Go 1.26 Interactive Tour

With the new GC I really love the trend of understanding memory bandwidth is the bottleneck for many things and the combination of locality and SIMD is a big performance unlock.

Reminds me of the WAND vs MAXSCORE discussion by turbopuffer: https://turbopuffer.com/blog/fts-v2-maxscore

> Optimized fmt.Errorf — With the new Go release, this debate should finally be settled. For unformatted strings, fmt.Errorf now allocates less and generally matches the allocations for errors.New. Specifically, fmt.Errorf goes from 2 allocations to 0 allocations for a non-escaping error, and from 2 allocations to 1 allocation for an escaping error […]

Thank you so much!!! I used to always take an extra second to decide which one to use based on the function’s parameter(s), and now I feel like I’m getting that time back, no sarcasm at all. This is a genuinely great improvement.

I'm salty they are embracing SIMD, but shove this proposal to the side:

https://github.com/golang/go/issues/58610

if you know the ISPC compiler, the proposal is basically that for Go

Oh wow the changes to new() are an enormous ergonomics unlock. ErrAsType will also replace a similar package for me (too bad value errors will still be a thing though, and beginning new type errors still so wordy).

Anton's interactive tour is the best way to read the release notes IMO! As mentioned on Cup o' Go.

This was a great read, even without having been working with Go for some time. It was well written, the examples demonstrated nearly everything, and being able to edit samples to test everything else, all on my phone, is quite delightful. Cheers to the author

I don't use Go often but that's an excellent summary article.

  https://antonz.org/go-1-26/#httptest-example

> To fix this issue, the HTTP client returned by httptest.Server.Client now redirects requests for example.com and its subdomains to the test server

I'm really conflicted about the idea that the https client should silently hijack requests to example.com in order to make a dummy certificate work... or am I just really misunderstanding?

The recursive type constraints are excellent.

Fed *bool `json:"is_fed"` // you can never be sure

felt that part

The differences btwn new() and make() is even lesser now. Is the goal is to deprecate make()?

Wow, big release. I wonder if it was AI facilitated?