TEtelotortium·Oct 8, 2021
The last two sections are gold:
> ### Wait, you did what?
> Time for the philosophizing I promised.
> People are often surprised and sometimes horrified when they learn that Tailscale maintains its own fork of the Go toolchain. Tailscale is a small startup. Isn’t that a horrible distraction, a flagrant burning of [innovation tokens](https://mcfunley.com/choose-boring-technology)?
> Maybe. But the thing is, you write code with the engineers you have.
> We had a problem: We kept crashing on iOS, and in addition to being awful, it was preventing us from adding features.
> Another team might have decided to cut even more features on iOS to try to achieve stability, or limited in some way the size of the tailnet that iOS could interact with.
> Another team might have radically redesigned the data structures to squeeze every last drop out of th
Read the rest
em.
> Another team might have rewritten the entire thing in Rust or C.
> Another team might have decided to accept the crashes and attempted to mitigate the pain by making re-establishment of connections faster.
> Another team might have decided to just live with it and put their focus elsewhere.
> The Tailscale team has Go expertise, spanning the standard library to the toolchain to the runtime to the ecosystem. [It’s an asset](https://danluu.com/in-house/), and it would be foolish not to use it when the occasion arises. And the fun thing about working on low level, performance-sensitive code is that that occasion arises with surprising frequency.
> Blog posts about how people solve their problems are fun and interesting, but they must always be taken with a healthy dose of context. There may be no other startups in existence for which working on the Go linker would be a sensible choice, but it was for us.
> ### Surprise twist ending
> With the Go linker work [newly completed](https://github.com/tailscale/go/pull/20), we set out to confirm our analysis. What would footprint say?
> As expected, footprint approved:
708 KB 1664 KB 0 B 218 __DATA_CONST
> Down from 1568KB to 708KB. Not bad. And did footprint still match iOS’s opinion?
> We expected to see our free memory reported go up by about 1MB. And instead it went up…36MB.
> Huh?
> While we were busy fixing the linker to save 1MB, iOS 15 launched and quietly gave us 35MB more.
> Thanks, iOS!