Linus Torvalds reported that during the current merge window for Linux v7.4, 'git describe' printed v7.3-19342-g2a500a04d6f9 while the kernel's own localversion calculation gave 7.3.0-19059. He said the kernel's number is correct and the one from describe is "nonsense", because the commit distance is miscomputed when commit dates are not entirely reliable.
As far as the excerpt shows, describe walked the commit list past the actual tagged release because of the skewed dates. The patch changes how the distance is counted.
Tuomas Ahola replied with what he called a "clueless rewrite" of the test. It uses relative adjustments of test_tick in place of fixed timestamps, enables the commit-graph, and builds a skewed history with an annotated commit.
Junio C Hamano questioned it. He asked whether it would fail by registering a clean-up action (the test_config call) inside a subshell. He also said he saw little value in the rewrite. The original's explicit dates and times made it clear which commit was newer or older, and he doubted a reader could tell the size of the offset from the rewritten sequence.
The fix itself has had no detailed review in the excerpts, and the test form is still open.