While I agree it sucks if the random generator sometimes gives you a bum situation, the solution to that is constrained generation. Which means many situations may just never occur. Since it's really impossible to determine actual winnability for a complex roguelike, you end up being over-constrained just in case, to ensure all builds are winnable. It's like saying "every movie must have a happy ending" and expecting that this isn't going to seriously restrict the types of stories that can be told.
If devs spent more time checking that 100% of random builds are winnable, that's a non-trivial AI type problem right there. It's like proving white or black can always win from a given chess position. There
is an answer, but it's beyond the most powerful supercomputers we have, so you'd only generate "trivial" chess positions if you wanted to be able to say "white (the player) can always win from this position".
A significant chunk of developer's time and money would need to be spent. The alternative would be massively simplifying the types of builds that can exist, so that they can be automatically checked more easily, or just making the game much easier so you know you can always win no matter what it throws at you. So, you say that the player can always run away and regenerate health, and the player can regenerate health faster than the hardest boss. Now, I've made a game you're guaranteed you can always win no matter what it generates. But is it a better
game?
EDIT:
I'm reading the article right now, and the guy doesn't even mention Diablo. Diablo was specifically developed to be a more AAA roguelike.
During development, Condor was eventually renamed Blizzard North.[20] The basis of Diablo was the 1980s roguelike game Moria;[21] Bill Roper later said that the team's "initial pitch, in a nutshell, was to take the excitement and randomness of games like Moria, Nethack, and Rogue, and bring them into the 1990s with fantastic graphics and sound". The switch from turn-based to real-time gameplay occurred roughly six months into production.
Diablo has hardcore mode as well, with permadeath, which is the main thing that the "Diablo isn't a roguelike" people would complain about.
The guy basically doesn't know what he's talking about, putting the entire roguelike resurgence down to a Nintendo DS port in 2008.
He then has a "why so popular" section and boils it down to two reasons. (1) it's easier (cheaper) for the devs to make random levels and (2) he thinks players are confusing it with other types of games. Neither of these things makes any sense. "Easier to make" doesn't automatically make something more popular, and the roguelike genre's been around long enough that just about anyone picking one up can't possibly be "confusing" it with the qualities or aspects of any other type of game. This article is so poorly researched as to make it worthless for toilet paper.
In Rogue and the first wave of roguelikes, permadeath was a simple byproduct of the fact that “saving” wasn’t a common concept in games yet; most games from the time -- even PC games with ready access to writeable media -- didn’t have saves. Shiren did have saves, but it treated the technological hindrance of its predecessors not having saves as a feature by adapting it into permadeath.
Yeah, bullshit. The very first version of Rogue didn't have saves because it was running on a mainframe, but as soon as they ported these games to PC, saves became normal. Permadeath was a deliberate design decision from the very first wave of roguelikes.