20m lines of code is not large enough to have 6 people in build tooling. A lot of companies at Twitter size end up over engineering and resisting buying vs. building in house big shot projects that often times can't compete with some 3rd party solution they could just buy.
For example my company is reinventing Next.js in house because the politics in the company won't allow us ditch this half-baked Next.js clone and just use the real thing. Our deployment pipeline will be much simpler if we simply bought Vercel. I'm sure we can negotiate prices with them that is better than hiring 12 engineers
I think the more relevant number is the number of ICs (around 2000, according to TFA). 12 build engineers (I don't know where you got 6 from) for 2000 ICs doesn't seem large at all to me; it's actually less than I would have expected.
They were not building/re-inventing Bazel (its Open Source from Google); they were just adopting it and ensuring the migration for the rest of the company (which is no small feat, since it affects all other developers in the company).
The weirdest part here is that the author worked on a migration from their NIH build tool to Bazel for 2 years and never once mentions why all of that was even happening, and by the end of his employment, it was still not done.
He links the pants-devel announcement and here is what it says:
> Twitter has made the decision to migrate from Pants to Bazel as our primary build tool, as we believe that we have unique needs that Bazel can better address
Pants was your own fucking build tool that you invented, controlled and maintained! How on earth can it not meet your unique needs.
I think you greatly underestimate how hard it is to find a good build engineer. Two of the smartest and hard working engineers did a bazel migration at one of my old companies and had to contribute heavily upstream to support all of our use cases.
It’s much cheaper in the long run to migrate to a well supported tool that you can extend than have to continuously staff some of the best engineers to keep building on your homegrown mess.
This comment is reminiscent of others i have seen about “building uber with 20 engineers” what could be so hard?
I find it weird that everyone acts like Twitter is a new startup that created all of it's infra 1-2 years ago, while they had 100 millions of users in 2012 - a lot of this stuff was already years in the making then.
I'm curious what you think it's possible to buy that replaces the need to invest on top of Bazel. Have you worked on a comparably sized codebase that used a simpler solution?
For example my company is reinventing Next.js in house because the politics in the company won't allow us ditch this half-baked Next.js clone and just use the real thing. Our deployment pipeline will be much simpler if we simply bought Vercel. I'm sure we can negotiate prices with them that is better than hiring 12 engineers