Reading a big file into a list can kill your program. A generator handles one line at a time. See the memory difference measured on a real file, learn what yield does, and chain small generators into a pipeline.
asyncio.gather runs many waiting tasks at once in one thread and returns results in order. Avoid await in a loop and blocking calls, keep failures with return_exceptions=True, and limit concurrency with a Semaphore.
Don't guess which part is slow. Run cProfile, read cumtime and tottime the right way, fix the one line that matters, and measure again. In the example, the obvious suspect isn't the problem.
Downloads spend most of their time waiting. ThreadPoolExecutor lets many waits overlap: 20 half-second downloads in 2 seconds instead of 10. Handle failures with as_completed, pick a sensible worker count and respect the server.
Six short snippets with setTimeout, promises, await and queueMicrotask. Predict the order, run it, then read why. The one rule: sync code first, then all microtasks, then the next task.
Send many requests, keep the successes, and report the failures instead of losing everything to one rejected promise. Plus limiting how many run at once.
fetch has no timeout option and can wait a very long time. AbortSignal.timeout(ms) fixes that in one line, and AbortSignal.any combines it with a cancel button.
Read a CSV too big for memory line by line with readline, total a column, and write a filtered copy with pipeline. Memory stays flat no matter how big the file is.
Trace a detached preview from its DOM node to an application Map, repair cleanup, and verify repeated open-and-close cycles with heap snapshots.
Record one slow click in Chrome's Performance panel, read the Bottom-up tab by self time, fix the function it points at, and record again to prove the fix worked.
Compute size-adjust and the vertical overrides from the font files, build the fallback face, and measure the layout shift before and after.
Measure the real slot width at each breakpoint, write sizes from it, and check which file loads at each screen density.