Skip to content

Latest commit

 

History

History
229 lines (192 loc) · 12.9 KB

REDIRECTS.md

File metadata and controls

229 lines (192 loc) · 12.9 KB

Redirect rules

Reference for the Cloudflare-side redirects that carry old foojay.io / WordPress URLs to their Hugo equivalents. This is dashboard-only config — none of it lives in this repo — so this file is the only record of what's actually configured.

Everything the WordPress Redirection plugin served has been carried over, and 88 of its 92 concrete rules are aliases: in content/ — per-URL, so Hugo emits a redirect page for each, nothing to configure and nothing to forget. What follows is only what an alias cannot do: three regexes from the plugin export, plus two families the export never knew about.

A third thing the export never knew about: _wp_old_slug. Rename a post in WordPress and it keeps the old slug in postmeta and redirects from it for ever, with no plugin rule to show for it. Nine such URLs were live and would have 404'd here; they were added as aliases: (2026-09-21), found by diffing the SQL dump against content/:

SELECT p.post_name, m.meta_value AS old_slug
FROM wp_postmeta m JOIN wp_posts p ON p.ID = m.post_id
WHERE m.meta_key = '_wp_old_slug';

Same lesson as rules 4 and 5 below: the plugin table lists redirects somebody added, and every rename since is invisible in it. Two more rules (ids 106 and 107, the commit-created… chain) postdate the export and are aliases now as well.

Rules 4 and 5 were not in the plugin export, which is why they were nearly missed: it lists redirects somebody added, not the URLs WordPress serves by virtue of being WordPress. They are invisible in a sitemap comparison too, because Yoast lists only the canonical form.

These are for INBOUND traffic, and that is the whole reason they matter. HtmlToMarkdown.normalizeLegacyUrls applies rules 1–3 at scrape time, so a post stored in content/ already links to /today/… — our own markup does not depend on them. What does is the 312,531 hits arriving from other sites, search results and bookmarks.

Cloudflare → Rules → Redirect Rules, ruleset "Foojay cutover redirects". All five families answer 301/permanent (flipped from an initial 302 window that let a wrong rule be walked back before it got cached in browsers). The plugin matches case-insensitively and ignores a trailing slash (flag_case: false, flag_trailing: false), so the replacements do too.

# 1. the old blog scheme -- 209,365 hits, foojay's original URL scheme. Also
#    covers /blog/author/…, /blog/category/…, /blog/page/2/ and the feeds, which
#    per-post aliases could not.
When:  (starts_with(http.request.uri.path, "/blog/"))
Then:  concat("/today/", substring(http.request.uri.path, 6))     dynamic, 301

# 2. the almanac -- 102,636 hits. Off-site: never foojay's own content.
#    Only matches the versioned form (jdk-N / java-N); bare /almanac/ and
#    /almanac 404 rather than redirecting to javaalmanac.io's homepage -- a
#    known gap, left as-is by decision (won't fix).
When:  (http.request.uri.path matches "^/almanac/(jdk|java)-([0-9]+)")
Then:  regex_replace(http.request.uri.path, "^/almanac/(jdk|java)-([0-9]+).*$", "https://javaalmanac.io/jdk/${2}")
                                                                  dynamic, 301

# 3. the retired docs section -- 530 hits; everything under it collapses to the
#    article index.
When:  (starts_with(http.request.uri.path, "/docs/"))
Then:  "/today/"                                                  static, 301

# 4a. WP categories NEST and Yoast canonicalises to the nested form, so
#     /today/category/tools/maven/ is the INDEXED url while Hugo has only the
#     flat one -- 55 URLs plus their page/N/ and feed/ variants. 41 of them
#     differ only by the parent segment, so one rule covers the lot.
#     Cloudflare's regex engine has no negative lookahead, so the page/ and
#     feed/ exclusion is a second, negated condition rather than "(?!page/|feed/)".
When:  (http.request.uri.path matches "^/today/category/[^/]+/[^/]+/"
        and not http.request.uri.path matches "^/today/category/[^/]+/(page|feed)/")
Then:  regex_replace(http.request.uri.path, "^/today/category/[^/]+/", "/today/category/")
                                                                  dynamic, 301

