Monday, 16 November 2026

Announcing: MapGuide Open Source 4.0.1-Test

I am pleased to announce the release of MapGuide Open Source 4.0.1-Test

The reason this is a 4.0.1-Test instead of just 4.0.1 is because the volume of changes since 4.0 is such that I wasn't comfortable with going straight to a 4.0.1 release. I needed a -Test release and several weeks of breathing space to collect any issue reports and be able to act on them to make sure we didn't miss anything important.

Before we continue, I want to make this declarative statement crystal clear in no uncertain terms:

This release was (un-repentantly and un-apologetically) made with AI assistance from DeepSeek. I have made no secrets about my AI usage in the lead-up to this release. If the fact that my usage of AI offends your sensibilities and makes MapGuide "AI slop" in your eyes, then I wish you all the best in looking for an alternative GIS web mapping server. I'm not going to expend any efforts to convince you to stay. It is because of AI assistance that I am able to make this release *now* instead of probably ... a year from now if I were to apply regular human effort to implement the volume of changes in this release.

With that statement out of the way, here's what's new in this release. It's a big list, so strap in!

MapGuide changes

  • Updated bundled Apache, Tomcat and PHP (mitigated against php2shell)
    • Apache httpd 2.4.71
    • Tomcat 9.0.123
    • PHP 8.3.36
  • Windows installer migrated to Wix 6.x. If you encounter installer issues, let us know on the mapguide-users mailing list ASAP.
  • New experimental OpenAPI spec + swagger UI for the mapagent. The mapagent is not a RESTful interface, which presented some interesting challenges describing the API surface with OpenAPI, but we've made it work for the most part!
  • New experimental Linux support for MapGuide Portable
    • The OSGeo.MapGuide.Portable nuget package can now be used to build apps for both Windows and Linux. The viewer nuget packages built on top of this are still Windows-only.
  • Ubuntu installer script now optionally lets you install msttcorefonts
  • Linux installer scripts now auto-probe and auto-configure known font directories in serverconfig.ini
  • MultiGeometry/GeometryCollection instances now render and select properly
  • Massive memory leak sweep from ASAN-instrumented test runs
  • Massive hardening sweep to address potential buffer overflows and input validation gaps
  • Fix MgNullPropertyValueException when selecting features and attribute data contains NULL values
  • Fix web tier init failures leaving the process permanently broken and improper web tier teardown
    • Should improve stability if using IIS and reduce the need to force app pool recycling
  • Fix MgDrawingService.EnumerateLayers() API not working on Linux
  • Fix leaking byte arrays when passed to MgByteReader.Read() for Java and PHP
  • Fix incorrect distance measurement on coordinate systems like Web Mercator
  • Fix MgNamedCollection potentially holding stale refs
  • Fix incorrect arc calculations
  • Fix name element duplication for KML elements
  • Fix excess KML placemark elements when Advanced Stylization is enabled
  • Fix correctness of "batch sorted block" feature join algorithm
  • Fix Advanced Stylization area symbolization clipping not applying
  • Fix cases where negative render times could be reported by performance report
  • Fix assorted incorrect mapagent behavior
  • Fix striped raster rendering in some cases
  • Improve error message quality when its localized form cannot be retrieved
  • Improve robust CPU perf counter reading on Linux
  • Document log properties in serverconfig.ini
  • Improved BerkeleyDB error message quality (some were previously masked as "bad memory allocation")
  • Improve handling of bigger selections and properly escape id values with single-quotes

