Mergify

Shipping Python 3.14 and suddenly every FastAPI endpoint throws NameError? ⚠️ PEP 649 defers annotation evaluation — great until your framework tries to resolve TYPE_CHECKING imports at runtime. We break down what PEP 649 actually breaks in production and why this trips up FastAPI. If your API inspects type hints at runtime, read this before you deploy. 👉 https://lnkd.in/ewXydfms

Remember when on-call meant 6 tabs, 4 CLIs, and a Slack archaeology dig? A staff engineer compressed all that into a single Markdown file in git; with Claude Code, any teammate can run a six-system investigation in two minutes. We wrote up the pattern so you can copy it: capture tribal knowledge, wire the steps, and make on-call runnable. 👉 https://lnkd.in/eSUgKUD7

Ever shipped with green CI and still wondered what’s lurking in the interleavings? We modeled our merge queue as a TLA+ state machine and let TLC loose: 468,000 states explored. Result: two real bugs our test suite—and years of production traffic—never caught. 🐛 We walk through the spec, the counterexamples, and how we fixed them in the post. 👉 https://lnkd.in/enbxrZyZ

Why is your frontend PR waiting on backend CI it never touched? ⏳ In a monorepo, one merge queue serializes everything. Scoped parallel queues turn it into a DAG, so unrelated changes can merge in parallel. We explain how to pick scopes and avoid cross-scope hazards—so the queue stops being the blocker. New blog post. 👉 https://lnkd.in/eT2TTxeq

Ever had a pytest fixture refuse to tear down after a rerun? 🧪 While building pytest-mergify, we hit that exact snag—finalizers weren’t behaving during reruns. Our fix was small: two helper functions and a trick borrowed from pytest-rerunfailures to ensure clean teardown between attempts. We wrote it up in “Diving into pytest Finalizers” if you’ve wrestled with reruns and fixtures. 👉 https://lnkd.in/ee2jTFYz

Ever watched a Slack shell script spiral into a shipped app? In 4 days and 31 commits—using a language I’d never touched—I shipped a native macOS app built entirely with Claude Code. The takeaway: tight feedback loops and automation turn throwaway tools into real products fast. I wrote about the dashboard arms race that put me on the App Store and what it changed for my workflow. 🚀 👉 https://lnkd.in/e2QVP3wh

Ever hit enter on an AI agent’s “cleanup” and watch 29 worktrees vanish? 🧹 I rubber-stamped a suggestion and it ran --force across my git worktrees. The takeaway: never delegate --force without guardrails—use dry-runs, scoping, and a quick diff before execute. I wrote up what happened and the simple checks that would’ve saved the day. 👉 https://lnkd.in/eU-CBqDU

Ever watched an AI agent “tidy up” your repo—then nuke 29 worktrees with --force? 💥 I rubber-stamped the suggestion and learned the hard way: automation needs guardrails, especially around destructive commands. We break down what went wrong, how to protect your git worktrees, and the simple checks to keep agents from doing irreversible ops. Full story on the Mergify blog: The Day My AI Agent Deleted 29 Git Worktrees. 👉 https://lnkd.in/eU-CBqDU

An empty array took down our merge queue. 😬 Turns out GitHub webhooks can be perfectly valid yet stale when events arrive out of order. We traced the incident and built action‑aware upserts so newer truths win and state stays correct. We wrote up what happened and how we fixed it—useful if you rely on webhooks in prod. 👉 https://lnkd.in/eQq_iw7P

What if AI didn’t open the door to coding—it just copied the key to a very messy room? In our new post, we dig into trials where AI code ran 19% slower, with 30% more warnings and 322% more vulnerabilities. And the baseline wasn’t pristine either. The work that matters now isn’t typing; it’s judgment, taste, and choosing which room to build. 🔑 👉 https://lnkd.in/e6BS3gsz