Bun 1.3.13 Stops Buffering Installs, and That Saves CI Jobs

Bun 1.3.13 changes how bun install handles a package tarball while it downloads. Previously the full archive buffered in memory before extraction started. Now Bun extracts it as the bytes arrive, so the peak memory an install needs drops significantly. Source maps use far less memory too.

On paper this reads as a minor plumbing fix, the kind of changelog line you skim past on the way to the benchmark numbers everyone actually shares. In practice it is the difference between a job finishing and a job dying.

CI runners and small containers are usually capped at 512MB or 1GB of memory, and a big install with a lot of dependencies used to be exactly the moment that limit got tested. Buffering a large tarball fully in memory before writing anything to disk is precisely the kind of spike that gets a process killed by the out of memory reaper, and the failure often looks unrelated to the actual cause, so people waste time chasing the wrong bug.

We would rather have this fix than another 20 percent faster install time. Speed is nice when everything already works. Not having your pipeline die halfway through a dependency install on a cheap runner is the difference between shipping and re running the job and hoping it passes this time. Bun quietly picked the more useful problem to solve.