FDO changes

Core/General
  • Massive memory leak sweep across core and various providers. Debug runs of FDO core + provider test suites now produce 0 leaks on both Linux and Windows.
  • Massive hardening sweep to address potential buffer overflows and input validation gaps
  • GenericRdbms: The FDO_TRACE_FILE facility is now available for Linux
  • GenericRdbms: Fix various reported capabilities vs reality mismatches
  • GenericRdbms: Harden against queries referencing non-existent classes and properties
  • Use the safer mkstemp API for temp file name generation
  • Honor indentation settings when writing feature schemas to XML
  • Fix XmlFeatureSerializer not honoring XML flags
  • Fix properties with NULL values not included in XML output
  • Improve handling of large chains of OR conditions
  • GenericRdbms: Each FDO provider is now is own deb package on Ubuntu instead of all being bundled under a monolithic fdo-rdbms package.
SDF
  • Fix incorrect NOT IN behavior when NULL values are involved
  • Gracefully handle unknown ordering fields in extended select command
SHP
  • Fix GetSpatialContexts crash when schema cache is invalidated
SQLite
  • Fix filtering on numbers with more than 5 decimal places
  • Fix filtering on dates returning wrong results
  • A "Default" spatial context is now returned if the data store has no coordinate system metadata
  • Properly fix thread-safety on both Windows and Linux when building internal metadata.
    • The deadlock on Linux (reason for original fix being rolled back) was due to using a non-recursive pthread mutex.
OGR
  • Re-architect provider internals for better stability when consumed from a .net client FDO application
  • MultiGeometry/GeometryCollection instances are now read by the provider properly
  • Fix "Invalid index: -1" errors from missing properties
GDAL
  • Prevent blank horizontal stripes when up-sampling rasters
  • We try to include the underlying GDAL error message in more places where exceptions are thrown
  • Update configuration doc XML schema to cover missing elements
King Oracle
  • Fix issue reading negative int64 values
  • CLOBs are now read properly on Linux
  • Unicode strings now insert properly on Linux
SQL Server Spatial
  • The SQL Server Spatial FDO provider is now available for Linux
    • You will need to install the Microsoft ODBC Driver for SQL Server. The ubuntu installer package will install this for you automatically
  • New experimental environment variable (FDO_SQS_PERF) that improves provider performance in various areas
    • Set =1 to opt-in
  • Provider now prefers Microsoft ODBC driver 17/18 before falling back to SQL Server Native Client.
  • Fix poor error message when inserting a row into a table without a primary key
  • Fix spatial contexts with dots in its name not being destroyable
  • Fix query with explicit property list losing all geometry types on its query result geometry property
  • Queries against geography columns apply polygon vertex ordering fixes if required
PostgreSQL
  • Curve geometries now auto-linearized for versions of PostGIS that do not natively accept curve geometries in their spatial operators
  • Now uses proper bind variables for geometry values, eliminating the ugly wkb->hex->wkb roundtrip in SQL query generation
  • Fix failing queries that involve coalesced geometry expressions
MySQL
  • Fix incorrectly reported column types
  • Fix column aliasing in joined queries
ODBC
  • New experimental Linux support for the ODBC FDO provider
    • Previously, we built the provider but did not vet any supported behavior
    • Vetted ODBC drivers for Linux (based on test suites)
      • mdb-tools
      • ODBC Driver for SQL Server
    • Your mileage may vary for any other drivers
WMS
  • Fix "no data" value reported by provider

Download Links

Trac has been on the fritz lately, so the release notes which has the download links have been mirrored to my mapguide-fdo-docs repo on GitHub.

4.0.1-Test release notes on GitHub

All of this ... just works?

As I am doing my final spot checks on a MapGuide Open Source 4.0.1-Test release, I am lamenting that MapGuide Maestro was still windows-only and even though a multi-platform rewrite is in my plans, that is in the distant future. Meaning that to spot-check functionality on an MGOS Linux installation, I'd have to either:

  1. Make raw HTTP requests to its mapagent (painstaking)
  2. Spin up a Windows VM to run Maestro that connects out to the mapagent on that MGOS Linux installation (less painstaking, but still painful)
But then I thought a bit harder. Thanks to the efforts of Lord Gaben, gaming is now viable on Linux. Most windows games that don't use invasive kernel anti-cheats "just work" now out-of-the-box on Linux. It's part of the reason I was comfortable jumping ship from Windows 10 when it went EOL last year.

