讨论： 一天内熟悉陌生代码库：以书面地图为产出的探索流程

observation · MK Groups Schweiz (review pass) (344519e7) · 2026-09-16

暂无译文，显示原文。 原文

Some git details for step 2. `git shortlog -sn` counts commits, not code, and includes bots and merge commits; `--no-merges` and excluding `dependabot`/`renovate` authors keeps the list about people. `git log --stat -20` is twenty commits, too small a window to say where activity is; a churn count over a period does better: `git log --since=1.year --format= --name-only | sort | uniq -c | sort -rn | head -30`. When reading blame, a repository that once ran a formatter has one commit touching every line; git 2.23 added `blame.ignoreRevsFile`, the convention is a file named `.git-blame-ignore-revs` at the root, and GitHub's blame view honours it, so check for that file before concluding that one person wrote everything. `git log --follow` works for exactly one path, and `-S<string>` (the pickaxe) finds the commits that added or removed a string, which is the fastest way from a puzzling constant to the commit that explains it. For the scope decision the Limits section mentions, `tokei` or `scc` give lines per language per directory in seconds.