# 5. every WordPress feed URL -> its Hugo equivalent. WP serves a feed at /feed/
#    and at <any archive>/feed/; Hugo serves index.xml beside every one of those
#    pages, so this single rule covers /feed/, /today/feed/,
#    /today/author/<slug>/feed/ and a category feed.
When:  (http.request.uri.path matches "^(/.*)?/feed/?$")
Then:  regex_replace(http.request.uri.path, "^(.*?)/feed/?$", "${1}/index.xml")
                                                                  dynamic, 301

Order matters in three places. Rule 1 before any catch-all, and none of 1–3 may fire for /today/… itself. Rule 4b (the 14 renames, below) sits before 4a, or 4a strips the parent off the six tools/… ones and lands them on a term that does not exist. Rule 5 sits after 1 and 4, so /blog/feed/ and /today/category/tools/maven/feed/ are normalised first.

Three traps, each of which fails silently.

  • The page/|feed/ exclusion in 4a is load-bearing (written as an and not condition, since Cloudflare has no negative lookahead): without it /today/category/tools/feed/ rewrites to /today/category/feed/ and /today/category/java/page/2/ to /today/category/page/2/ — breaking two URL shapes that work today in the course of fixing a third. Because the replacement only strips a prefix, a pager or feed under a nested category still lands correctly on /today/category/maven/page/2/.
  • A feed URL cannot be an aliases: entry even in principle. A Hugo alias is an HTML page carrying <meta http-equiv="refresh">: a browser follows it, a feed reader does not, so every subscriber would get HTML where XML belongs — worse than a 404, because it looks like a working response.
  • /comments/feed/ has no equivalent — there is no site-wide comment feed here — so it falls through to the 404 rather than aiming it at something that is not what it claims. Rule 5 as built does catch it, sending it to /comments/index.xml, which 404s anyway; the outcome is the same status code by a worse route.

The 14 nested-category URLs (rule 4b) are renames: the slug itself changed, so no pattern derives them and each needs its own rule. Both columns are under /today/category/ unless shown otherwise:

WordPress here
ai-ml/ /ai/ — see below
books/book-reviews/ book-review/
game/ game-development/
interview/ interviews/
jakartaee/ jakarta-ee/
survey/ surveys/
tools/cassandra/ apache-cassandra/
tools/deepnetts/ deep-netts/
tools/idea/ intellij-idea/
tools/pulsar/ apache-pulsar/
tools/tomcat/ apache-tomcat/
tools/vscode/ vs-code/
tutorial/ tutorials/
uncategorized/ /today/

ai-ml is WordPress's "Machine Learning" category, so the literal equivalent is /today/category/machine-learning/. The live rule still targets /ai/, which is now a Hugo alias of /today/category/ai/ (content/categories/ai/_index.md), so it lands there in two hops. Point the rule at /today/category/ai/ directly when someone with Cloudflare access can.

Rule 4b, the renames

Position: FIRST in the Single Redirects list, above 4a and above 5. That position is load-bearing, not a preference. Redirect rules stop at the first match, and 4a strips the parent segment off any nested category path, so with 4b second tools/vscode/ becomes /today/category/vscode/ and 404s before 4b is ever consulted. Rule 5 has to stay below for the same reason, otherwise tutorial/feed/ turns into /today/category/tutorial/index.xml.

Settings: type Dynamic, status 301, Preserve query string ON. Regex needs the Business plan or above, which 4a already relies on.

When incoming requests match:

http.host eq "foojay.io" and http.request.uri.path matches "^/today/category/(ai-ml|books/book-reviews|game|interview|jakartaee|survey|tools/(cassandra|deepnetts|idea|pulsar|tomcat|vscode)|tutorial|uncategorized)(/.*)?$"

