¡Albricias!

When April began, kettle-api was a promising prototype with a long list of caveats in its README. By the end of June it was version 1.0, with the caveats struck through and, in several cases, the reasons for them removed from the code.

The quarter's work fell into three parts. April went to the data layer, where Tobias Wren replaced a hand-rolled query builder with something that did not leak. May was given over to authentication, and to the discovery that a timezone bug had been misfiling midnight bookings for a year. June was for the things nobody enjoys: documentation, a deprecation policy and a release checklist.

The numbers are tidy enough. Six hundred and twelve commits, forty-four pull requests merged, nine contributors besides the author, and a retry loop that retried forever, taught at last to stop after five attempts.

The tag went up at ten past six on the thirtieth, followed within the hour by a changelog of unusual candour. It lists, under “Known problems”, three things that are known problems.

The first day brought 380 downloads, a figure the author called “respectable for a Tuesday”.

Most were small: a typo, a missing example, a flag renamed for clarity. One, from a contributor in Lisbon, rewrote the pagination and was merged after a single round of review. All were thanked, which costs little and is remembered for years.

A May change brought the suite from ninety seconds to twenty-six. Continuous integration, which had been taking the afternoon off, was seen back at its desk. Left undone: the flaky test in the booking module, which failed twice in June and was mentioned in the commit log in language unfit for print.

The quarter did not run evenly. April, spent on the data layer, was the heaviest month by some way; May, with its authentication work and its long review threads, the quietest in commits and the loudest in comments; June, with the release at its end, a steady climb.

The installation section now lists the commands that work. The author reports no regrets.

Anything deprecated will stay for two minor versions and be named in the changelog. The policy is four paragraphs long and has already been quoted back to the author twice.

Of the pull requests that came to kettle-api this quarter, forty-four were merged and eleven were sent back for changes at least once. The most common reason, by a margin the reviewer found embarrassing, was a missing test.

Discovered in May by a customer's complaint and a single unlucky booking at 00:30, the bug had been quietly misfiling midnight appointments for twelve months. The fix is nine lines; the post-mortem is two pages and the better read.

Three repositories caught the author's eye this quarter, all of them in the author's own line of work: a query builder, a testing library and a small server that does almost nothing, very quickly.

Weekly downloads, a few dozen at the start of April, passed three hundred in the week of the release and settled, by the quarter's end, at a figure the author calls “enough to be paged about”.