As far as "modern techniques" are concerned, I don't know that there's been a lot of scientific progress in that sense, and, obviously no one's making tools for such platforms -- at least not anywhere close to the kind of development we have for exponentially more advanced modern systems where that complexity is not only worthwhile (solve more complex problems) but necessary (the machine itself is more complex).
These examples are going to be entirely hand written ASM; or, the critical parts are, at least. And with possible "racing the beam" graphics engines, the whole game state machine may need consistent execution time, too; these may be extraordinarily fragile systems indeed, and the marvel is at how much can be accomplished, pushing the envelope (of CPU and bus cycles, memory capacity and speed, etc.) as perfectly as possible to the point of the whole thing falling over into a glitchy crashing mess. This can only be achieved through slowly won, extremely deep understandings of the base systems. And which is nothing different over the now multi-generational history of these systems.
So what's new? Just historical perspective, I think. We have some benefit from more clever code generation tools, but these too are hard-won and bespoke among demoscene circles, and are nothing new; there is room for incremental improvement, however. (For example, say you have a situation where certain operations need to be completed within some number of cycles, and interleaved perfectly with other operations; a tool could search ever-larger configuration spaces to determine how to optimally divide up and interleave those operations. Even very poor performing algorithms (exhaustive search?) can be afforded these days, which weren't at all feasible back then; granted, exponential (or worse) execution time is still very much against you, but you can afford another order of magnitude every couple of years -- it's something.
And the other is just, we're more comfortable with games, or GUI interfaces, and what goes for a minimum feature set. Even kaizo romhacks have gotten more popular -- and less punishing. As a microcosm, consider the notorious early ones like:
https://www.smwcentral.net/?p=section&a=details&id=16059to recent ones like:
https://www.smwcentral.net/?p=section&a=details&id=28951I imagine, due in no small part to the success of Celeste (2018), a precision platformer (on PC, Switch, etc.) that, while extremely challenging, is also extremely gracious to the player, and has a fabulous kicking soundtrack, art style, and story. It's perhaps the archetype from which precision platformers and kaizos will forever after draw their inspiration, and perhaps, in a sense, a peak the whole genre had been building to, and will forever be changed by. In short, I think it's fair to draw a line from classic insultingly-punishing romhacks, to this non-hack game, back to romhacks and beyond, in the present and future.
(And, perhaps not even very surprising, there is also celeste.smc:
https://www.smwcentral.net/?p=section&a=details&id=29613 )
And I think it's fair to generalize this point, in that we will find many other examples where the same sequence fits. FPS games were around in a basic sort of way since the mid-late 80s, but were forever changed after Doom (1993), and so on by the Quakes, Unreal, Half Lifes, and beyond. We can draw a line back to Doom, now that those of us who grew up with it (Hi!) are getting nostalgic for it -- but are looking for something with more shine, complexity, style -- and less unfair mechanics than the classics. There's a whole subgenre of "boomer shooters" now, imitating the low-poly or paletted pixel-art style of those old games, but remixed with the hindsight of modern games, their ease of use, gentler skill curves, and so on.
And so we can return to the case of something like Sonic: at the same time the series has been going forward by official development (Sonic's been in 3D for decades now, matter of fact), it's also been going backwards, in modding and porting by fans, and indeed new remixes and challenges by them as well -- in fact, an official title even started as a fan project and SEGA liked it so much they picked it up and finished it as a full title:
https://en.wikipedia.org/wiki/Sonic_ManiaI can't speak for SEGA legally of course, but they have a history of being quite lenient of fan projects. Sonic fandom is... notoriously quirky, shall we say. Good on them for tolerating, even encouraging it, despite how cringy it can be at times.
Which, not to waste a callback -- there's a
Sonic mod for Doom, or I guess it's a standalone thing all by itself, just based on the Doom engine, even.
And, needless to say, Doom is the poster child of porting things to unusual platforms; there are precious few on anything 8-bit, though. It seems to be just complex enough of a pile of algorithms, that there's almost no hope of porting much of it to such platforms, even stripping down the graphics and levels as much as you can. That said, consider this example:
There's also a clone (not really Doom engine as I understand it, but ground-up new along similar lines?) for Amiga 500, evidently quite a challenge:
For my part, I've put a flat-shaded raycaster (ala Wolf3D, without all the Wolf..) on an AVR:
also a basic DSP effects box, and I've demonstrated a few-MS/s-equivalent time sampling method on one; pushing certain limits of the system, but, granted, nothing as deep and comprehensive as the above examples.
Tim