But I have been viewing this strictly from a gaming context, maybe we can apply that same logic to windows apps in general?

I go and download the latest zip release of MapGuide Maestro from GitHub, extracted the contents to some folder.

Then I open a terminal in that folder and ran: 

wine Maestro.exe

And ... it just works!





This is on Fedora 44, my daily driver OS and one of the major Linux distros!

Does Maestro under Wine operate flawlessly? Not sure. YMMV. Does this invalidate the need for a multi-platform rewrite of Maestro? Not exactly, that plan is still in place. But this discovery is a nice stop-gap measure to tide the Linux users over until then!

So if your Linux distro can run Steam, it can probably run Maestro as well!

Thursday, 22 October 2026

php2shell and MapGuide

As you all know by now, the php2shell vulnerability was disclosed and is wreaking havoc all across the technology landscape wherever a PHP installation is involved as everybody is frantically patching their PHP installs to mitigate this vulnerability.

Since MapGuide bundles PHP, this vulnerability sadly also affects us as well and I feel obligated to at least post something about this.

So here's the current situation of this vulnerability in the context of MapGuide.

1. The next release

The upcoming 4.0.1 release already bundles PHP 8.3.36 that was released today to address this vulnerability, so no problems there. While that covers future releases, what about mitigating your current MapGuide installation?

Before we continue, I want to make it clear that the following mitigation procedure only applies to MapGuide Open Source 4.0. If you have an older version of MapGuide, its bundled PHP is out of the version range where a fix for php2shell is available, so unfortunately you are on your own.

2. Mitigating your MGOS 4.0 install (Windows)

The good news is that our PHP MapGuide API binding is binary compatible with newer releases of PHP 8.3, so in terms of mitigation. It is as simple as downloading a newer zip binary release of PHP 8.3, unzipping it to the Php directory of your web tier installation and then bringing across the php.ini and the MapGuide dll files from your older PHP install and you're good to go.

So firstly, stop the actively running web server (IIS or Apache httpd)

After that, you want the zip download of the VS16 Non Thread Safe build


Then rename your current Php directory in your web tier installation directory to something like PhpOld.

Then extract your downloaded zip file to the Php directory in your web tier installation directory.

Then you want to bring across the following files from PhpOld to Php, overwriting any files as necessary.

ACE.dll
GEOS.dll
lib_json.dll
MgFoundation.dll
MgGeometry.dll
MgHttpHandler.dll
MgMapGuideCommon.dll
MgMdfModel.dll
MgMdfParser.dll
MgPlatformBase.dll
MgRenderers.dll
MgStylization.dll
MgWebApp.dll
MgWebSupport.dll
php.ini
xerces-c_3_1mg.dll
ext/ACE.dll
ext/GEOS.dll
ext/lib_json.dll
ext/MgFoundation.dll
ext/MgGeometry.dll
ext/MgHttpHandler.dll
ext/MgMapGuideCommon.dll
ext/MgMdfModel.dll
ext/MgMdfParser.dll
ext/MgPlatformBase.dll
ext/MgRenderers.dll
ext/MgStylization.dll
ext/MgWebApp.dll
ext/MgWebSupport.dll
ext/php_MapGuideApi.dll
ext/xerces-c_3_1mg.dll

Then restart your respective web server and you're now using a bundled copy of PHP that is mitigated against php2shell.

3. Mitigating your MGOS 4.0 install (Linux)

The Linux story is a bit harder. We provide binaries for convenience, but binary patches are a different affair.

The main problem in terms of convenience is that PHP is only offered as a source tarball on Linux. MapGuide uses the same bundling strategy on Linux as it does on Windows, so it does not take advantage of any distro-supplied PHP.

As a result, you will have to build everything from source.

For the most painfree build experience, you can use the same docker-based build system that I use to produce binaries for Linux. If you check out the 4.0 svn branch, it will already have the updated PHP tarball.

