Development Blog

Taking a different tack from last time, I decided that it would be worth getting really dirty with low-level engine work for the next attempt at map zooming. One of the main reasons we’d need to go this route if there’s ever to be hope of reasonable performance in software mode: Dirty rects. If we play by the engine rules we get to keep that functionality in its existing simple package, which generally means massive savings on CPU cycles.

REX

It had been a while since I’d done any serious tinkering in “Rogue Engine X” (REX), Cogmind’s underlying game engine. The acronym you might recognize from REXPaint, the engine’s ASCII painting software I built with it for my own use and later released (dang that’s been out for over 10 years now, too, with many of its own users).

I do very occasionally add a little REX functionality here and there to cater to Cogmind needs (or REXPaint for that matter), but it’s been mature for like 12 years so there’s never been any huge developments in that time.

My plans this time were for a pretty big one: Add a third type of glyph size.

To summarize, in traditional terminal style the display is just a uniform grid of monospace glyphs, each with a foreground and background color. Less traditional, and needed to produce Cogmind’s map with square spaces as opposed to rectangular ones more appropriate for text elements, two adjacent text cells can be occupied by a single “wide” glyph.

The concept is simple, though does require that text characters take up about half the width that tiles do, which can be a little restrictive at certain sizes.

So the terminal has a base cell size, though doubling the width of that base size gives another wider type of glyph that can be used as well. (I also shared a larger diagram and some related ideas under the section “mixed fonts” in my Fonts in Roguelikes article.)

Notice how the tiles in this Cogmind screenshot each occupy two cells, delineated by the partial grid overlay.

In practice it gets a little more complicated than one might imagine from the above description, because glyphs are not drawn directly and immediately to the visible console as shown, but instead first drawn to their own subconsole, and numerous subconsoles can overlap one another at different positions. This is great for organizing an interface, though when it comes time to merge everything to create the final view, partial overlapping means you can have pieces of larger glyphs showing through, etc.

The idea is to now add something even bigger than wide glyphs, but a key point is that whatever the new dimensions are they must still be a multiple of the base cell size. We have the regular base cell size used for text characters, a wide glyph size used for map tiles/characters, and what can we extrapolate comes next for a zoomed map if we want it to retain a square aspect ratio? Enter: the quad.

Big chonker tile has arrived.

Doubling the map tile size turns 1 wide tile into 4 (2×2), so while a wide cell occupies two base cells, a “quad” glyph would occupy 8 base cells, 4 in the first row and 4 in a second row. This behavior is similar to the wide glyph, just wider, while also expanding in a second dimension as well, so introducing it to the engine logic is, uh, fun :P

I had to rewrite most of the wide glyph support in order to add quads, but having wide support already there to reference was helpful, and merging everything under the same umbrella kept the overall complexity from expanding much.

To design and test quads I loaded up my old REX testing environment, which contains a random assortment of little test consoles and behavior samples to ensure everything is working properly. One of the important things to test beyond basic functionality (which itself took a little while to get down) is quad overlap with other consoles of different types, and screen edge overlaps.

Been many years since I used this thing! It was put together as the engine features were coming together back in 2011 (my first post about it).

Some of the environment is animated/dynamic, though with quads it’s more about rendering and alignment issues, and confirming that underlying data values are correct.

I got pretty excited seeing the quads appearing normally each time a new test was devised and (finally) passed.

It’s official: REX has quad support!

Fonts

REX/Cogmind/REXPaint/etc use bitmap fonts, so if we’re adding a new glyph size that means we also need to accommodate that size in the font files.

While I allow quad fonts to be loaded from file, and that’s what I worked with for the initial implementation, it seemed unnecessary for our needs as far as providing this zooming feature in Cogmind, since our main goal is to simply allow the upscaling of map tiles. Therefore another part of this engine rework was to allow quad bitmaps to be generated as needed. Basically quads don’t have to exist until a given font set is actually set to be used, at which time the bitmap will be generated in memory by upscaling a specified source bitmap which has already been loaded.

Cogmind

Then there’s Cogmind over here not having any idea what’s about to hit it. Hm, what will the impact be? My first quick test was to simply switch the map font to a quad and just… see what happens!

