libutil: Avoid repeated memory reallocation in BaseNix32::decode

[Zaczero]

assert(used <= maxSize);
res.resize(used);

Downsizing is essentially free whereas res.resize(i + 1) causes heavy memory realloc.

Also fixed a small bug where encodedLength(0) would return a really big number (no real imapct).

Plus made some minor type adjustments where made sense to me (size_t is more idiomatic for len/size).

Motivation

Context

---

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

[edolstra]

Hm, where did the repeated allocation occur? Didn't the res.reserve(...) in decode() prevent that?

[Zaczero]

Agreed, I misunderstood the workings of resize (I'm new to C++). So I suppose the primary difference now is the performance of 1 bulk memset vs individual writes.

[xokdvium]

Where does the memset happen here?

[Zaczero]

std::string res(maxSize, '\0');

whereas resize writes null byte on each call to make it proper cstr.