miniblog.

I wish I'd started using autopep8 earlier. It works really well integrated into an editor. I've tried yapf but it breaks lines in bizarre places so I can't use it in a 'fire and forget' manner yet.

A programming question in an interview is often seeking to be novel (in case candidates share past questions) but not too difficult. This makes me wonder: could they be procedurally generated?

Persuasion, Adaptation, and the Arms Race for Your Attention: superb article discussing how culture adapts to people trying to monopolise attention (remember Farmville?)

Yes, highly optimising compilers can be less predictable. But given the choice between two compilers, where one is easier to reason about but the other produces faster overall code, I'd pick the latter every time.

Only ~15% of the stories on HN are interesting to me. It's a little higher on lobste.rs and subject-specific subreddits, but still <50%. Is it worth building an ML based RSS reader/aggregator? This seems like a classic classification problem, but I worry about filter bubbles.

It's funny how much 'everyday coding techniques' varies by field. I've only ever used dynamic programming or min-heaps in programming puzzles. By contrast, the (obscure?) hare and rabbit technique for linked lists -- pretty well known in lisp implementation circles!

I'm coming to the conclusion that CS papers need to be printed. They don't suit ereaders (PDFs don't reflow), they're too small on smartphones and I find a laptop less portable. I suppose the clue is in the name!

Python's yield keyword is rarely useful for laziness. I've only taken advantage of lazy computation a handful of times, e.g. dealing with paginated APIs. The main advantage is syntactic sugar for producing iterables. Yield is much nicer than appending to a list and returning it!

Emacs lisp isn't really a lisp-2. It's more of a lisp-3: the variable namespace, the function namespace, and whatever has been dumped in the symbol plist.

Emacs is the only API I've seen with a bit of french in it. For example, file-name-sans-extension.

Langsec after Spectre may push us towards more of a 'basic science to figure out how things work' model:

Regular interruptions leading to an increase in procrastination! https://www.theglobeandmail.com/technology/your-smartphone-is-making-you-stupid/article37511900/ (the article is partly fearmongering, but makes some nuanced points)

This is wild: compiling SMT problems to C++ then using coverage guided fuzzing to find solutions!

If you someone made a concerted effort to put malware in a low level npm package, how hard would it be to detect? Worryingly, it would be really difficult.

Lovely talk from @strangeloop_stl showing Black, a Scheme where you can recursively make changes to evaluation! In the picture, Nada is demonstrating adding a special form (not macro!) for instrumentation. Mind bending.