Re: Bug#881845: rustc: FTBFS on mips*: test failures
On 02/02/18 13:14, James Cowgill wrote: > Hi, > > On 11/01/18 20:16, Emilio Pozuelo Monfort wrote: >> On 17/11/17 13:25, James Cowgill wrote: >>> Hi, >>> >>> On 15/11/17 17:30, Emilio Pozuelo Monfort wrote: Source: rustc Version: 1.21.0+dfsg1-3 Severity: important Hi, Sometime ago I asked about rustc bootstrap status on mips*: 17:08 < infinity0> mips* fail many tests last time i tried, ~3 months ago, i didn't want to ship it, i haven't had time to check since >From >https://buildd.debian.org/status/fetch.php?pkg=rustc=mips=1.14.0%2Bdfsg1-3=1484077706=0 test net::tcp::tests::timeouts ... FAILED test net::udp::tests::timeouts ... FAILED test sys::imp::ext::net::test::timeouts ... FAILED Looks like timeouts are broken on rust/mips? mips64el has different errors: test f32::f32::tanh_0 ... FAILED test f64::f64::tanh_0 ... FAILED test io::error::Error::from_raw_os_error_0 ... FAILED Maybe the mips porters can take a look? (debian-mips@ on Cc). Note those errors are for 1.14.0, you'll want to try with a newer version first and see what's the current status. >>> >>> I just tried building the latest rustc on mips64el. As I expected, I hit >>> this LLVM bug again where any code using atomics will hang: >>> >>> https://bugs.llvm.org/show_bug.cgi?id=32020 >>> >>> I'll see if I can get Simon to have a look at it again. >> >> FTR: >> >> According to our conversation over IRC, there's a patch for the above bug, >> but >> mips64el is now affected by https://github.com/rust-lang/rust/issues/47290 >> >> mips(el) may be fine now with that patch, or they may not :P > > I have been looking at the state of rust support on mips over the last > weeks or so. These are the major issues: > > Hanging atomics (mips, mipsel, mips64el) > > This is an LLVM bug which happens on all mips. > > https://github.com/rust-lang/rust/issues/39013 > https://bugs.llvm.org/show_bug.cgi?id=32020 > > I have attached the most recent iteration of Simon's LLVM fix for this > along with some of my adjustments to it and a backport to LLVM 4.0 I > used for testing with rust. > > LLVM 128-bit integer isues (mips, mipsel) > > 128-bit integer arithmetic is broken in LLVM 4 on 32-bit MIPS. This has > been fixed in LLVM 5. Unfortunately this bug has caused the upstream > stage0 compiler to break badly so you may need to apply the fix and > cross build your own stage0 until upstream binaries have moved to LLVM >= 5. > > https://github.com/rust-lang/rust/issues/41222 > https://bugs.llvm.org/show_bug.cgi?id=32713 > https://reviews.llvm.org/rL305389 > > llvm.powi.f64 broken (mips64el) > test f32::f32::tanh_0 ... FAILED test f64::f64::tanh_0 ... FAILED > > This is caused by the llvm.powi.f64 intrinsic being broken for negative > powers. Thankfully this is not a very serious bug. > > Fixed here: > https://bugs.llvm.org/show_bug.cgi?id=36061 > https://reviews.llvm.org/D42537 > > No 64-bit Rust ABI support (mips64el) > > The biggest issue with Rust itself is the lack of working FFI support > for 64-bit mips. > > My attempt at implementing this: > https://github.com/rust-lang/rust/issues/47290 > https://github.com/rust-lang/rust/pull/47964 > > Other issues > test net::tcp::tests::timeouts ... FAILED test net::udp::tests::timeouts ... FAILED test sys::imp::ext::net::test::timeouts ... FAILED > > As I wrote in https://github.com/rust-lang/rust/issues/39013, these are > big endian specific. I haven't had the chance to test rust on big endian > yet. > test io::error::Error::from_raw_os_error_0 ... FAILED > > Fixed in https://github.com/rust-lang/rust/pull/47874 > > Some FPU errors only happen on Loongson. In theory, enabling FPXX in > LLVM should have fixed this, but I have not investigated it very far. > > There are a few other failing tests where the test itself is broken. I > will try to submit fixes to ignore / adjust the tests soon. Any progress on these issues? I see that rust built on mips64el but it now hangs. And mips/el are still uncompiled. Cheers, Emilio
Re: Bug#881845: rustc: FTBFS on mips*: test failures
