previously in this series, I had to take a momentary detour to restructure the viewer to use Redux for streamlined application state management as the viewer has grown to the point where having a Flux architecture was essential for viewer development to continue at a maintainable level.

The challenging part of this process was to get the map viewer component to play "dumb" (so that we can wrap a "smart" redux-aware presentational component around it), namely because this component is the one that wraps OpenLayers, and we had to modify our usage in a way that facilitates uni-directional data flow that is required by Redux. Instead of making direct calls to OpenLayers to pan/zoom/etc, it now pans/zooms/etc based on the component props and any changes to those props.

This presents a slightly awkward at first glance (but necessary) situation where any events from OpenLayers that would change the map view and display parameters, we have to intercept these changes and flow it back to the Redux store first, that would then propagate back to its "smart" parent component that will set the necessary prop changes to carry out the map actions.

This work is now mostly complete, so now we can focus back on continuing evolving the viewer.

As I've been implementing the following features described below, I've gotten this great feeling of validation in my choice of using React, because there is actually high conceptual synergy between React and our current Fusion viewer framework.
  • Fusion is based on widgets. React is based on components
  • A Fusion application is composed by widgets described by a Flexible Layout document. Our react-based viewer is composed by various components, which I intend to compose together through a configuration abstraction similar to a Web/Flexible Layout document in combination with our (now) centralized application state in the Redux store.
But unlike Fusion, we have a single source of truth for application state. In Fusion, application state is all over the place with some state requiring manual event subscription to be notified of any changes. Our react-based viewer has none of these problems thanks to Redux.

So now, here's what I've achieved since the last post in this series.

Layout Templates

We now have the notion of layout template components. This is the Fusion template analogue ported over to our React-based viewer. To support this template concept in the React world, we need a supporting cast:

1. We have various registry modules that allow various components and commands to be registered. This is also an extension point that allows for external components and commands to be registered as well.

2. We also have a notion of "placeholder" components which reference a particular component by its registered id. Layout template components consists of these placeholder components and various layout/placement code, which once the application has been initialized, these placeholder components will mount and will render the actual components from the registry by their component ids. The placeholder allows for a clean error boundary that allows us to show informative errors in its place should we try to render a component that has not been registered.

In practice what placeholder components give us can be illustrated like so:

Here's the viewer using the "AJAX Viewer" layout.



Should some components not be registered, say the Task Pane and Legend components, we get a clear error message that spells it all out.


In Fusion, such a situation would probably be a silent failure. Even in this state, the new viewer is still mostly functional as all these components are now independent of each other. Their only shared dependency is the centralized redux store. So with this clean separation of components and presentational layout now in place, I've also explored some other template designs. In particular, a port of my dream responsive map viewer layout.

Say hello to the "Sidebar" template, named after the sidebar-v2 template it was originally derived from:


Unlike the original template, all the jQuery that drives the original sidebar behavior has been ripped out and replaced with React components that do the same thing. Also it has some extra niceties, such as the main toolbar docked vertically, and some in-built smarts to auto-flyout Task Pane content when URL commands are invoked, which is important on mobile devices, as illustrated below when I try to do a buffer on an emulated smartphone display.


Web Layout compatibility

Here's something about the two templates above. They are both driven by the same Web Layout document. Why invent a new configuration mechanism when our existing Web Layout and Flexible Layout documents already do this for us?

Being able to support Web Layouts (and in the future, Flexible Layouts) is important as it means the authoring process for MapGuide applications using this new viewer remains unchanged. The only difference is you point the Web/Flexible Layout to this viewer instead of your AJAX/Fusion viewer. Now this support is not 100% compatible, there's still missing commands I've yet to port across, but feed a Web Layout document to this viewer and it should be able to recognize most of the commands and configuration options within.

As hinted in the previous paragraph, I intend to support Flexible Layouts as well. As mentioned previously, Fusion has such conceptual similarity to React that most of the things that drive Fusion should be easily portable to our new viewer as well.

Other cool stuff

Thanks to ol3-contextmenu, we got a functional context menu now. This is driven by context menu settings from the Web Layout.



The viewer now has basic modal dialog support as well, which is an important component for an "Aqua" style layout template when I get round to it, and also as an alternative target for Invoke URL command content.


Current weigh-in

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


I have no doubt we'll cross the 1MB barrier soon, but once I start seriously looking at "paying for only what we use" with regards to all the libraries we're using, the final bundle size should hopefully approach something more reasonable

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 (epsg.io 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.



Toolbar

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.


Legend

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.