Compilation Issue on rust 1.41.0

[dbcfd]

error[E0428]: the name `imp` is defined multiple times
   --> .cargo/registry/src/github.com-1ecc6299db9ec823/memchr-2.3.1/src/lib.rs:148:5
    |
139 |     fn imp(n1: u8, haystack: &[u8]) -> Option<usize> {
    |     ------------------------------------------------ previous definition of the value `imp` here
...
148 |     fn imp(n1: u8, haystack: &[u8]) -> Option<usize> {
    |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ `imp` redefined here
    |
    = note: `imp` must be defined only once in the value namespace of this block

[dbcfd]

System info

info: using existing install for 'stable-x86_64-apple-darwin'
info: default toolchain set to 'stable-x86_64-apple-darwin'

  stable-x86_64-apple-darwin unchanged - rustc 1.41.0 (5e1a79984 2020-01-27)

[dbcfd]

This also happens on beta and nightly.

[BurntSushi]

I'ved yanked 2.3.1 from crates.io until I have time to properly investigate this. cc @ldesgoui

(It was almost certainly caused by #67.)

[BurntSushi]

I'm particular, I would like to know why CI didn't catch this.

[CryZe]

https://github.com/BurntSushi/rust-memchr/pull/67/files#diff-b4aea3e418ccdb71239b96952d9cddb6R145

This should be not(miri)

Though the problem seems to be in more places.

[ldesgoui]

My bad! I'll be home in 10 to fix, not(all(a, b, c)) should be all(not(all(a, b)), not(c))

[BurntSushi]

No rush. It is yanked so there should be anyone impacted by this now. But yeah, thank you. I also need to figure out how to make sure this case is tested in CI.

[kali]

If it can help, I see it from a github action CI here https://github.com/snipsco/tract/runs/441345982?check_suite_focus=true

[ldesgoui]

The compilation error is caused by the logic detecting when the C version should be used, which is memchr_libc AND (NOT (x86_64 AND memchr_runtime_simd)) AND (NOT miri) which translates to
cfg(all( memchr_libc, not(all(target_arch = "x86_64", memchr_runtime_simd)), not(miri) )
Here's the truth table comparing what we have/what we want:

x64	simd	miri	!(x64 && simd && miri)	!(x64 && simd) && !miri
f	f	f	t	t
f	f	t	t	f	wrong
f	t	f	t	t
f	t	t	t	f	wrong
t	f	f	t	t
t	f	t	t	f	wrong
t	t	f	t	f	wrong
t	t	t	f	f

4 wrong cases, the one hit by users was most likely features = "libc", x64 = t, simd = t, miri = f

[ldesgoui]

This case is not tested in there:
https://github.com/BurntSushi/rust-memchr/blob/2.3.1/.github/workflows/ci.yml#L100-L122
Incoming PR with fix and according CI test