[stdsimd] cargo fmt --all succeeds but subsequent cargo fmt --all -- --check fails

[gnzlbg]

With this PR cargo fmt --all currently outputs nothing and returns 0 (EXIT_SUCCESS). A subsequent call to cargo fmt --all -- --check returns 101 and sometimes (1/5 times) prints the following diff. Other times it doesn't print anything, just fails with that error (e.g. like here: https://travis-ci.org/rust-lang-nursery/stdsimd/jobs/392688468#L472):

Diff in stdsimd/crates/stdsimd-verify/src/lib.rs at line 95:
 
 fn to_type(t: &syn::Type) -> proc_macro2::TokenStream {
     match *t {
-        syn::Type::Path(ref p) => match extract_path_ident(&p.path).to_string().as_ref() {
-            "__m128" => quote! { &M128 },
-            "__m128d" => quote! { &M128D },
-            "__m128i" => quote! { &M128I },
-            "__m256" => quote! { &M256 },
-            "__m256d" => quote! { &M256D },
-            "__m256i" => quote! { &M256I },
-            "__m64" => quote! { &M64 },
-            "bool" => quote! { &BOOL },
-            "f32" => quote! { &F32 },
-            "f64" => quote! { &F64 },
-            "i16" => quote! { &I16 },
-            "i32" => quote! { &I32 },
-            "i64" => quote! { &I64 },
-            "i8" => quote! { &I8 },
-            "u16" => quote! { &U16 },
-            "u32" => quote! { &U32 },
-            "u64" => quote! { &U64 },
-            "u8" => quote! { &U8 },
-            "CpuidResult" => quote! { &CPUID },
-            s => panic!("unspported type: {}", s),
-        },
+        syn::Type::Path(ref p) => {
+            match extract_path_ident(&p.path).to_string().as_ref() {
+                "__m128" => quote! { &M128 },
+                "__m128d" => quote! { &M128D },
+                "__m128i" => quote! { &M128I },
+                "__m256" => quote! { &M256 },
+                "__m256d" => quote! { &M256D },
+                "__m256i" => quote! { &M256I },
+                "__m64" => quote! { &M64 },
+                "bool" => quote! { &BOOL },
+                "f32" => quote! { &F32 },
+                "f64" => quote! { &F64 },
+                "i16" => quote! { &I16 },
+                "i32" => quote! { &I32 },
+                "i64" => quote! { &I64 },
+                "i8" => quote! { &I8 },
+                "u16" => quote! { &U16 },
+                "u32" => quote! { &U32 },
+                "u64" => quote! { &U64 },
+                "u8" => quote! { &U8 },
+                "CpuidResult" => quote! { &CPUID },
+                s => panic!("unspported type: {}", s),
+            }
+        }
         syn::Type::Ptr(syn::TypePtr { ref elem, .. })
         | syn::Type::Reference(syn::TypeReference { ref elem, .. }) => {
             let tokens = to_type(&elem);

We should add to CI a check that after formatting cargo fmt --all -- --check should return true and if not fail the integration tests.

[gnzlbg]

cc @nrc

[gnzlbg]

update: the integration tests for cargo fmt --all -- --check have been merged in 75d46b1

Once this is fixed those tests should prevent it from becoming broken again.

[topecongiro]

This is still reproducible with the latest rustfmt (rustfmt 0.99.1-nightly (da17b689 2018-08-04)).

Why do cargo fmt --all or cargo fmt --all -- --check return 101 in the first place? AFAICT in the latest rustfmt there is no code path that returns 101 exit code...

[topecongiro]

I figured that rustfmt silently crashes when formatting stdsimd/crates/src/lib.rs. In particular, this module declaration seems to break rustfmt.

[topecongiro]

EDIT: Ignore this comment, this is wrong 😞

---

Ok, so the actual file that breaks rustfmt is stdsimd/stdsimd/mod.rs, and the relevant code is this:

/// ```ignore
/// #[cfg(all(any(target_arch = "x86", target_arch = "x86_64"),
///       target_feature = "avx2"))]
/// fn foo() {
///     #[cfg(target_arch = "x86")]
///     use std::arch::x86::_mm256_add_epi64;
///     #[cfg(target_arch = "x86_64")]
///     use std::arch::x86_64::_mm256_add_epi64;
///
///     unsafe {
///         _mm256_add_epi64(...);
///     }
/// }
/// ```

Note that the third line should be prefixed with #. I have sent a PR that fixes this to stdsimd (rust-lang/stdarch#548).

[topecongiro]

I am still unsure why rustfmt started to fail silently against invalid code, I guess there is something weird going on inside syntax::with_globals?

[topecongiro]

#2787 (comment) was wrong 😞 It was a pure bug in the rustfmt side. #2895 should fix this.

[topecongiro]

Closed via #2895.

[gnzlbg]

Did this land on the nightly preview yet? Or could you ping me when it does?

[nrc]

It hasn't because we've had some problems in the Rust repo. It is in the queue, so as soon as there is a new nightly, it should have this fix in it.