std.net: Replace ArrayLists with fixed size arrays

[usebeforefree]

I have resolved an old TODO, replacing some ArrayLists in std.net with fixed size arrays. The changes have been tested with std/net/test.zig. No changes made to the test itself, since the behaviour stays the same.

[linusg]

Having to deal with a null state for each ns item seems a bit awkward. The BoundedArray replacement of maintaining a separate length field from https://github.com/ziglang/zig/pull/24699/files#diff-e6dad51212e62a28f43ec6ec4c94e539c08bc1e481db4de2bd49b03d85ea53e2 might be a bit nicer:

- analysis_roots: std.BoundedArray(*Package.Module, 4) = .{},
+ analysis_roots_buffer: [4]*Package.Module,
+ analysis_roots_len: usize = 0,

[AndrewKraevskii]

@linusg suggestion is good with u2 instead of usize.

[usebeforefree]

Not sure if I should include a ns_addr_len as well, or just keep using rc.ns_len.

[jeroendecuyper]

Doesn't this PR throw an error if there are more then 3 nameservers defined vs glibc where it just ignores from the fourth one on?
Due to this line 1829.

[usebeforefree]

Doesn't this PR throw an error if there are more then 3 nameservers defined vs glibc where it just ignores from the fourth one on? Due to this line 1829.

You are completely right, I'm just not sure how to handle it. Returning an error makes the CI fail, but silently not returning anything is deceiving, and certainly not good.

[jeroendecuyper]

To me, this either goes in either of 2 ways =>

• similar to glibc, ignore the fourth and so on. just from compability reasons.
• otherwise write a modern resolver and allow all nameservers to be used that are specified.

[usebeforefree]

Thank you for your help and patience, the CI should pass now. :)

[andrewrk]

I'd like to be in charge of merging this please because I have some related changes in a branch to be merged soon.