Well for the most part it Just Works. Wow. No crashes, just big tiles.

There’s some obvious kinks like the fact that I didn’t even change the map view dimensions, causing the map view to also quadruple in pixel size and extend off behind the HUD and off screen, therefore “centering” Cogmind in the bottom right corner. That’s to be expected, along with other issues like console alignment and any other source code references assuming the map view is using wide-type glyphs.

But the important thing is that IT WORKS.

Cogmind’s very first use of the new “quad” glyph support added to REX.

That ain’t no mockup. Also because it’s playing by the engine rules there is zero performance hit from this feature. Zero.

You can see the UI jank--to record that I had to turn off autocentering and use the mouse for directional input, plus the misalignments and weird stuff in various locations (check out the items in the inventory xD). BUT IT WORKS.

There is clearly still a lot to do. Manually test swapping the font is literally all I’ve done so far on the Cogmind side of things--the size can’t yet be toggled dynamically, but before starting this whole adventure I did prove it could work in theory by testing whether the game would explode if I tried to destroy the entire main map interface and recreate it on the fly.

The disparity between the surrounding UI text size and map tile size when zoomed is kind of annoying--it’s not quite the same aesthetic, but if it means some people who otherwise might not be able to play could now do so, I guess that’s a good thing! Also again I find myself wondering what portion of potential players will find this sufficient since it doesn’t address text, but maybe in combination with the Terminus font it will work for most people. We’ll just have to find out.

While doing the latest map zoom experiments I also came up with an initial list of complementary feature ideas, those that could help blunt the negative impacts of having a much shorter view range than usual.

• Cogmind may not necessarily be centered when zoomed, instead having the view gradually shift so that you can see further and further in your general direction of travel, out to your actual sight range. Cogmind’s unmodified sight range (16) while truly centered in a zoomed view would extend at most about 4 spaces out of view in the worst case scenario--a north/south direction, so Cogmind would generally be within that distance of the center unless sight range is further boosted. (East-West direction is less of an issue since the view is a horizontal rectangle for most people’s screens.) I can see this dynamic view positioning being fairly complicated to implement well, but a good formula and related behaviors there could save the player a lot of time that would otherwise be spent scrolling around.
• The above feature is likely more appropriate for keyboard users, not mouse users who wouldn’t generally be happy with a map sometimes shifting to a different position under their cursor during successive movements. For that type of input it would be nice to have a way to quickly set your own relative centering position, depending on the direction from Cogmind you wish to see more of while moving or performing other actions.
• Labels for important things, especially hostiles, that enter FOV but are not currently in view can be shown at the edge of the view in their direction. Or perhaps not the whole label, but more like the floating indicators that appear to denote an offscreen drone or Cogmind location. Cogmind already stops and labels new hostiles, so this would just be an extension of that feature to accommodate zoomed folks who want to have a little more info about the cause. Heck, maybe in temporarily pausing the action it could even shift the view over a bit to directly see the cause?
• For new players, zooming the map could perhaps assume they would like everything to be larger or more readable, in which case maybe it’d be a good idea to also automatically switch the font to Terminus at the same time? Just a thought though, not a fan of this approach, and I think it won’t be nearly as relevant given the nature of future planned UI updates…
• This one’s just fluff, but I can see feedback SFX and an optional very fast animation for the transition between zoom and standard view, for people who want to use both and do it in style :)

Lots of optional features out there, it appears, though exactly how many of them are actually useful, and more importantly can actually be implemented in a way that brings out that usefulness, remains to be seen.

Anyway, those are just some general notes for now, and I haven’t done any real playing with this feature active, but later once it’s actually built and not hacked together I’ll definitely be trying out some runs to see what about this setup irks me and if there’s anything I can do about it.

Although a zoomed map this isn’t the kind of feature I want to use, I imagine it could be useful to others, and look forward to seeing where development takes it. I’ve always loved working on UI to begin with :)

This is the second in a five-part adventure through the process of putting all this together:

• Realtime Image Scaling
  
• Engine-level Architecture
• Implementation
• Polishing
• QoL

A few years back I introduced an experiment to demonstrate a potential “overmap” implementation that still obeys the rules of the terminal interface. What about the opposite--a way to zoom the map itself? Obviously this would be intended for a completely different purpose, addressing one of the more common complaints about Cogmind, that on some displays and play environments everything is rather hard to see.