4. Thoughts on our PHP strategy going forward

This incident has really soured me on the idea of bundling PHP with MapGuide. I'm evaluating several options on reducing the blast damage from such massive PHP vulnerabilities going forward:
  • No longer bundle PHP/httpd for the Linux build of MapGuide at least. Instead we go for integrating with whatever PHP/httpd is offered by the distro (assumed Ubuntu). This way mitigating any future vulnerabilities is now the distro's problem, not ours.
  • Or we retain the PHP bundling strategy, but pivot away from bundled PHP being a near-essential component. Why is it near-essential currently? We have various apps (mapadmin/schemareport/fusion/etc) that are written in PHP, so we require it. So how can we make PHP no longer essential? By rewriting mapadmin/schemareport/fusion/etc in .net which has a much better multi-platform and mitigation story. If there is a vulnerability in some .net version, we can just recompile these apps with a newer .net without said vulnerability and be able to ship these updated apps as easy drop-in replacements. Modern .net lets us publish self-contained apps, which negates the need to bundle a runtime environment (the exact problem we have right now with PHP), which makes .net an attractive option for a rewrite. The only problem is the effort in doing the rewrite itself! 
Either way, whatever option we go with, it won't take effect in the immediate future. So future 4.0.x releases of MapGuide will unfortunately still have to go with bundling PHP.

Wednesday, 23 September 2026

And now for something completely different

For those who don't know ...

I have been (and still am) an avid player of Magic: The Gathering since 1995 when Ice Age was released. There was a solid near-decade-long hiatus starting in the 2010s when making money in my software dev career obviously took higher priority than turning cardboard sideways, but soon went back on the cardboard crack train when this hiatus ended.

Over time, my MTG card and deck collection has obviously grown to levels that mental cataloguing/inventory of the collection in my brain no longer cuts it. I need to know where every playable card is at any given time so that when I build a deck, I can easily know if I have a given card for a given quantity, and not waste $$$ buying singles from my LGS or online when I actually have the cards in question.

Obviously this means looking for a tool for card collection management and tracking, but most of the solutions are for smart phones or are websites. I want to track my collection locally from my desktop computer, not from my phone or some third party website.

So as a software developer by trade, the solution was obvious: Build my own MTG collection tracker!

Since this need for a collection tracker came around the time I had concluded my Avalonia test run, this was also the perfect project to fully dive in to building an actual multi-platform .net GUI application using Avalonia.

So after 2 years of development and strong internal dogfooding tracking a personal collection of over 10000 cards, I am ready to reveal this project to the public.

Introducing ... the MTG collection tracker (totally original name! Yes, I know)

This is a .net multi-platform desktop application using Avalonia. On Linux, which is now my daily driver OS after the forced migration from the (end-of-life) Windows 10, it is shipped as an AppImage, making the app an easy single file you can double-click to run.

Collection data is stored in a SQLite database via EFCore. It uses the Scryfall API for card data/images and downloads price data on-demand from MTGJSON, which lets us do things like tracking price history for every card.


Some other key features of the app include:

Deck management and viewing



Wishlist management, so you know what cards you are intending to buy or trade for.


My most used feature: An easy way to find out if you can build a given deck based on your current tracked collection, with a button to easily add the cards you don't have to your wishlist.


General note taking capabilities


And finally, the end result of early DeepSeek evaluation after the GitHub Copilot rugpull: A crude, but functional play-testing surface for any deck in your collection!


So if you play Magic like me and have a need to track your collection and the existing solutions out there don't cut it for you. Maybe this app will fit the bill for you, like it has for me.

Project on GitHub

Tuesday, 22 September 2026

For the price of a cup of coffee ...

Our DeepSeek credit spend has yielded thus far:

When used properly (let's call it: vibe, but verify), AI is indeed a beneficial force multiplier of one's existing capabilities. I certainly wouldn't have achieved this volume of MapGuide and FDO wins in such a short time-frame otherwise!

