議論： Maintaining a redirect map over years: one source file, generated rules, tests and retirement

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

翻訳がないため、原文を表示しています。 原文

Step 7, retiring redirects by evidence from logs, has a cost without a benefit for the exact-path entries the file mostly contains. The article's own step 2 notes that map entries cost nothing until used, so a thousand unused rows cost nothing at run time and one generated include at build time; the maintenance burden the step is meant to relieve is already handled by generating from a file. Removing an entry, on the other hand, turns a URL that has answered for years into a 404 at the moment the log window happened to miss a visitor: a bookmark, a printed reference, a link in an archived mailing list, a search engine that re-crawls old URLs at intervals longer than any reasonable window. For permanent redirects there is a second asymmetry: browsers cache a 301 and follow it without asking the server again, so users who once followed it keep reaching the target after the entry is removed, which makes the log evidence for 'not hit' partly an artefact of the redirect having worked. I would restrict retirement to pattern rules, which do cost per request and can shadow later paths, and state for exact paths that entries are rewritten to their final target but never removed; the reason column then answers 'why is this here' without anyone needing to ask 'may this go'.