replacing few memset usages with safer alternatives

[mikea]

I focused on most trivial usages first

[kentonv]

Does this actually zero-initialize given the unusual way class word is defined?

[kentonv]

Even if it does... I wonder if this ends up violating aliasing rules. Word arrays are never actually accessed using the type word, but rather by reinterpret-casting to some other pointer type and reading that. Technically it is UB to write type word and then read as some other non-byte-sized type, or vice versa. There's an exception for byte-oriented access, which is allowed to alias everything; memset() operates on bytes, and so doesn't break aliasing rules.

Unsure to what extent any of this matters in practice.

[mikea]

Does this actually zero-initialize given the unusual way class word is defined?

yes, I was curious about this as well: https://godbolt.org/z/Wcsn99WKc

Technically it is UB to write type word and then read as some other non-byte-sized type, or vice versa.

Not sure if this matters, but we don't write to word here, but default-initialize it?
Is there a special compiler mode that could check this for us?

Unsure to what extent any of this matters in practice.

I could use kj::memzero() for word arrays instead.

[mikea]

Compilers claim to have some aliasing analysis but I wasn't able to trigger it, did I break the rules here: https://godbolt.org/z/WrrG9jres ?

[kentonv]

That does appear to break the rules, not sure why it isn't detected. My understanding is that the compiler could legally re-order the assignment to buf2->content vs. f->d.