This week the php-src repository tagged php-8.2.35 — no features, no launch event, a changelog with the emotional register of a building inspection — and the release climbed to the top of Hacker News, where it collected 912 points and 510 comments.
That is not a comeback. It is an inventory count.
The Crowd Cheering 8.2.35 Is the Crowd That Can’t Move
Read the thread and a specific population emerges: people running 8.2. Not people who picked 8.2 last quarter out of conviction. People whose code is welded to a content management system, a checkout extension, or a plugin whose last commit predates the current junior developer’s college degree.
The PHP Group has had the 8.2 branch in security-only maintenance for over a year now, and its security-support window closes on Dec. 31. So a point release on that branch is, by construction, a gift to the people who have not moved — and the applause is gratitude with a little dread underneath it. The tourists show up for feature releases. The installed base shows up for patches.
This is what makes the standard reading so lazy. Every few years the same column gets written: PHP is back, boring technology wins, the haters were wrong. But nobody upvotes a point release because they’ve been persuaded by an argument. They upvote it because a stranger in a different timezone fixed something in a branch they cannot leave, for free, at the last possible moment.
An End-of-Life Date Is a Staffing Forecast
Large organizations have spent a decade converting “end of life” into a procurement artifact — a checkbox on a vendor security questionnaire, a line in an audit evidence folder, a slide in a board deck with an amber status light. Compliance departments love this framing because it turns an engineering problem into a document problem, and document problems can be delegated to whoever answers the questionnaire.
The framing is comfortable and it is wrong. A support date is a forecast of unscheduled labor. It tells you how many developer-weeks your organization intends to spend, in a compressed window, on work it has not budgeted, using people who already have a full queue and no incentive to touch a system that is currently, technically, working.
That is the real cliff. Not the date — the queue. Every institution that has quietly decided to run a decade-old CMS until the wheels come off has also decided, without saying so, that its two most competent engineers will spend a quarter of next year doing archaeology instead of the project their boss promised the board. The compliance box gets ticked. The roadmap does not survive contact with reality.
The contractors already know. “Nobody in that room is arguing about whether we upgrade,” said one who maintains the public-facing sites for a cluster of county offices, standing in the parking lot after a change-advisory meeting. “We’re all arguing about whether the upgrade comes out of the same line item as the roof. It doesn’t. That’s the meeting.”
Funding the Maintainer Is the Cheap Half of the Invoice
The respectable answer — the one I gave for years, and the one most of the right has settled on — is that large companies free-ride on volunteer maintainers and should cut checks. Fine. True. And almost beside the point.
Sponsoring a maintainer is cheap and it is upstream. The migration is downstream, and downstream is where the money actually goes: the extension nobody wrote a replacement for, the integration test suite nobody trusts, the intern who has to learn why the invoice generator rounds that way. You can fund every maintainer on the php-src commit log and still leave a thousand IT departments holding a bill they cannot pay, because the constraint was never the patch. The patch is the part that works.
This is where conservatives ought to feel a flicker of recognition and mostly don’t. Maintenance is the most conservative activity in software: it preserves what exists, it distrusts the new, it values the working system over the elegant one, and it is performed by people with no marketing department and no conference slot. If you believe institutions decay when nobody tends them, you should be furious that the entire maintenance economy — patch releases, end-of-life calendars, migration labor — is treated as janitorial work by an industry that gives keynote stages to whatever shipped in the last eighteen months.
Meanwhile the attention economy and the maintenance economy keep diverging. The conference circuit debates inference costs and agent frameworks. The asset base — the actual web that takes payments and files permits and publishes school lunch menus — ships a changelog with no adjectives and gets 912 upvotes from people saying thank you to a stranger.
That is the closest thing to a maintenance-economy press conference you will get this year. It arrived without a press release, on a branch that expires on Dec. 31, and almost everybody who clicked was someone who cannot afford to be wrong about what happens next.
Nobody gets to upvote the migration.