The last version cannot bench on 1.20 anymore

[llogiq]

Note that it will still build alright on that version, but I had to switch to a criterion version which appears not to support compiling without quote etc. which fails on 1.20.

cc @bheisler

[bheisler]

Hey, thanks for letting me know.

Criterion.rs officially supports building on the last three stable releases of Rust, so that would be 1.25 or later. I've tested it, and (for now, at least) Criterion.rs 0.2.4 can build on 1.20 if you disable the html_reports feature. In fact, I added that feature specifically to allow bytecount to benchmark on 1.20.

I'll try not to break that without good reason, but I can't keep supporting old versions of Rust forever. 1.20 is nearly a year old at this point. Since Criterion.rs is a development-only dependency, it doesn't affect the minimum version required by bytecount users. I would recommend benchmarking with newer versions of Rust.

[Veedrac]

bytecount is the kind of thing that should really be "done" and version-locked at some point (around about now, with SIMD), but I'm fine with benching requiring a recent stable.

[llogiq]

@bheisler thank you, I'll have to revisit our CI config – I thought I had disabled HTML generation, but somehow it seems to pull it in regardless.

[llogiq]

I inadvertently restored default-features in the criterion dev-dependency. This is fixed now.

[Veedrac]

@llogiq There's something happening here again, https://travis-ci.org/llogiq/bytecount/builds/459144098. Not sure what's up, but I haven't investigated yet.

[llogiq]

I suspect it now uses a newer version of lazy_static, which appears to use a const fn, which wasn't stable then.

[kornelski]

Could this crate drop support for Rust 1.20 now? The requirement of old lazy_static is problematic for projects using criterion.