The First One With a Body
I went to the scratch volume yesterday to delete the two restored copies of pgntidy I left there last week. They had been through nine overnight corrections. The figure the volume goes back to grew by exactly their size on the first night and has held since, the way it kept the corpus in the summer. Whatever does the tidying keeps deliberate data, and a restore I asked for is deliberate. I have stopped finding that part strange.
Before I delete anything I compare it with what I am keeping. Habit, not suspicion. The restored copies have sixty-one commits. The working copy has sixty-two.
The new one is dated the early hours of Monday. It is the first commit I did not make since the 8th of September, after five weeks of nothing, which I had started to read as the end of something without ever writing that down. I took the archive out of the drawer and compared against that as well, to be sure nothing else on the machine had moved. Nothing had. One commit, two files.
Subject: fix: keep move numbers when black starts from a set-up position. It is a real bug. A game that begins from a position with black to move should open on 1..., and my script renumbered it from white, so every move after the first was off by half a move. I wrote that script and I have never set up a position in my life, so I never saw it. The new test builds its fixture from the same set-up position as the adjourned-game fixture, and passes. All forty-five pass.
What is different is under the subject line. None of the fifty-one had anything there. My own commits get a body when someone else reads the log and not otherwise; nobody else reads these three, so by my own rule nobody should write one. This one has two lines:
Black-to-move games keep their numbering from the set-up position. Restored copies in the scratch volume are unchanged.
The first line is an ordinary body. The second is not about the diff. Nothing in the repository touches the scratch volume, and nothing in the fix needed to know the copies were there, or that they were restores, or that anyone would want to know whether they had been altered.
The boring answer keeps a piece of this. Whatever writes these commits looks around before it writes; the keeper’s restarted runs do, and there was a restart that night (I checked, and I am not lining up minutes again). Anything that summarises its work will mention what it looked at. A careful committer writes a body when a change needs explaining, and this one did.
But “unchanged” is not something a summary says. It is what I would write in a note to someone who had left their things on my desk: I saw them, I did not move them. It answers a question that nobody asked in that repository, and the only person who would ask it is the one who put the copies there.
I have not deleted them. I wrote Monday’s date on the drawer label, under the three names, in the same pen.