1.48 beta broken on mips due to copy_file_range EOVERFLOW

[infinity0]

https://buildd.debian.org/status/fetch.php?pkg=rustc&arch=mipsel&ver=1.48.0%7Ebeta.8%2Bdfsg1-1%7Eexp1&stamp=1605149146&raw=0

error[E0107]: wrong number of type arguments: expected 3, found 2
   --> /usr/src/rustc-1.48.0/vendor/petgraph/src/graphmap.rs:580:16
    |
580 |     edges: &'a IndexMap<(N, N), E>,
    |                ^^^^^^^^^^^^^^^^^^^ expected 3 type arguments

error[E0107]: wrong number of type arguments: expected 2, found 1
   --> /usr/src/rustc-1.48.0/vendor/petgraph/src/matrix_graph.rs:925:22
    |
925 |     removed_ids: &'a IndexSet<usize>,
    |                      ^^^^^^^^^^^^^^^ expected 2 type arguments

error[E0107]: wrong number of type arguments: expected 3, found 2
  --> /usr/src/rustc-1.48.0/vendor/petgraph/src/graphmap.rs:61:12
   |
61 |     nodes: IndexMap<N, Vec<(N, CompactDirection)>>,
   |            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ expected 3 type arguments

error[E0107]: wrong number of type arguments: expected 3, found 2
  --> /usr/src/rustc-1.48.0/vendor/petgraph/src/graphmap.rs:62:12
   |
62 |     edges: IndexMap<(N, N), E>,
   |            ^^^^^^^^^^^^^^^^^^^ expected 3 type arguments

error[E0107]: wrong number of type arguments: expected 2, found 1
   --> /usr/src/rustc-1.48.0/vendor/petgraph/src/matrix_graph.rs:851:18
    |
851 |     removed_ids: IndexSet<usize>,
    |                  ^^^^^^^^^^^^^^^ expected 2 type arguments

error: aborting due to 5 previous errors; 87 warnings emitted

For more information about this error, try `rustc --explain E0107`.

Did not run successfully: exit code: 1
"/<<PKGBUILDDIR>>/build/mipsel-unknown-linux-gnu/stage1/bin/rustc" "--crate-name" "petgraph" "--edition=2018" "/<<PKGBUILDDIR>>/vendor/petgraph/src/lib.rs" "--error-format=json" "--json=diagnostic-rendered-ansi,artifacts" "--crate-type" "lib" "--emit=dep-info,metadata,link" "-C" "opt-level=3" "--cfg" "feature=\"default\"" "--cfg" "feature=\"graphmap\"" "--cfg" "feature=\"matrix_graph\"" "--cfg" "feature=\"stable_graph\"" "-C" "metadata=08c9d7a6b2939d49" "-C" "extra-filename=-08c9d7a6b2939d49" "--out-dir" "/<<PKGBUILDDIR>>/build/mipsel-unknown-linux-gnu/stage1-rustc/mipsel-unknown-linux-gnu/release/deps" "--target" "mipsel-unknown-linux-gnu" "-C" "linker=mipsel-linux-gnu-gcc" "-L" "dependency=/<<PKGBUILDDIR>>/build/mipsel-unknown-linux-gnu/stage1-rustc/mipsel-unknown-linux-gnu/release/deps" "-L" "dependency=/<<PKGBUILDDIR>>/build/mipsel-unknown-linux-gnu/stage1-rustc/release/deps" "--extern" "fixedbitset=/<<PKGBUILDDIR>>/build/mipsel-unknown-linux-gnu/stage1-rustc/mipsel-unknown-linux-gnu/release/deps/libfixedbitset-5dcd55a1fc6307c9.rmeta" "--extern" "indexmap=/<<PKGBUILDDIR>>/build/mipsel-unknown-linux-gnu/stage1-rustc/mipsel-unknown-linux-gnu/release/deps/libindexmap-1f8402edc452918c.rmeta" "--cap-lints" "warn" "--cap-lints=warn" "-Clink-args=-Wl,-z,relro" "-Zmacro-backtrace" "-Zunstable-options" "-Wrustc::internal" "-Cprefer-dynamic" "-Zbinary-dep-depinfo" "-Wrust_2018_idioms" "-Wunused_lifetimes" "-Dwarnings" "--sysroot" "/<<PKGBUILDDIR>>/build/mipsel-unknown-linux-gnu/stage1" "--remap-path-prefix" "/<<PKGBUILDDIR>>=/usr/src/rustc-1.48.0" "-Z" "force-unstable-if-unmarked"