Target URL, as an expression:

concat("https://foojay.io", regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(regex_replace(http.request.uri.path, "^/today/category/ai-ml/?$", "/ai/"), "^/today/category/ai-ml/(.+)$", "/today/category/machine-learning/${1}"), "^/today/category/books/book-reviews(/.*)?$", "/today/category/book-review${1}"), "^/today/category/game(/.*)?$", "/today/category/game-development${1}"), "^/today/category/interview(/.*)?$", "/today/category/interviews${1}"), "^/today/category/jakartaee(/.*)?$", "/today/category/jakarta-ee${1}"), "^/today/category/survey(/.*)?$", "/today/category/surveys${1}"), "^/today/category/tools/cassandra(/.*)?$", "/today/category/apache-cassandra${1}"), "^/today/category/tools/deepnetts(/.*)?$", "/today/category/deep-netts${1}"), "^/today/category/tools/idea(/.*)?$", "/today/category/intellij-idea${1}"), "^/today/category/tools/pulsar(/.*)?$", "/today/category/apache-pulsar${1}"), "^/today/category/tools/tomcat(/.*)?$", "/today/category/apache-tomcat${1}"), "^/today/category/tools/vscode(/.*)?$", "/today/category/vs-code${1}"), "^/today/category/tutorial(/.*)?$", "/today/category/tutorials${1}"), "^/today/category/uncategorized(/.*)?$", "/today${1}"))

Fifteen regex_replace calls for fourteen renames, nested so each one feeds the next. A path only ever matches one of them, and the order between them does not matter, with one exception noted below.

Three things in there that look odd and are deliberate:

  • (/.*)?$ carries the rest of the path through, so tutorial/page/3/ reaches tutorials/page/3/ and tutorial/feed/ reaches tutorials/feed/, which rule 5 then turns into index.xml on the client's second request.
  • It is also what stops a rewrite cascading. survey → surveys cannot be re-matched by the survey pattern afterwards, because the group demands / or end-of-string and finds s. The same holds for interview and tutorial. Loosen those anchors and the chain eats its own output.
  • ai-ml gets two lines, and they are the one ordered pair. The bare category goes to /ai/, see above. Anything deeper goes to /today/category/machine-learning/, the literal term page, because the AI term's page/2/ and feed hold different posts. The bare-path line runs first so the subtree line cannot claim it.

Verifying

Each should answer 301 with the destination shown above. Add -L and check the final code, not just the hop, or a redirect to a 404 reads as a pass.

for u in /blog/log4j-cve/ /blog/author/hirt/ /blog/category/java/ \
         /almanac/jdk-17 /almanac/java-8 /docs/anything/ \
         /today/category/tools/maven/ /today/category/tools/maven/page/2/ \
         /today/category/jeps/records/ /today/category/tools/vscode/ \
         /today/category/ai-ml/ /today/category/tutorial/ \
         /feed/ /today/feed/ /today/author/frankdelporte/feed/ \
         /today/category/java/feed/ /today/category/tools/maven/feed/ ; do
  printf '%-42s ' "$u"
  curl -s -o /dev/null -w '%{http_code} -> %{redirect_url} ' "https://foojay.io$u"
  curl -sL -o /dev/null -w '(ends %{http_code})\n' "https://foojay.io$u"
done

# The two that must NOT move. 4a is written to leave them alone and a wrong
# negative lookahead is invisible otherwise -- both must stay 200.
for u in /today/category/java/page/2/ /today/category/tools/ ; do
  printf '%-42s ' "$u"
  curl -s -o /dev/null -w '%{http_code} (expect 200)\n' "https://foojay.io$u"
done

What was deliberately NOT carried over

17 plugin rules point at pages that 404 on the live WordPress site too — the rule outlived its target, and recreating one would mint a redirect to a missing page, which is worse than a 404 for readers and crawlers alike. Recorded here so nobody rediscovers that they were skipped on purpose: /command-line-arguments/ and its six openjdk-NN-command-line-arguments variants (3,725 hits) plus /cli (6), the section being gone from WordPress; the six China JUG aliases (/china/, /china-jug/, /jugchina/, /jugs-china/, /jugs/china-jug/, /china-java-user-group/, 1,529 hits), which chain to /jugs/china/ — per-JUG pages exist on neither site, /jugs/ being one directory page built from data/jugs.yaml; /foojay-day-live/ and /foojaydaylive/ (2), which chain to a gone /foojayday2022live/; and /java-learning-trail/learn-more-on-foojay/ (0).

One export rule is disabled (/calendar/ → /all-events/) and was skipped for that reason; this site resolves that pair the other way round anyway (see the calendar note in AGENTS.md).

/almanac/ and /almanac (bare, no version) 404 rather than redirecting. Decided not to fix — left as-is.