Three bugs in FreeBSD's ig4 driver, one dead touchpad
A touchpad that died on every resume turned out not to be a suspend bug at all. FreeBSD never started the I2C controller the touchpad is wired to, for three separate reasons.
From our projects
Release notes and engineering write-ups published across the Tradik projects. Every headline opens the original article on the site that wrote it.
A touchpad that died on every resume turned out not to be a suspend bug at all. FreeBSD never started the I2C controller the touchpad is wired to, for three separate reasons.
1.8.62 fixes fourteen things that never errored: pages dropped without a word, a 404 in the wrong language, a header that refused geolocation.
1.8.60 is about the work an agency does between builds: letting a client fix their own headline, and a GTM install that was only ever half there.
Click-to-edit is easy to build and hard to make safe. Here is the rule that decides which paragraph a click means, and why the editor sometimes refuses to guess.
We built an incremental build for 1.8.60. It made a five-thousand-page site five per cent faster. Here is why we shipped it anyway, and what it is actually for.
An agency has a different relationship with sitemap.xml than a site owner. Split sitemaps answer 'is it indexed' per section, without forty dashboards.
Almost every fix in four releases came from one place: a third-party crawler, weeks after a build that reported success. What those failures had in common.
A site audit finds copied pages by comparing words. The pages that actually compete with each other share an intent and almost no vocabulary, and word comparison is blind to them.
MDDB 2.14 speaks the stateless MCP revision without dropping the old one, and fixes three ways retrieval was quietly returning the wrong document first.
A ten-document corpus, one ordinary question, and a semantic search that ranks the right answer nowhere. What actually fixed it was not a parameter.
A generator that reads the clock cannot promise the same bytes twice. Which is why the copyright year in your footer is a harder problem than it looks.
Two optimisations to MDDB's vector loop predicted 33% and delivered 1.5%. The benchmark had been saying why for months, in a column nobody read as a diagnosis.
Posts written on tradik.com, and the archive of the blog that ran on pushpanel.io between 2018 and 2020 — infrastructure notes from the work of the time, kept because the problems did not expire with the domain.
The tools we write about are the same ones we bring to client work. Tell us what you are building.