globwalker: avoid spurious allocations in iterator

[lucab]

This tweaks the GlobWalker iterator logic in order to avoid a spurious allocation for each processed directory entry (including ignored ones).

Verified by running the current testsuite under heaptrack and observing allocations stats:

$ heaptrack target/debug/deps/globwalk-baseline
[...]
heaptrack stats:
        allocations:            7176
[...]

$ heaptrack target/debug/deps/globwalk-patched
[...]
heaptrack stats:
        allocations:            7081
[...]

[Gilnaa]

Neat! Thanks

[lucab]

Thanks for the quick merge!
For context, I came to this while looking into the allocation profile of https://github.com/getzola/zola (through its https://github.com/Keats/tera dependency).
It would be great if you could tag a bugfix 0.9.1 release with this, whenever you have time, so that I can pick up the improvement there.
Thanks again for maintaining this crate 🎉

[Gilnaa]

No problem, published now.
Do note that tera is probably going to stay on globwalk 0.8 for now;
I admit I'm not well-versed with how the resolver works, but you might need to force its hand.

Purely out of curiosity, can I ask you to spare a bit more details on what you tried to solve?
How much of an effect did those allocations have?

[lucab]

Do note that tera is probably going to stay on globwalk 0.8 for now

Thanks for the heads-up, I guess I'm going to wait for the next major version then.

Purely out of curiosity, can I ask you to spare a bit more details on what you tried to solve?
How much of an effect did those allocations have?

I just wanted to have a look at zola build CPU/memory usage patterns and noticed this low-hanging fruit.
Impact is pretty limited but it linearly grows with the number of files. On the CI testsite (small-medium size, ~150 files) it accounts for ~1% of temporary allocations (allocations directly followed by their deallocation).