Extremely long compile times when compiling examples/sync.rs with nightly compiler

[earthnuker]

Hello everyone,

after copying the content of examples/sync.rs into an empty rust project and adding the following versions of the dependencies:

[dependencies]
anyhow = "1.0.48"
async-std = { version = "1.10.0", features = ["attributes"] }
futures = "0.3.18"
ipfs-embed = "0.22.4"
libipld = "0.12.0"
rand = "0.8.4"
tracing-subscriber = {version="0.3.2", features=["env-filter"]}

when trying to compile it with either the latest stable (rustc 1.56.1 (59eed8a2a 2021-11-01)) or nightly (rustc 1.58.0-nightly (dd549dcab 2021-11-25)) nightly-x86_64-pc-windows-msvc toolchain on Windows 10 i get compile time exceeding 2 hours, i can provide a verbose compiler log if required.

Best regards,

Earthnuker

[dvc94ch]

Do you have insane compile times compiling other rust projects? What's the CPU/mem utilisation during that time?

[dvc94ch]

I'll guess you have a very low powered device and 90% of your CPU/mem is being used by windows to monitor you and send a real time voice/video feed to the central machine learning center.

[earthnuker]

Hello, thanks for the quick reply,

i just woke up and it did finish compiling after 6 hours and 40 minutes, but when i try to run it the executable overflows its stack and crashes.

Do you have insane compile times compiling other rust projects? What's the CPU/mem utilisation during that time?

no, this is the only dependency that has caused the issue so far.

my CPU is a Intel i7-4770K (8) @ 3.5GHz, 32 GB of DDR3 RAM

rustc uses one CPU core and about 750MB of memory during the phase where it is slow

I'll guess you have a very low powered device and 90% of your CPU/mem is being used by windows to monitor you and send a real time voice/video feed to the central machine learning center.

the same behavior occurs on a Linux machine with a Intel i5-2400 (4) @ 3.400GHz, 20 GB of RAM, and don't you think the "haha windows 10 is spying on you all the time" joke is getting a bit old?

Best regards,

Earthnuker

[dvc94ch]

on a fresh checkout it builds for me in a bit over a minute.

time cargo build
________________________________________________________
Executed in   87.85 secs    fish           external
   usr time  566.68 secs   65.78 millis  566.61 secs
   sys time   48.83 secs    6.16 millis   48.82 secs

after that I built the examples

time cargo build --examples
________________________________________________________
Executed in   34.02 secs    fish           external
   usr time  149.15 secs    0.00 micros  149.15 secs
   sys time   11.37 secs  746.00 micros   11.37 secs

don't you think the "haha windows 10 is spying on you all the time" joke is getting a bit old?

well, I doubt android, ios or macos are much better. why do you think amazon built alexa if not to spy on you?

[earthnuker]

on a fresh checkout it builds for me in a bit over a minute.

time cargo build
________________________________________________________
Executed in   87.85 secs    fish           external
   usr time  566.68 secs   65.78 millis  566.61 secs
   sys time   48.83 secs    6.16 millis   48.82 secs

after that I built the examples

time cargo build --examples
________________________________________________________
Executed in   34.02 secs    fish           external
   usr time  149.15 secs    0.00 micros  149.15 secs
   sys time   11.37 secs  746.00 micros   11.37 secs

don't you think the "haha windows 10 is spying on you all the time" joke is getting a bit old?

well, I doubt android, ios or macos are much better. why do you think amazon built alexa if not to spy on you?

on a fresh checkout

• cargo build runs without issue in about 2 minutes

• cargo build --examples still hangs

i tried switching to the latest nightly-x86_64-pc-windows-gnu toolchain, same result

[cryptoquick]

@earthnuker, you can try cross-compiling from linux or WSL:

https://github.com/rust-embedded/cross

Also, @dvc94ch, you gotta remember, not everybody is uncomfortable being spied on. Some people don't care how much corporations are being paid to manipulate them as long as they're handing out free stuff.

[dvc94ch]

We can probably leave the spying stuff, it is admittedly off topic. You could try using the lld linker. Linking takes a lot of ram, I assume that's where it's hanging.

…

On Sat, Nov 27, 2021, 21:26 Hunter Trujillo ***@***.***> wrote: @earthnuker <https://github.com/Earthnuker>, you can try cross-compiling from linux or WSL: https://github.com/rust-embedded/cross Also, @dvc94ch <https://github.com/dvc94ch>, you gotta remember, not everybody is uncomfortable being spied on. Some people don't care how much corporations are being paid to manipulate them as long as they're handing out free stuff. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#109 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFVDL7TYP5ZF52MF2SI6GDUOE5GRANCNFSM5I3THPXQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.

[earthnuker]

@earthnuker, you can try cross-compiling from linux or WSL:

https://github.com/rust-embedded/cross

Also, @dvc94ch, you gotta remember, not everybody is uncomfortable being spied on. Some people don't care how much corporations are being paid to manipulate them as long as they're handing out free stuff.

using a cross compiler does not solve the issue

We can probably leave the spying stuff, it is admittedly off topic. You could try using the lld linker. Linking takes a lot of ram, I assume that's where it's hanging.

as i said before, this machine has 32 GB of memory, the server i'm compiling on has 20 GB of memory, and the rust compiler barely uses a gigabyte of memory

[cryptoquick]

@earthnuker if you can use WSL to cross compile the program, and it can run on Windows, I think that's a reasonable workaround.

