zit 0.3.41

zit 0.3.41 makes the crosslink line counts opt-in. A forge that shows "+15 −3" on its timeline asks for them with zit crosslink --lines; every other caller gets the stream 0.3.39 printed, at 0.3.39's cost. And zit merge-tree --commit now records a clean merge, so a forge can merge from the web without a checkout.

No artifact, tag namespace, wire, or on-disk repository format changed. The crosslink record format is still crosslink 1.

Action for forge clients

If you read lines lines from zit crosslink, add --lines to the call. Without it, 0.3.41 prints no lines lines at all. A consumer that does not read them needs no change.

Line counts are opt-in

0.3.40 counted every changed file of every check-in the first time crosslink ran over a repository, whether or not the caller wanted the counts. On a long history that first run took many times longer than it used to. Now:

• zit crosslink counts nothing, prints no lines lines, and reads and writes no line-count cache.
• zit crosslink --lines prints them exactly as 0.3.40 did: the counts zit diff --numstat --checkin gives, - - for a binary file, 0 0 for a mode-only change, and no lines line when the store cannot produce a side's content.
• A cache written by 0.3.40 is reused as it is.

A shun no longer recounts everything

A check-in's counts are fixed by its name, so they are counted once. Under 0.3.40 a shun threw every stored count away. Now a shun recounts only the check-ins that touch a path a shunned blob was written to: the check-in that wrote it, and any later one that read it as the parent side. Those lose the lines line for that file; everything else is served as stored.

The one exception is a history that holds a check-in whose changed files are unknown. A shunned blob could sit at a path nothing records, so the counts are taken again in full.

merge-tree --commit records a clean merge

Before, --commit recorded a merge only when it had conflicts to settle, and refused a clean one with "there is nothing to resolve". The clean case is the common one, and it had no path that worked on the store alone. Now:

zit merge-tree feat --into trunk --commit -m "merge feat" --user u \
  --expect-current TRUNK_TIP --expect-other FEAT_TIP

records the merge on trunk, the check-in zit merge feat && zit commit would make in a checkout of trunk. --expect-current and --expect-other still guard it, and any --resolved file or --delete path is refused, since nothing conflicts. The source may be any check-in, not only a branch tip.

Compatibility

• zit crosslink without --lines is byte-identical to 0.3.39.
• zit crosslink --lines is byte-identical to 0.3.40, except for files a shun has made uncountable.
• --lines --no-cache still matches the cached run byte for byte.
• merge-tree --commit on a conflicting pair is unchanged.