I think the concept of the "ego item" in gaming can be credited to Gary Gygax, back in 1E AD&D if not before. Literally, he meant an object with an ego, with its own personality. In gaming terms, these were powerful items whose egos meant they were not to be trifled with, expressed their views and could in extremis refuse to obey the wielder.
As ever with Angband the original intent of the designers is unclear, but over the years the term has come to mean "an item with one or more properties which its base item type doesn't have, excluding plusses to-hit/dam (on weapons) or to AC (on armour pieces)". The original concept of powerful arcane items with their own personalities is closer to what we think of as artifacts, and indeed some roguelikes actually express the personality of some artifacts quite explicitly (e.g. the Singing Sword in Crawl).
Having collected the stats from 25million dungeon levels in 3.3.0 we're working towards a fairly significant overhaul of item generation, to try and rebalance the prevalence of weapons and armour which has crept up on us over the last few years. As many people have observed, there are some deficiencies in the current system - two main ones being the distinction between "special" and other artifacts, and the granularity of the alloc_prob variable. There are also obvious problems with distribution: the vast majority of stuff is findable in the first half of the dungeon, making the second half less interesting and making potentially interesting items obsolete too quickly.
It seems to me that the current system of templates for ego items is quite limiting - you need a new template for every variation of an ego, such as if you want it to be quite common on boots and quite rare on armour, or permit a higher pval in one slot than another. The potential for random variation is currently limited to a single random sustain, random high resist or random ability, and even these are implemented in a spectacularly hackish way. There's also no reliable way of providing a smooth power curve from {average} through {good} to low-end ego to high-end ego. Although you ought to be more likely to find Slay Troll than Holy Avenger on dl20 and the reverse on dl80, that isn't noticeably true.
I've had item affixes in the corner of my mind for a couple of years now. EyAngband introduced a number of useful innovations, this one clearly inspired by the prefix/suffix system used in many modern CRPGs (Diablo, Dungeon Siege, Titan Quest et al.). While EyAngband used them as a bolt-on to the existing framework, it would seem eminently plausible to replace Angband's template system with an affix-driven system for magical and ego items. What follows are my initial thoughts on how this could work and why it would be an improvement - feedback would be most welcome.
First, let's distinguish between craftsmanship and magic. A set of affixes like sharp/keen/brutal of piercing/slashing/bludgeoning could refer to additional damage-dealing properties of weapons which result from superior craftsmanship rather than magic (e.g. +dam, or chance of 2x dam, or whatever). Similarly, affixes like superior or of deflection could imply similarly-derived properties for armour (e.g. increased AC, reduced weight). As an aside, this opens up interesting possibilities for making detection slightly less dominant if these items did not show up as magic.
Next, you have all the magical affixes, which should be divided into groups. These would include slays, brands, resists, stat boosts, protections, movement effects (speed, stealth, FA, lower weight, etc.) and so on. I envisage a one-to-one mapping of affixes to object flags, so that any code for picking affixes can provide any individual flag. (Note to self: get the OFT_ types right in list-object-flags.h and this grouping is already done.)
That gives you the basic framework for creating the equivalent of {good} items and low-end egos. What we now think of as {good} would be those with a non-magical affix (occasionally two) which purely boosted offence or defence, while those with one magical affix would be the equivalent of today's single-slay weapons and other one-trick egos.
Where it gets interesting is combining multiple affixes. You can combine a prefix and a suffix to get items ranging from mildly exciting to endgame (anyone remember Godly Plate of the Whale?). You can combine more than one of each to re-create the current high-end egos, with hundreds of other possibilities. You can combine craftsmanship and magical affixes to get even more permutations (sure, Gondolin weapons are special, but some of the smiths were better than others). The key thing about this is that it provides a route to a smoother power curve - the number of affixes would be a function of depth, with shallow items rarely getting more than one or two and really deep items rarely getting fewer than four or five.
Not all properties are equal so it would be sensible if there was some way of telling the generator that "of Dragonbane" is a more useful affix than "of Troll-slaying". I know that some people intensely dislike numerical power ratings, so I remain open to alternative ways of achieving this. In the meantime, list-object-flags.h already provides a rating for each flag, and a multiplier for each slot, which could be used as a starting point.
To avoid a total mish-mash of random permutations, we could tell the generator to recognise certain combinations and steer its later choices. This would enable us to retain the flavour of existing ego types without their limitations - we could specify the likelihood of each affix on a per-slot basis, and limit the later choices as much or as little as we wanted. (Hmmm, this item has one more affix to go, and it's already got SLAY_ORC and SLAY_TROLL - looks like a weak Westernesse weapon to me, so let's give it STR or DEX ... oooh we rolled a 1_in_6, let's bung in a sustain for free ...) It would also go hand-in-hand with naming items with multiple affixes - so for example anything with two of SLAY_EVIL, SLAY_UNDEAD and SLAY_DEMON gets the prefix "holy".
This proposal will mean rethinking the concept of randarts. With more randomness in non-artifact items, it becomes less important in artifacts. I think there is still a place for randarts, but how to distinguish them from these new ego items is a topic for another day. (My initial instinct is to randomise only *some* of the properties of each artifact ...)
So how are we going to balance this to make sure it doesn't flood the dungeon with egos (er ...) or, even worse, leave people struggling with a +0 whip until 2000'. That's what the stats module is for - rather than agonise over balance before writing it we can write it, run some stats, adjust it and iterate. It's non-trivial (the code for picking names is going to be tricky - fortunately there's a handy precedent in Sangband's class titles), but I think it's possible.
What do you think?
