Pages redirects

This blog uses CloudFlare Pages to host all its static content. To redirect old URLs to new ones, I use a _redirects file at the root of the repository . You should inspect that file to see how it works.

Note that this kind of redirects are not working at the domain-level: Pages rules match on path only, never on host. If you want to redirect https://example.com/ to https://www.example.com/ , you need a zone-level Single Redirect matching on http.host (I used Page Rules for this before Cloudflare deprecated them).

Here are some of the most useful rules:

• By default, pages are redirected for normalization  :

  From	To
  /contact	/contact
  /contact/	/contact
  /contact.html	/contact
  /about	/about/
  /about/	/about/
  /about/index.html	/about/

  Note how folder roots with an index.html are always redirected to an URL with a trailing slash, while HTML files are stripped of their extension.

  This behavior cannot be changed.

• Redirect a single URL:

  /old-url /new-url 301
  

• Redirect empty folders to the root of the site:

  /empty-folder      /   301
  /empty-folder/     /   301
  

  I use this double rule to catch both the folder named empty-folder , and an hypothetical file named empty-folder.html . That way I am sure a parasitic empty-folder.html won’t be served.

The rule budget is positional

Update from 2026: I lost the last 18 rules of my _redirects file for years to an accounting subtlety the documentation does not state. I only found it by reading the parser in cloudflare/workers-sdk ( parseRedirects.ts , also bundled in wrangler and Miniflare):

• A rule only counts against the static budget (2,000) while it appears before the first rule containing a * or a :placeholder  .

• From that first dynamic rule on, every line burns the dynamic budget (100), even plain /old /new 301 ones.

• At rule 101 of that mixed stream the parser does not skip a line: it stops reading the file , and everything after is silently discarded. Nothing in the deploy output tells you.

So the shape of a large _redirects file is not a style choice: put every exact rule first, every wildcard and placeholder rule second, or the tail of your file does not exist.

Two matching facts from the same source ( rules-engine.ts ) worth knowing:

• Sources are matched as anchored regexes: :name compiles to [^/]+ (never empty, never crosses a slash) and * to .* (may be empty). /a and /a/ are therefore different sources, and neither matches the other.

• Because a splat may match empty, a rule ending in /* also answers the bare trailing-slash URL. WordPress URLs all ended with a slash, so for old blogs this empty-splat behavior is what keeps two decades of inbound links alive.

I keep a faithful Python replica of the engine in this blog’s repository, and a test suite that parses my _redirects with it and probes every rule against production. The full write-up lives in docs/redirects.md  .