Prior to this I’ve always framed the zooming discussion as a full-UI thing, where not just the map but everything needs to be larger, which is kind of a show-stopper when there is a minimum number of text elements required to be visible at all times for proper play as designed. But maybe if we only zoomed the map it’d work for some people who otherwise can’t play?

It’s hard to say whether this would satisfy some people since there’s still the text elements, but maybe for example using an alternative font like Terminus is sufficient for those parts, and the map is what we should focus on. Anyway, it could be worth experimenting with, and I’ve moved up the timeline for doing that.

Why now?

I’ve always been interested in experimenting with larger alternative interface layouts, though because I didn’t see much promise in them, and doing so deviates from the core design, the idea was to wait until at least a likely engine update down the road, as well as the completion of most of Cogmind’s content.

Well, this year money issues have had an influence on my near-term direction :P

Revenue has fallen quite a lot, I mean it has been over 10 years of dev at this point, and now that Cogmind is being developed at a bigger loss I need to start worrying about revenue again… (this will also likely lead to some release timeline adjustments in the future, too).

Anyway, to the issue at hand, hopefully with a larger map view Cogmind will be able to appeal to enough additional people who are otherwise okay with the rest of the game--I know there are some out there, and it’ll be better for revenue going forward!

On that note, I must thank all patrons for making the ongoing expansions much more feasible. I’ll admit expansion-level content releases for a niche game without explicitly charging for them isn’t really feasible forever, but I don’t want to split the game world into DLCs--it’d be bad for the design so it’s all or nothing, and there is just so much cool stuff still to add. It must be done. It will be done :)

It’s time to experiment!

Mockups

Any proper UI work is likely to start out with mockups--might as well play around with relatively simple images before investing a greater effort into code…

There you have it, a mockup depicting Cogmind using a size 18 font (Terminus for better readability overall) combined with all map tiles doubled in size (1080p@16:9, the most common Cogmind player resolution). (open for full size)

Okay so each tile actually occupies four times the usual amount of space, but what we mean here is that the cell dimensions are doubled. Something in between 1.0x and 2.0x might be more ideal from a size and visual balance perspective, but doubling is more feasible for retaining the actual aesthetics, both in terms of cell alignment and pixel accuracy, and might also provide us with other benefits later.

The mockup is also missing some components that eventually must be considered, such as what text over the map might look like as far as object labels and other info overlays, but that’s not important right now, more of a detail to consider when the time comes, assuming the fundamental feature even works out at all.

Architecture

It’s time to enter… the zoomiverse!

Okay this is an early blooper, we’ll get to that in a moment ;)

Just how to implement selective zooming in a terminal emulator that is made to do no such thing is a bit of a dilemma.

Terminal-based engines don’t behave like a normal game engine where you have individual windows represented essentially as images layered on top of one another and their contents can therefore be scaled individually. Instead there are many layers of cell data from different subconsoles that feed into a single root console which is then converted to the final image.

Under this kind of architecture it’s not all that reasonable to insert images into the mix, or make modifications to entire subconsole properties of the variety which are easy and obvious to apply to images. We can’t go “oh sure just tell the computer to zoom that window and we’re done with it!”

Sticking to the terminal system’s constraints is great for helping maintain visual consistency, and keeping the overall architecture and interface relatively simple, but if we want to zoom the map things are going to get complicated beyond the scope of what the engine can normally do on its own.

Over time I’ve brainstormed 4 different theoretical approaches to zooming the map, and most recently expanded that to 5~6 (an exact count depends on how different one needs to be in order to be considered unique). Some are more involved than others, and each comes with their own tradeoffs, though having no actual experience with this feature in practice, its true complexity and scope are not immediately apparent. Therefore it makes sense to start with the easiest, least intrusive option regardless of all other factors, just as an experiment to gain a better understanding of what the results feel like, and collect a list of design issues that would need to be tackled to make this a reality.

Realtime Scaling

The first and simplest method is absolute brute force (of course :P). Let’s stretch some pixels.

