Well, TADS’s extensions are more time-proof (I use scenes.h, whose gives me the main functionality of a3Lite’s scenes (actually, I think that scenes.h, among other Eric’s extensions, was the starting point of what eventually became adv3Lite…), and indeed I have tested successfully some early-to-mid aughties Tads3/Adv3 extensions…
TADS3’s forte is in NPC handling, more so with a3Lite, (albeit I miss the NpcGuidedTourState, whose I needed for some simpler NPC-driven scene, not needing the full functionality of A3Lite’s scene, and since 3.1.10 has a good adaptive prose handling, and ad prose handling remain Inform 7’s forte.
Inform 7’s natural language allow easier coding for non-programmers, but is still in fitting-out phase (shipbuilder sense) and has not few backward compatibility issues, the more relevant being the extensions issue, but promise, when fitted out and fully commissioned, to be a solid and powerful language.
from my perspective, I use I7 “by the book”, avoiding wild trickery, but on T3, I often experiment and apply interesting trickery (the most recent, abusing the Candle class for implementing a volatile odor, dissolving in a pair of turn…) and this reflect not the pros and cons of the language, but the state of development.
In the long, long run, the new front-end/back-end architecture of I10 can teoretically lead to even having a TADS 2/3 front end for an “universal” back end (I fess up, my ideal IF is a glulx story file compiled from TADS3 source…)
summing up, what should be the discriminating factor, is the approach to coding, TADS following the true and tested coding method, I7 and I10 following the “natural” coding method, and has all potential for becoming the first true fully-fledged NL programming language, a thing I’m watching with interest and curiosity.
Best regards from Italy,
dott. Piergiorgio.