We took back a glob shortcut

A Next.js app keeps its routes in directories named like app/[id]. In
almost every glob tool, [id] is a character class: one character, i
or d. So a glob written for that route matches nothing until you
escape it as app/\[id\].

vx once took a shortcut: a [...] in a task glob was literal. A route
directory needed no escape. It looked friendly, and it was a mistake.

A glob should mean the same thing everywhere

The shortcut broke the other half. src/*.[jt]s is an ordinary way to
say "the .js and .ts files", and in vx it matched nothing. A glob
copied from turbo.json, an Nx project or a shell meant something else
once it reached a vx.config.ts. A config you cannot read the same way
in two tools is a config you have to test twice.

So brackets are a character class again, everywhere a task glob is read:

inputs: { files: ['src/**/*.[jt]s', 'app/\\[id\\]/**'] }

src/**/*.[jt]s matches .js and .ts. A literal route directory is
escaped, as it is in Turbo, Nx and most glob tools (the double backslash
is the TypeScript string's escape). @vzn/vx-migrate passes Turbo's and
Nx's classes through unchanged, and escapes the brackets in any path it
writes.

Why every key changed

A glob is part of what a task's key folds in. After the change, the
same glob string can select different files: app/[id]/** used to mean
the route and now means app/i/** or app/d/**. A key that stayed the
same over different inputs could replay outputs built from the wrong
files. That is the one failure a cache must never have, so 0.0.644
moved the cache version to vx-cache-v48, and every cached task misses
once after the upgrade.

vx is pre-1.0, and a breaking change ships as one: no alias, no setting
to keep the old reading. The glob rules are in
Caching. More on vx at
vznjs.github.io/vx.

---

Originally published on the vx blog. vx is open-source (MIT) task orchestration with a build cache: vznjs.github.io/vx · GitHub.

_{Written with AI assistance.}