Sounds easy enough to take the map area and blow it up, yeah? Well, not really xD. Cogmind doesn’t know anything about images, and the engine doesn’t know anything about Cogmind or its UI structure, so we’re going to need a little extra communication between the two on this point.

Basic steps describe the cooperative process:

1. Cogmind registers a callback function with the engine, letting it know that it wants to zoom an area of the interface every frame.
2. When the engine is about to render a frame, it first lets this function know about it and expects to be handed an image in return. That image is created by Cogmind itself by forcing the interface to render [mostly] normally in between frames, copying a central portion of the map view directly to an image, then scaling it up to fit the normal view dimensions.
3. The engine finishes rendering its normal frame, then at the end of that frame takes the zoomed image sent by Cogmind and copies it over to the desired area before displaying the final results on the screen.

If this sounds terrible, that’s because it is :P

The performance of realtime software scaling is no good, with a first iteration tanking my FPS from 240 to 40, and that one wasn’t even working right. Once it got “fixed” the real FPS was more like 24, definitely far below acceptable.

But it did work! Hacky and incomplete though it may be…

Still image of a zoomed Cogmind map, working in game, based on realtime software scaling.

Here it is in action:

Them’s some big ASCII--a working realtime software-scaled map view in Cogmind.

I say “in action,” but really only keyboard input would work in this test since the mouse still doesn’t know anything about this image resize weirdness.

Essentially in this form it’s nowhere near complete, and not performant either. While there are plenty of potential optimizations, optimizing this kind of architecture makes little sense since much of the work would become obsolete by an inevitable switch to hardware acceleration, yeah? Scaling and copying a few images would be nothing for a GPU, but for now Cogmind is CPU-bound and that isn’t changing in the near term.

A few other problems I noted:

• Tons of artifacts created when toggling the zoom (I believe any such zoom feature would need realtime toggling).
• An image-based approach is not directly compatible with some other interactive visual systems that expect to have cell-specific knowledge at various locations, so there would need to be a layer of translation that tends to complicate things.
• Not only is an image-based map unable to take advantage of the normal dirty rect system, in fact it requires turning off that system completely, meaning the engine is always rendering every frame in full. That’s pretty slow, compounding with the software scaling work. In my fullscreen tests, forcing a full render every frame in the normal game gives an FPS of 60*, while realtime image scaling drops it to 25. I managed to up it to 30 with one optimization, but it’s still far from ideal, plus you don’t really want to constantly be rendering at max speed in the first place, even as a cost of getting a map zooming feature. *These speeds are in my dev build, which has a lower FPS than released versions, so I’m just looking at it for relative comparison. (Aside: “Dirty rects” are an important concept in gamedev, whereby you keep track of known areas of the screen that have changed since the last frame, for example defined by a list of rectangles, and only update those areas during the render, since any unchanged areas should remain the same and don’t need to be updated. This is also where artifacts may originate in a game’s display, where perhaps an area changed but was never marked “dirty” for updating along with everything else.)
• The zoomed area would need more nuance, probably in the form of a mask, since it doesn’t distinguish what’s in the target area at all and simply scales everything, so you try to hack a machine and get this…

Many different UI elements can appear over parts of the map view, not just the map itself :/

As with any project dealing with rendering work, this one had its fair share of funky bloopers along the way. One of the main issues was a logic error causing the map to repeatedly zoom itself, resulting in a recursive scaling effect.

Had some funky problems with coordinates for zoom centering, too :P

The Next Step

The map zooming implementation as shared here isn’t ideal, or even acceptable, though the good news is I do have that longer list of possible methods, and will be working with one of the better, if more involved, ones that could solve all the issues presented here.

A smarter approach should play by the engine’s rules, but will take a while longer to implement. The important thing is that playing with this idea showed me what it would feel like and gave me an opportunity to formulate concepts for some promising complementary features which might make this more feasible in a UX sense as well.

Before properly testing out those ideas, however, I’ll need to put together an implementation that won’t try to fry my CPU ;)

This is the first in a five-part adventure through the process of putting all this together:

• Realtime Image Scaling
• Engine-level Architecture
• Implementation
• Polishing
• QoL