Online coding and Python classes in Headingley

Code-Point Open arrives as a zip of 120 CSV files holding 1,749,109 postcodes. Each row carries the postcode, a grid reference and the codes of the areas it sits in, including its council ward. The learner writes the same search twice. Version one reads every row into a Python list and then filters the list. Version two is a generator: a function that opens each file in turn and yields a single row, so the filter sees one postcode, decides, and lets it go before the next arrives. Python's tracemalloc module records the peak memory of each.

Both versions find the same 885 postcodes in LS6, and the same 491 whose ward code is Headingley and Hyde Park: 490 of those are in LS6 and one is in LS2, a reminder that postcode districts and wards do not line up. The difference is in the cost. The list version peaked at more than a gigabyte, roughly ten thousand times the generator's peak, and on our machine it was also slower, because building and later discarding 1.7 million row objects takes time. The times will differ on another computer; the shape of the result will not.

Generators are not free. You get one pass: once a row has gone by, it is gone, so anything that needs the data twice, such as sorting, must store it after all. Choosing between the two is a design decision, and the honest way to make it is to measure.

A Headingley teenager who thinks about scale before running code will catch what a chatbot's tidy example leaves out, and Python is where that instinct is built. The longer argument is in why teenagers should still learn to code in 2026.