Migrating from the Old Site

The old site is live and has bookmarks, search results and links from elsewhere. A migration that just switches it off turns every one of them into a 404. This chapter does what the Django course's chapter on migration did, in TypeScript and in Next.js's own configuration: measure exactly what the old site answers, suggest redirects for a person to approve, keep them in the app's configuration, check the links, and move one site at a time. Because the Django project has done the same measurements, every number here can be compared with an independent implementation.

Step 1: Measure, Don't Remember

The old site is a folder of built files (/hungary/x/y/ is the file hungary/x/y/index.html), so that folder is the complete list of old addresses. A check asks, of each one, whether the new sites have it, without starting a server: from their pages, the folders those pages sit in (each has a listing), the files that will be copied, and the decided redirects. An address is same, redirected (a decided redirect that ends on something that exists) or missing. A redirect to a page that does not exist is not success.

So most of the site already works at the same address, because the folder structure was kept. Reading the 80 missing pages: 32 are lessons whose folder was renamed or moved, 23 are _print pages (the old site had a printable twin of many pages), and 25 are the index pages of renamed folders, removed “tentative” course folders, and japan/kanji-tiles. These are exactly the numbers the Django project found, from a separate implementation.

Step 2: Redirects You Review First

A permanent redirect is remembered hard by browsers, so a wrong one is hard to undo. So: suggest, then review, then build.

1. Suggest. If exactly one new page has the same file name (ignoring capitals and - against _), propose it; if several do, list them; if none, say so; a name ending in _print falls back to the page without it. Result: 55 suggestions, 0 ambiguous, 25 with nothing to suggest, and all 55 point where the Django project's do.
2. Review. The suggestions go in a file whose decision column is empty. Only rows where you write yes become redirects. Loading the file before any review gives 0 redirects (55 rows skipped), so the lists that are committed are empty.
3. Build. next.config.ts reads the site's list and Next answers the redirects itself. A redirect's source is a pattern in Next, so the characters ( ) : * + ? { } in an old address are made plain. A page that moved to another site gets a full address.

Over Real HTTP

To see the mechanism work, I made a scratch list with every suggestion accepted (--accept-all, which prints “NOT reviewed”: 19 for languages, 8 programming, 11 systems, 11 AI, 6 humanities; not committed), built the languages app with it, and asked the running app about every old address the languages site owns:

The HTTP answers and the check from Step 1 agree on all 329 addresses. All redirects are 308. Note that an old link with a trailing slash takes two hops: first Next's own tidy-up to the address without the slash, then the real redirect. It works and each hop is cheap, but it is not one.

Step 3: Links

Broken links

Every absolute internal link of every page is checked against the pages, folders, files and redirects: 11,751 links, 483 broken (305 to pages, 178 to files), in a few seconds (not timed precisely). That is the Django project's result, category by category. 282 of the page links are to the hiragana and katakana tile pages, which exist on the old site and not yet on the new one; 174 of the file links are solution files under /web-development/scripting-and-backend that were moved or never made. It is not set to fail the build: 483 known problems would stop every build until each is decided.

Links between sites

A page written for the old single site that links to /linux/… breaks after the split, because /linux/ is on the systems site. The fix is made when the page is built, not in the files: a link into another site's folders gets that site's address in front. External links, // links, anchors, /search/ and the site's own links are left alone. On the real content this found 0 links to rewrite: every page links only inside its own site. So the function is tested only on made-up examples, and a function that has never met real data deserves suspicion.

Step 4: One Site at a Time

While the old site still answers on osztromok.com, a site moves by telling the old host's Apache to send that site's folders to its new subdomain. To go back, remove the line.

Without Apache, the same pattern was run as a regular expression over every old address: of 8,892 that a site owns, none was matched by the wrong site or by two, and none of the 144 that no site owns was matched by any rule (the Django project's numbers).

Hands-On Exercises