[earthnuker]

@earthnuker if you can use WSL to cross compile the program, and it can run on Windows, I think that's a reasonable workaround.

thing i tried so far:

• nightly and stable compiler natively on windows 10 (both msvc and gnu targets): hangs
• nightly and stable compiler natively on linux: hangs
• nightly compiler targeting x86_64-pc-windows-gnu through cross running on a linux machine: hangs

why would using WSL fix the issue?

update: compiling in WSL also hangs at Building [=======================> ] 410/411: ipfs(bin) (ipfs is the name of my test crate containing the sync.rs example code)

[dvc94ch]

don't know what to tell you other than maybe try a different linker

[earthnuker]

tried with

[target.x86_64-pc-windows-msvc]
rustflags = [
    "-C","linker-flavor=lld-link"
]

no change

update: did some more testing, stripped out all dependencies that are not needed, so now my Cargo.toml is

[package]
name = "ipfs_test"
version = "0.1.0"
edition = "2021"

[dependencies]
async-std = { version = "1.10.0", features = ["attributes"] }
ipfs-embed = "0.22.4"
libipld = "0.12.0"

and my code is

use ipfs_embed::{generate_keypair, Config, Ipfs};
use libipld::store::DefaultParams;

#[async_std::main]
async fn main() {
    let mut config = Config::new("out".as_ref(), generate_keypair());
    config.network.kad = None;
    let a = Ipfs::<DefaultParams>::new(config).await.unwrap();
}

that seems to be the minimum necesarry to reproduce the issue for me, as soon as i remove the line that calls the new method of the Ipfs struct it compiles fine

[wngr]

Running your sample on Linux (target x86_64-unknown-linux-gnu) works as expected:

✗ time cargo build
// ...
cargo build  675.98s user 52.79s system 634% cpu 1:54.87 total
✗ rustc --version
rustc 1.56.1 (59eed8a2a 2021-11-01)

Maybe you have some global .cargo settings or something? I guess this can only be debugged if you're able to reproduce this on a vanilla host, maybe a public github repo using github actions?

Edit: Added said repo at https://github.com/wngr/test-109, CI job works even w/ x86_64-pc-windows-msvc.

[earthnuker]

did some more testing:

• clean docker container using https://github.com/Earthnuker/ipfs_embed_109 and docker build .
• tested on:
  • github CI
  • the linux machine with 20 GB of RAM, hangs
  • a VM on my brother's VM host, hangs, killed after ~20 minutes at Compiling ipfs_test
  • a friend's Linux machine, hangs, killed after ~20 minutes at Compiling ipfs_test
  • another friend's Linux machine, hangs, killed after ~20 minutes at Compiling ipfs_test
    • perf trace has the following to say:

      204698.147 (500.112 ms): futex(uaddr: 0x55e1db6c4bcc, op: WAIT|PRIVATE_FLAG, utime: 0x7fff7e6b3ed0) = -1 ETIMEDOUT (Connection timed out)
      205198.273 ( 0.002 ms): futex(uaddr: 0x55e1db6c5590, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      205198.286 (500.083 ms): futex(uaddr: 0x55e1db6c4bcc, op: WAIT|PRIVATE_FLAG, utime: 0x7fff7e6b3ed0) = -1 ETIMEDOUT (Connection timed out)
      205698.382 ( 0.001 ms): futex(uaddr: 0x55e1db6c5590, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      205698.395 (500.084 ms): futex(uaddr: 0x55e1db6c4bcc, op: WAIT|PRIVATE_FLAG, utime: 0x7fff7e6b3ed0) = -1 ETIMEDOUT (Connection timed out)
      206198.490 ( 0.001 ms): futex(uaddr: 0x55e1db6c5590, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      206198.504 (500.113 ms): futex(uaddr: 0x55e1db6c4bcc, op: WAIT|PRIVATE_FLAG, utime: 0x7fff7e6b3ed0) = -1 ETIMEDOUT (Connection timed out)
      206698.628 ( 0.001 ms): futex(uaddr: 0x55e1db6c5590, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      206698.642 (500.061 ms): futex(uaddr: 0x55e1db6c4bcc, op: WAIT|PRIVATE_FLAG, utime: 0x7fff7e6b3ed0) = -1 ETIMEDOUT (Connection timed out)
      207198.713 ( 0.001 ms): futex(uaddr: 0x55e1db6c5590, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      207198.726 (500.125 ms): futex(uaddr: 0x55e1db6c4bcc, op: WAIT|PRIVATE_FLAG, utime: 0x7fff7e6b3ed0) = -1 ETIMEDOUT (Connection timed out)
      207698.864 ( 0.001 ms): futex(uaddr: 0x55e1db6c5590, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      207698.880 (500.074 ms): futex(uaddr: 0x55e1db6c4bcc, op: WAIT|PRIVATE_FLAG, utime: 0x7fff7e6b3ed0) = -1 ETIMEDOUT (Connection timed out)
      208198.964 ( 0.001 ms): futex(uaddr: 0x55e1db6c5590, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      

i also i pushed a second commit to the test repo that removes the codegen-units = 1 options from Cargo.toml, although that has made no difference on my windows machine

[earthnuker]

the github CI job with codegen-units=1 finished after 4 hours and 20 minutes, the job without codegen-units=1 is still running, clocking in at over 5 hours now and will likely be killed by a timeout before it finishes.