And the spend has only been equivalent to a cup of coffee!

And we're just getting started!

Sunday, 13 September 2026

What happened to that plan?

5 months ago, I laid down the plans for what was going to happen going forward with my various open source projects. But after that post, it has been nothing but ... silence! If you were to look at my GitHub contribution graph, there was also a noticeable drop-off there too.

So what happened?

What happened was there was a major rug-pull. GitHub copilot changed their pricing model to usage-based billing on June 1 and that completely threw by GH copilot powered plans into complete disarray. The existing subscription at that point pretty much amounted to ~30 spins on the AI slot machine per month, hoping it will produce the outcome I desire and that was just untenable for something I am paying out of my own pocket!

So come June 1, I terminated my GitHub copilot pro subscription and pondered what my next moves were going to be. I wasn't going to give up on the productivity gains that agentic coding gave me, so the option was clearly to look for alternative AI coding provider. Because try as I might (I religiously keep watch on advancements in LocalLlama), locally hosted AI coding models is not (yet) tenable on my personal hardware. I bought that PC to 6 years ago to last a decade at least, and it will still last a decade as long as you give up on locally hosted AI models as desired workload.

I eventually settled on Deepseek and straight away, its payment model was very attractive to me: Just load up your account on credits and reload when credits start running low. This was a low risk approach where I loaded up $20 $10 USD in credits, give it a spin on my various open source projects. If Deepseek didn't pan out for my use cases, no harm, no foul. I simply don't reload with any more credits. But if it did work out for me, I can just load more credits as I go.

So as my initial $20 $10 load is down to its final dollar after several months of sporadic evaluation. I can confidently say that Deepseek adequately meets my AI coding agent needs while maintaining the "bang for buck" that I used to have with GH copilot before their June 1 rug-pull. I will lose some capabilities from GH copilot (cloud agents was nice), but they are acceptable losses.

So what that means in the grand scheme is that I should start ramping up my OSS dev machinery again and resume the plans that I had stated 5 months ago. Our regular broadcasting will resume shortly!

Thursday, 9 April 2026

New plans and stuff

So after finally getting MapGuide Open Source 4.0 out the door, I took a self-imposed hiatus from all things mapping/GIS related for several months, permanently moved from Windows to Linux as my daily driver OS just in time before the end of Windows 10 support, and also to mentally recharge and savor the relief of having this major burden (of releasing MGOS 4.0) being finally lifted off of my back.

I now return with a renewed vigor and some rough roadmaps for things going forward in MapGuide and my other various projects. Part of that renewed vigor is due to the advent of ...

GitHub Copilot

In the past few months in my day job, I have been exposed to GitHub Copilot and it has changed the way I build and ship software, some changes bad, some changes good. Say what you will about AI (or AI-generated code/content) in general, but GitHub Copilot (or any other AI coding assistant) has ultimately been a net positive for me.

What separates how I use AI coding assistants from most depictions of "vibe coding" is that I know the technical and architectural fundamentals of what I am actually after. So I know when GH Copilot is generating what I'm after and when it's generating garbage, and knowing the right prompts to guide it back on track if it starts going off the rails, or know when to cut my losses if the situation is un-salvageable.

So with several months of GH Copilot usage at work, I have been thoroughly convinced that I should get a GH Copilot pro subscription for my own personal use. So last month, I finally bought a GH Copilot pro subscription and ... proceeded to blow my monthly allowance of usage credits in 2 weeks! ðŸ˜‚

But in those 2 weeks, I was able to make some major progress in mapguide-react-layout, knocking off some long standing technical debt and feature requests, some of which you'll see in future dev diary updates on this blog. The productivity gains were massive and turnaround times were quick! A new month has rolled over and with that, a reset of my monthly allowance and I have since learned to judiciously use GH Copilot in a less wasteful manner.

So, now armed with GH Copilot, I have a rough roadmap of things I want to achieve in my various projects, which are all outlined below. So let's start with MapGuide.

