Creating a vector in C++ has never been easier

[vitaut@mastodon.social]

Creating a vector in C++ has never been easier

[funkylab@mastodon.social]

@vitaut oh that really captured my attention.

[vitaut@mastodon.social]

@funkylab captivating, innit?

[jsbarretto@social.coop]

@vitaut Is there some magic here I'm too Rusty to understand? How is `v` in scope?

[vitaut@mastodon.social]

@jsbarretto it is captured by &

[jsbarretto@social.coop]

@vitaut ...before it has been defined? So it gets default-constructed, captured by the closure, then moved back into the original place? Gosh, that's... not nice

[funkylab@mastodon.social]

@vitaut yes, it is.

I thought I was pretty on top of what happens here. But when I change the line to
`std::vector<int> v = [&]() -> std::vector<int> { v.push_back(1); return {}; }();`
then it stops working, making me think I have the value category of your original lambda's return value wrong. Hm.

[vitaut@mastodon.social]

@jsbarretto The name becomes visible right after declaration. It's actually worse than you think because it is captured uninitialized.

[lisyarus@mastodon.gamedev.place]

@vitaut But...why

[vitaut@mastodon.social]

@lisyarus ¯\_(ツ)_/¯

[funkylab@mastodon.social]

@vitaut oh no. The program's output depends on whether I create `v1` after changing `v`, doesn't it… oh noooo

[vitaut@mastodon.social]

@funkylab

[jsbarretto@social.coop]

@vitaut Oh lawd, what a mess

[janriemer@floss.social]

The picture shows a C++ code snippet with the following content...

@altbot

[smurthys@hachyderm.io]

@vitaut "reference capture considered harmful"

[oschonrock@mastodon.social]

@jsbarretto @vitaut

Ewwwww...

This better be UB!

[pervognsen@mastodon.social]

@vitaut @jsbarretto The interaction with captures is cursed but you can do stuff like

Node n = {.value = 42, .next = &n}

in C to statically initialize self-referential pointer-based data structures. The more interesting case is something like

Node n[] = {{.value = 1, .next = &n[1]}, {.value = 2, .next = &n[0]}}

where you effectively get the linker/loader's relocation engine to deserialize cyclic pointer-based graphs. For statics you can also just extern declare the variables.

[pervognsen@mastodon.social]

@vitaut @jsbarretto Although this makes me realize I've never actually checked the spec wrt the semantics for stack/automatic storage duration variables for the self-referential initialization pattern (it's hard to see a use case) but I guess it's still alright?