Cheers for clarifying :)

I'll admit that surprises me greatly, I can't see why it's considered more efficient, but hey, Google.

The first program you build in Noogler training takes more compute and io to build than the Linux kernel. The distributed build systems laughs at such a trivial program and barely breaks a sweat at programs 10x that size.

A poor little laptop would break down and cry.

Google has a giant monorepo. It is too big for git. (Virtually) everything is built from source. Building a binary that just runs InitGoogle() is going to crush a laptop.

I believe that there are also a bunch of IP reasons for this policy, but from a practical perspective doing everything with citc and blaze is really the only option.

Can you share the order of magnitude we're talking about here?

probably efficient because their codebase size would outstretch local disk space requirements, and also would allow for comprehensive access controls