Hasnain says:

“Two weeks ago, the Supreme Court proved that it valued Trump over the rule of law when it granted him absolute immunity. Today, Aileen Cannon proved that she values Trump over the rule of law when she dismissed the charges against him. The rule of law is not failing—it has failed.

We live in frightening times because the rule of law has failed. America is right now, functionally a failed state. It is wrong and intellectually facile to pretend otherwise. I don’t have the solution, but I know we’ll never find one if we don’t accept the reality of our situation.”

Hasnain says:

“So Blue Maga may not be violent or authoritarian like Red Maga, but the consequences of a second major political party in the United States succumbing to a weird online cult of personality could be disastrous for our democracy.

Democrats who have spent the past decade (rightly) attacking Republicans for their blind and zealous devotion to Trump would do well to heed the advice of Friedrich Nietzche: “He who fights with monsters might take care lest he thereby become a monster.””

Hasnain says:

This certainly started with a compelling hook.

“So how successful has your program been?

The results speak for themselves. A bunch of our people went to the Olympics in 2021. At the most recent world championships, every female American gold medalist in individual events was a UVA athlete. Kate Douglass showed up here at UVA a few years ago, swimming the 200-meter breaststroke in two minutes and 30 seconds. Now she’s the American record holder, with a time of two minutes, 19.30 seconds. She just broke the U.S. Olympic trials’ all-time record, and she’s a favorite to win the Olympics this year.

Nine UVA athletes, including Kate, just became U.S. Olympians — one-fifth of the U.S. team! Gretchen Walsh won the 100-meter butterfly, setting the world record. Paige Madden got second in the 400-meter freestyle, right after Katie Ledecky; Paige is now a two-time Olympian.”

Hasnain says:

“While Russia’s invasion of Ukraine “has complicated things,” Elliott said, the collaboration between the U.S. and Russia on BEST is still ongoing, for now. Barinov says the team at Baksan is considering using a new source of neutrinos, such as zinc, to further test the result. They may even construct a third chamber of gallium around the source. For now, the anomaly remains unsolved, with no sign of a resolution on the horizon. “It has us all puzzled,” Haxton said.”

Hasnain says:

“Was there an earlier moment when you realized you wanted to be a scientist?

The honest truth is that I did not always want to be a scientist. At the point of entering college, I was like, maybe I will double major in biology and English. I loved biology in particular, and I’ve always loved writing, so I thought maybe I’d be a writer.

I have always been very interested in everything. It’s a common refrain for astronomers to say, “Oh, ever since I was a little kid, I absolutely loved space, and I knew that’s exactly what I wanted to do when I grew up.” And I definitely loved space as a little kid, but I also loved rocks, and dinosaurs, and lizards. Salamanders in particular. If anything, it all started with looking under rocks for salamanders with my sisters in the backyard in Virginia. It’s just a curiosity about nature and a love of learning, and that’s what you get to do as a scientist.”

Hasnain says:

“I spent a long time on this post. It was difficult to balance realism against ease of understanding, but I feel good about where I landed. I'm hopeful that being able to see how these complex systems behave in practice, in ideal and less-than-ideal scenarios, helps you grow an intuitive understanding of when they would best apply to your workloads.”

Hasnain says:

“And this is how you properly test concurrent data structures.

Postscript
Of course, this is just a toy. But you can see some ways to extend it. For example, right now our AtomicU32 just delegates to the real one. But what you could do instead is, for each atomic, to maintain a set of values written and, on read, return an arbitrary written value consistent with a weak memory model.

You could also be smarter with exploring interleavings. Instead of interleaving threads at random, like we do here, you can try to apply model checking approaches and prove that you have considered all meaningfully different interleavings.

Or you can apply the approach from Generate All The Things and exhaustively enumerate all interleavings for up to, say, five increments. In fact, why don’t we just do this?”

Hasnain says:

This approach is nuts. Learnt a lot about algorithm design and also about high performance micro-optimizations

"The program is over-fit to the input spec and the particular host it runs on (Intel Xeon E3-1271 v3 @ 3.60GHz, 512MB RAM, Ubuntu 20.04). Given the CPU, it only uses SIMD instructions up to AVX2, no AVX512. It assumes the input is exactly according to the spec and hence does zero error handling and even on such input will only produce correct results with probability < 1, though very close to 1, depending on the parameters you choose."

Hasnain says:

"17 TiB of tablebases are unwieldy, so to do this migration without hours of downtime, we set up a second server with the new approach. This also allowed us to run controlled benchmarks on the full set of tablebases, before finally doing the switch and retiring the old server."

Hasnain says:

“Aaaand the quick and easy solution is rarely quick or easy. This was never meant to be so arduous - this whole “let’s make a primitive function” idea was meant to take an hour tops to avoid spending a couple hours implementing it at the bytecode level, but it ended up taking much longer! I don’t think I could have easily predicted this seemingly easy change would be so hard: this requires a lot of knowledge about the system you’re working with and about interpreter design, yet I’m but a humble PhD student. We’re getting there though.”