Dice Lock's large-text board now turns on only when a player asks for itDice Lock's bigger-text mode now turns on only when a player asks for it
October 28, 2026 · Dice Lock's large-text board drops its OS auto-detect for a one-time pinch gesture, and leaked headless Chrome processes get killed for good in two repos.Dice Lock's bigger-text mode drops its automatic phone-setting check for a one-time on-screen offer, and a leftover background browser gets reliably shut down in two different tools.
last updated
Large Board answers to the player, not the phone
Dice Lock has had a large-text mode for its board, Large Board, sitting behind a flag called JIO-2392 for a while. The original plan let it turn itself on: at boot, the page would read the phone's own accessibility setting, iOS's content-size category or Android's font scale, and if that setting already asked for large text, Large Board came up without the player touching anything. Today that plan was pulled, before it ever reached a player. The boot-time OS probe is gone, and so is the three-state key it wrote (on, off, or never-chosen), replaced by a plain yes-or-no: Large Board is on only if a player turned it on. A phone already carrying an old build's stored "on" still loads as on; anything else loads as off.
What stayed is a gesture, not a default: spreading two fingers past 1.2x mid-match, on an upright phone, with Large Board off, pauses the board under one offer: BIGGER BOARD?, with NO THANKS and LARGE BOARD, and a line explaining that the mode swaps the Raise and Roll buttons for a swipe and a double-tap. Only choosing LARGE BOARD writes anything; declining, pressing Escape, or pinching back in leaves the setting untouched, and the offer itself only fires once.
Turning it on exposed two layout bugs that needed their own fixes the same day. On a Tower step at the narrowest supported width, 320 pixels, the step's floor pill crowded a name, a die icon, and a time into one row and clipped the lettering; it's now a two-column grid, with the name on its own line above the die and time. Separately, under 375 pixels wide, the Pause button sat directly over the board's top-left corner, because the game hadn't reserved it any room; the layout code now holds that space open on narrow Large Board screens. A third fix gave the RESTART? confirmation sheet a description a screen reader actually announces, rather than just the dialog's name.
A leaked browser was the other half of a load spike
Two unrelated pipelines, Dice Lock's own docs renderer in the bigbrain-core repo and a browser test in this site's repo, both drive headless Chrome, and both had the same leak: Chrome was killed with SIGTERM on a timeout, the polite "please exit" signal, and a Chrome instance wedged in its own shutdown just ignores that. If the parent process was gone by the time Chrome finally gave up, the browser kept running under launchd, invisible to whatever started it. The two were contributing to a load average of 30 to 90 on a 10-core Mac.
Both got the same fix the same day. The docs renderer and its media-capture script now send SIGKILL, which a process cannot ignore, and take an explicit timeout instead of relying on the library default. The site's own browser test goes further: it now tracks the browser's PID directly and kills it on a failed connection, not only on a clean close, and on SIGINT, SIGTERM, or SIGHUP, so a runner timeout or a cancelled run can't leave Chrome behind either.
A bigger-text mode for Dice Lock now turns on only when a player asks
Dice Lock, the dice puzzle game, has a mode that shows everything on the board in larger text, called Large Board. The original plan was for the game to turn this on by itself: when it started up, it would check the phone's own built-in large-text setting, and if that was already turned on, Large Board would come on too, with no action from the player. That plan was scrapped today, before it ever reached a real player. The game no longer checks the phone's setting at all, and a three-way memory of "on", "off", or "never asked" was replaced with a simple on-or-off switch: Large Board is on only if a player actually turned it on themselves.
What stayed is a gesture, not an automatic switch: if a player spreads two fingers apart on the screen during a match, enough to count as a deliberate stretch, and Large Board is currently off, the game pauses and offers it once, with a button for "no thanks" and a button to turn it on, plus a note that turning it on swaps the usual buttons for a swipe and a double-tap. Saying no, or pinching back the other way, changes nothing, and the offer only ever shows up once.
Turning this feature on uncovered two display bugs that got fixed the same day. On the narrowest phone screen size the game supports, a label showing a level's name, a small die icon, and a time all got squeezed into one line and the text got cut off at the top; it now shows the name on its own line above the die and time. Separately, on phones a little wider than that but still narrow, the pause button was sitting directly on top of the corner of the game board, because the game hadn't left it any room; that's fixed too. A third small fix means a screen reader now reads out what a "restart this run" confirmation screen actually means, instead of just announcing its title with no explanation.
A browser left running in the background was adding to a slowdown
Two separate systems use an invisible, script-controlled copy of the Chrome browser to automatically generate things: one renders this project's internal documentation diagrams, and the other is a test that checks pages in a browser for this website. Both had the same problem. When either one tried to close Chrome after a set amount of time, it sent the polite "please close now" signal, but a Chrome process that gets stuck while trying to shut itself down just ignores that signal and keeps running. If the program that started it had already quit by the time Chrome finally gave up, that Chrome process kept running in the background completely unnoticed. Together, these leftover Chrome processes were part of why the Mac mini running all of this was seeing a workload level of 30 to 90, on a machine with only 10 processor cores to spread that across.
Both got fixed the same way on the same day. The documentation tool now sends the forceful "stop immediately" signal instead of the polite one, which a process cannot ignore, and gives itself a clear time limit instead of trusting a default one. The website's own test goes a step further: it now keeps direct track of exactly which browser process it started and shuts that one down no matter how the test ends, whether it fails, times out, or gets cancelled, instead of only cleaning up when everything goes according to plan.