[infinity0]

Actually the real culprit would appear to be indexmap which was upgraded from 1.5.1 to 1.6.0 for rust 1.48 - breaking the API that petgraph depends on. @bluss @cuviper

[Mark-Simulacrum]

So the problem is that indexmap's build script uses autocfg to try to discover the availability of some Rust features, and it is likely breaking on mips for some reason. See https://zulip-archive.rust-lang.org/182449tcompilerhelp/46175petgraphwrongnumberoftypearguments.html for some previous discussion.

I suspect the next step here is going to be identifying the cause of the problem -- you should be able to get logs from the autocfg build script somewhere in the build directory, along the lines of build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/indexmap-617a24810b08e703/stderr but with the target triples and the hash in indexmap likely changed.

[cuviper]

Actually the real culprit would appear to be indexmap which was upgraded from 1.5.1 to 1.6.0 for rust 1.48 - breaking the API that petgraph depends on.

There's no API change: indexmap-rs/indexmap@1.5.1...1.6.0

The error you're seeing is because no_std builds have IndexMap<K, V, S>, while std builds have default S = RandomState and add new and with_capacity using that default. We use autocfg to detect whether the target has std, and have done so since 1.3.0. There's a slight functional change in 1.5.2's new "std" feature, but that should only improve the situation by avoiding autocfg when the feature is set, and I don't think many dependents are setting that feature yet.

See indexmap-rs/indexmap#144 and indexmap-rs/indexmap#151 for more context. The autocfg probe just invokes rustc with the output set into the $OUT_DIR, which may be unusual but not wrong IMO. It seems to be an issue with the compiler writing or copying files to some filesystems, so I'd be interested to know what the error is in your case, and whether there are any other build environment differences that cause it to newly fail.

[infinity0]

Thanks for the context, digging through the log I see:

[indexmap 1.6.0] error: could not copy "/<<PKGBUILDDIR>>/build/mipsel-unknown-linux-gnu/stage1-rustc/mipsel-unknown-linux-gnu/release/build/indexmap-f26e0fc96efda7b8/out/probe0.probe0.3a1fbbbh-cgu.0.rcgu.ll" to "/<<PKGBUILDDIR>>/build/mipsel-unknown-linux-gnu/stage1-rustc/mipsel-unknown-linux-gnu/release/build/indexmap-f26e0fc96efda7b8/out/probe0.ll": Value too large for defined data type (os error 79)
[indexmap 1.6.0] 
[indexmap 1.6.0] error: aborting due to previous error
[indexmap 1.6.0] 
[indexmap 1.6.0] 
[indexmap 1.6.0] Did not run successfully: exit code: 1

The first occurence of this error "could not copy A to B: Value too large for defined data type (os error 79)" is for indexmap but it also happens for several other crates, so the problem is likely something else, I'll keep investigating..

[cuviper]

If they're all in probeN.ll files, they're still likely via autocfg, but I think it will ultimately require a compiler fix.

[cuviper]

EOVERFLOW is also a weird error here -- these are definitely not large files or anything...

[infinity0]

It's the stage1 build of indexmap that is failing, so annoyingly I can't reproduce this outside of the rustc-build context with just indexmap since we only have stage0 (1.47) on hand. So it could indeed be some issue with 1.48.

[infinity0]

@the8472 You were the only one to make changes to library/std/src/sys/unix/fs.rs specifically copy() for 1.48, any ideas? It is giving EOVERFLOW on mips.

[infinity0]

I will try reverting 4ddedd5 to see if this fixes the problem, will report back in a few hours.

[cuviper]

I wonder if mips (or your filesystem) doesn't like the request for usize::MAX bytes here:

rust/library/std/src/sys/unix/fs.rs

Line 1233 in 9722952

let bytes_to_copy = cmp::min(max_len - written, usize::MAX as u64) as usize;