On 17/11/17 13:25, James Cowgill wrote: > Hi, > > On 15/11/17 17:30, Emilio Pozuelo Monfort wrote: >> Source: rustc >> Version: 1.21.0+dfsg1-3 >> Severity: important >> >> Hi, >> >> Sometime ago I asked about rustc bootstrap status on mips*: >> >> 17:08 < infinity0> mips* fail many tests last time i tried, ~3 months ago, i >> didn't want to ship it, i haven't had time to check since >> >> >From >> >https://buildd.debian.org/status/fetch.php?pkg=rustc=mips=1.14.0%2Bdfsg1-3=1484077706=0 >> >> test net::tcp::tests::timeouts ... FAILED >> test net::udp::tests::timeouts ... FAILED >> test sys::imp::ext::net::test::timeouts ... FAILED >> >> Looks like timeouts are broken on rust/mips? >> >> mips64el has different errors: >> >> test f32::f32::tanh_0 ... FAILED >> test f64::f64::tanh_0 ... FAILED >> test io::error::Error::from_raw_os_error_0 ... FAILED >> >> Maybe the mips porters can take a look? (debian-mips@ on Cc). Note those >> errors are for 1.14.0, you'll want to try with a newer version first and >> see what's the current status. > > I just tried building the latest rustc on mips64el. As I expected, I hit > this LLVM bug again where any code using atomics will hang: > > https://bugs.llvm.org/show_bug.cgi?id=32020 > > I'll see if I can get Simon to have a look at it again. FTR: According to our conversation over IRC, there's a patch for the above bug, but mips64el is now affected by https://github.com/rust-lang/rust/issues/47290 mips(el) may be fine now with that patch, or they may not :P Cheers, Emilio
Re: Bug#881845: rustc: FTBFS on mips*: test failures
On Fri, 17 Nov 2017 12:25:35 + James Cowgillwrote: > Hi, > > On 15/11/17 17:30, Emilio Pozuelo Monfort wrote: > > Source: rustc > > Version: 1.21.0+dfsg1-3 > > Severity: important > > > > Hi, > > > > Sometime ago I asked about rustc bootstrap status on mips*: > > > > 17:08 < infinity0> mips* fail many tests last time i tried, ~3 months ago, > > i didn't want to ship it, i haven't had time to check since > > > >>From > >>https://buildd.debian.org/status/fetch.php?pkg=rustc=mips=1.14.0%2Bdfsg1-3=1484077706=0 > > > > test net::tcp::tests::timeouts ... FAILED > > test net::udp::tests::timeouts ... FAILED > > test sys::imp::ext::net::test::timeouts ... FAILED > > > > Looks like timeouts are broken on rust/mips? > > > > mips64el has different errors: > > > > test f32::f32::tanh_0 ... FAILED > > test f64::f64::tanh_0 ... FAILED > > test io::error::Error::from_raw_os_error_0 ... FAILED > > > > Maybe the mips porters can take a look? (debian-mips@ on Cc). Note those > > errors are for 1.14.0, you'll want to try with a newer version first and > > see what's the current status. > > I just tried building the latest rustc on mips64el. As I expected, I hit > this LLVM bug again where any code using atomics will hang: > > https://bugs.llvm.org/show_bug.cgi?id=32020 > > I'll see if I can get Simon to have a look at it again. Any luck there? (I don't see a ping on the bug report). librsvg is now using rustc. We're stuck with an old version for now, but when we upgrade to the rust version, I wouldn't want to have to remove it, along with all the rdeps, from a bunch of architectures... Hence my insistence :) Cheers, Emilio
Re: Bug#881845: rustc: FTBFS on mips*: test failures
Hi, On 15/11/17 17:30, Emilio Pozuelo Monfort wrote: > Source: rustc > Version: 1.21.0+dfsg1-3 > Severity: important > > Hi, > > Sometime ago I asked about rustc bootstrap status on mips*: > > 17:08 < infinity0> mips* fail many tests last time i tried, ~3 months ago, i > didn't want to ship it, i haven't had time to check since > >>From >>https://buildd.debian.org/status/fetch.php?pkg=rustc=mips=1.14.0%2Bdfsg1-3=1484077706=0 > > test net::tcp::tests::timeouts ... FAILED > test net::udp::tests::timeouts ... FAILED > test sys::imp::ext::net::test::timeouts ... FAILED > > Looks like timeouts are broken on rust/mips? > > mips64el has different errors: > > test f32::f32::tanh_0 ... FAILED > test f64::f64::tanh_0 ... FAILED > test io::error::Error::from_raw_os_error_0 ... FAILED > > Maybe the mips porters can take a look? (debian-mips@ on Cc). Note those > errors are for 1.14.0, you'll want to try with a newer version first and > see what's the current status. I just tried building the latest rustc on mips64el. As I expected, I hit this LLVM bug again where any code using atomics will hang: https://bugs.llvm.org/show_bug.cgi?id=32020 I'll see if I can get Simon to have a look at it again. James signature.asc Description: OpenPGP digital signature