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

For the price of a cup of coffee ...

Our DeepSeek credit spend has yielded thus far:

• The MapGuide Windows installer successfully migrated over from Wix 3.x to Wix 6.x
• Added missing support for rendering MultiGeometry/GeometryCollection instances in MapGuide
• Added missing support for evaluating spatial predicates involving MultiGeometry instances at the FDO level across various FDO providers. This means the MultiGeometry/GeometryCollections that MapGuide can now render, can now also be selected properly!
• Fixed a 12-year old bug where we couldn't enumerate layers of a DWF drawing source in MapGuide on Linux (it was another 64-bit Linux portability bug!)
• Fixed an OGR provider bug where stale CPL errors were not cleared upon constructing and throwing an FDO exception, resulting in these stale CPL errors repeating the same FDO exceptions being thrown, even if the underlying OGR API calls no longer error.
• A solid action plan on a 14-year old bug with striped raster images. This one has lots of moving parts around testing and/or test data which I still need to figure out, so no actual fix has been prescribed yet.

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!

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!