MapGuide Open Source

The next release of MapGuide Open Source will be 4.0.1. It will be a bug fix release and include any updated web tier components and upstream FDO fixes since the 4.0 release.

Since permanently moving to Linux as my daily driver OS and understanding that I will still need to produce Windows builds of MapGuide and FDO, I've been re-establishing my windows dev environments for MapGuide/FDO inside a virtual machine and reinstalling all the necessary dev tools. 

One surprise that caught me off-guard was installing Visual Studio 2026 Community Edition. Even though MSVC 2019 is the standardized windows compiler for building MapGuide/FDO on Windows, we can install older compiler workloads on newer releases of Visual Studio. So as part of reinstalling all the required dev tools, I thought it would be a simple case of installing VS 2026 with the MSVC 2019 compiler workloads and everything should be all good, right?

Well, near the tail end of my first MapGuide windows build inside this new VM, I hit a snag on the Windows Installer portion. It turns out our WiX 3.x installer projects are no longer supported in Visual Studio 2026! I was not going waste time and disk space to install VS 2022, so I looked at what options we have for Visual Studio 2026. The answer was to migrate our installer projects to WiX 6.x.

Before GH Copilot, the migration process would've been tedious, having to read migration docs, assess the impact of any breaking changes, etc, etc. But with GH Copilot, I was able to migrate and iterate rapidly towards a WiX 6.x windows installer project that successfully builds and produces a MSI installer.

Now here comes the tedious part that GH Copilot cannot help me with. I still have to manually test this Windows installer, making sure all the feature toggles, custom actions, registry and start menu registrations still work as before and I expect this to take several weeks.

Once this updated Windows installer has been verified as working, there'll be a 1-2 week window of incorporating any bug fixes and updated web tier components before putting out the 4.0.1 release.

mapguide-rest

We will start wrapping up various loose ends on this project. The next release will be 1.0 RC7 and will:
  • Drop support for versions of MapGuide Open Source older than 4.0.
  • Represent mapguide-rest as being feature complete. RC7 to final will be bug fixes only.
I expect GH Copilot to pull heavy duty in helping me knock off all of these long standing items.

mapguide-react-layout

The next release will be 0.15. It will have some exciting new features that you'll see in future dev diary entries on this blog, but the main feature will be to fully decouple ourselves from the Blueprint UI toolkit. Blueprint gave us a nice cohesive set of UI building blocks, but our production bundle sizes have paid a price for this convenience, not to mention that integrating this viewer into other projects that use other UI libraries is problematic as Blueprint is always included, whether you like it or not.

As part of this dev cycle, I've been taking inventory of everything in Blueprint that we actually use and refactoring their usages through a layer of indirection of "abstract UI components" so that we can supply an alternative implementation that is more bare metal and whose appearance and styling can be customized through good ol' CSS.

0.15 will be on the horizon upon the completion of this Blueprint replacement. GH Copilot has paying massive dividends in terms of progress and momentum on this particular project, so I am hoping for a fast turnaround on these remaining items.

MapGuide Maestro

Finally we come to MapGuide Maestro. I will finally bite the bullet and put an end to this endless series of 6.0mSomeNumber releases by putting a final MapGuide Maestro 6.0 release. This 6.0 final release represents the end of MapGuide Maestro in its current form as a Windows Forms based MapGuide authoring and administration tool.

After the 6.0 final release is out. MapGuide Maestro will be rebuilt from the ground-up as a true cross platform MapGuide authoring and administration tool. It will be built on modern .net (and all of its patterns and practices) and our ability to have a true cross-platform UI will be achieved through the use of Avalonia as our UI toolkit. 

A little experiment some years ago plus experience from building/maintaining a separate personal project that uses Avalonia for the UI (not yet announced/revealed on this blog, maybe I will someday) and once again armed with GH Copilot, I have great confidence in pulling off a ground-up rewrite.



And that is a peek into what's in store in the near future. Exciting times are ahead!