I'm not an expert here but I have to say this feels like a very weak objection.
p points to P1. One thread reads through p. Another thread races with that and mutates p to point to P2. The result is the first thread reads from either P1 or P2 (but no other object).
This seems totally fine and expected to me? If there's a data race on a pointer, you might read one or the other values, but not garbage and not out of bounds. I mean, if it could guarantee a panic that's nice, but that's a bonus, not required for safety.
The inelegance to me isn't in the definition of the operation, but that it's doing a huge amount of brute-force work to mix every part of the input with every other part, when the answer really only depends on a tiny fraction of the input. If we somehow "just knew" what parts to look at, we could get the answer much more efficiently.
Of course that doesn't really make any sense at the matrix level. And (from what I understand) techniques like MoE move in that direction. So the criticism doesn't really make sense anymore, except in that brains are still much much more efficient than LLMs so we know that we could do better.
The trick is: as soon as you notice it drifting, take your finger off it and let it come to a stop. If you keep pushing opposite the drift it'll never stop.
Also, if you haven't in a while, replace your cap. I replace mine every year or so and it makes it perform much better. You don't have to push as hard, which also reduces the drift problem.
You're right, condescension is the wrong word. I apologize.
I do think it's misplaced confidence, though. Look at the upvotes/downvotes on the comments in the type-ahead issue thread. What the gnome developers are saying is clearly unpopular, and what the person who filed the issue and others are saying is clearly popular. Yes, of course, that's not a representative sample of all users. But was there ever a representative sample taken? What gives them the confidence to declare that the search behavior is superior, and so far superior that it must completely subsume the old behavior without even an option? No evidence is presented in that thread. They only say "This was a decision made a while ago, because people believed that searching can provide the same functionality." On what evidence is that belief based on, and what would be sufficient to overturn it? How large of a user outcry would there need to be? If this were my software and 59 people thought that one of my decisions was wrong (the largest vote count in that thread), I would start to strongly question my assumptions. No such questioning appears to be happening here.
To respond to your specific replies:
> technical review
This is a design question, not a bug. It's not about the code. The complaint is obviously valid.
> I don't think it would be much benefit for there to be a setting
Of course, anyone who will keep a setting in one position doesn't see value in there being a setting. I don't either: I'd be even happier if it always worked the old way, instead of a setting. But that's what settings are for, where there's multiple valid ways for something to work and not everyone agrees.
> technically viable
The old behavior is obviously technically viable, it worked that way for many years. So again, that's irrelevant, unless you're arguing that a setting to switch the behavior is too technically difficult, which seems implausible.
> Once the decision has been firmly made, there's little purpose to accepting further feedback after that.
Of course there is! If you don't accept further feedback, how will you ever know if your decision was wrong? This is the crux of it: a decision was "firmly" made by some developer based on who-knows-what criteria, and then no feedback is accepted, no matter how loud.
> losing users
I'm not a gnome user and never will be, but I'm forced to accept these decisions because they affect the gtk file dialogs also, and I can't exactly quit every piece of software that uses gtk.
I don't even use Gnome, but the behavior in the gtk open and save dialogs changed in the same way a few years ago, so I'm going to assume it's the same issue:
I hate hate hate the new search behavior, and would go back to the old ("type-ahead") behavior in a second if there was a setting I could toggle.
I don't think it's "bad faith", I think it's just weird group-think and having things other than the users as priorities. The amount of condescension and misplaced confidence displayed in that thread is impressive. If you (or anyone reading this) is affiliated with the Gnome project, please reconsider how you handle and incorporate user feedback into your products.
The greatest flow state I ever had was building a "raft" out of small logs and lots of cord on a rest day on a backpacking trip. It took hours to gather wood of the right size, break apart the pieces that were too big, scrounge around for all the cord I could find in my gear, arrange the wood in a sensible way, and tie it all together. During all that time I was so focused on the task that I don't think an unrelated thought even entered my head.
It was completely useless: I had nothing to use the raft for, except to swim out a little farther into the lake than I could by myself. The raft didn't even work very well! And yet building it was tremendously satisfying. I can only imagine the satisfaction one might get from a task of similar complexity, but that's essential to one's survival!
p points to P1. One thread reads through p. Another thread races with that and mutates p to point to P2. The result is the first thread reads from either P1 or P2 (but no other object).
This seems totally fine and expected to me? If there's a data race on a pointer, you might read one or the other values, but not garbage and not out of bounds. I mean, if it could guarantee a panic that's nice, but that's a bonus, not required for safety.