i do wonder if the test styles in use aren't keeping up with catching unknown unknowns introduced by ongoing development though: myself and my colleagues on the lix project has spent weeks chasing down three separate bugs in curl since the start of this year (two of which were iirc regressions) relating to various misbehavior related to http/2 and stream pauses. i suspect we simply exercise that feature far more than either anyone else or the test suite.

i wonder if some sort of generative property tests (maybe using a grammar to give possible actions of the server and curl user which are then selected at random) is necessary to reduce the number and severity of state machine bugs reaching users. or maybe doing this genre of property testing with antithesis.com.

Antithesis: autonomous software testing

Try the Antithesis autonomous testing platform and find bugs in your software with perfect reproducibility

^{antithesis.com}