"Fancy new algorithm" doesn't work on big-endian machines about bytecount HOT 13 CLOSED

from bytecount.

Ah dangit, thanks so much for reporting this. I'm guessing the issue is

let mask = !(!0 >> ((utf8_chars.len() % chunksize) * 8));

and I have a patch at PR #54. I'm installing some cross platform testing stuff and should have confirmation soon.

from bytecount.

Yep, I can repro the issue and the fix works for me. Thanks again!

@llogiq We should CI this.

from bytecount.

Is there a CI that gives us native big-endian machines? Or do we need to setup qemu on a Travis job?

from bytecount.

https://github.com/japaric/trust looks nice. Looking through it mostly just seems to wrap https://github.com/rust-embedded/cross, which is what I used locally (so maybe just use cross?).

from bytecount.

The fix works for me over here too, thanks for the quick response!

from bytecount.

version 0.5.1 with the fix will be published soon.

from bytecount.

0.5.1 is published to crates.io.

from bytecount.

What architectures should we add to CI?

from bytecount.

I used mips, but any big-endian arch should be fine. https://github.com/BurntSushi/byteorder/blob/64c03fb383289f8e42b2892bf72297879777a238/.travis.yml#L8

from bytecount.

ppc64 and s390x if available are typically much faster than mips.

from bytecount.

Looking at that travis patch, does CROSS_TARGET actually run the tests? I would not have normally expected so... the bad code does compile successfully, but the test fails.

from bytecount.

FWIW, travis recently gained s390x support, so testing there should be easy-ish.

from bytecount.