Weird behavior when building sys crate with multiple make jobs in docker container

[Christian-Prather]

Hello,
I am running into a weird problem that I believe will go beyond just Corrosion however as I am only seeing it when using Corrosion I figured I would start here and see if anyone had any advice. Thank you in advance!

Problem

The problem occurs when using corrosion to build a rust bin that relies on the libffi-sys crate https://crates.io/crates/libffi-sys/2.0.0. This external libffi-sys crate internally builds the libffi C library https://github.com/libffi/libffi using the cc crate https://crates.io/crates/cc as a dependency. The weirdness comes in the fact that libffi-sys fails to build under specific circumstances however this particular setup is critical for me to be able to use Corrosion.

Setup

I am using the following build setup:
Docker image based on Ubuntu 18.04
CMake version: 3.30.5
rustc version: rustc 1.78.0 (9b00956e5 2024-04-29)

My test application can be found here but is essentially hello world bin with a basic cmake.
https://github.com/Christian-Prather/test-corrosion

Here are my commands I use to build the bin.

cd test-corrosion
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Debug ..
make -j4

This will fail with the following log.
multiple-2.txt

However this is where it gets weird, if I run the same command without -j4 so just make or make -j1 it succeeds and produces and executable binary? I also am able to build with make -j4 just fine if I am NOT in a docker container.

Here is the log for make -j1 inside the docker image which succeeds:
single-2.txt

I know that since this only appears to occur in Docker that the problem likely does not fall entirely on corrosion however I was hoping anyone might have some thoughts on this. Unfortunately running in Docker for our builds is a hard requirement so even though it works without that's not an option.

I am wondering if this has something to do with the make jobserver not being handled correctly in the Docker image however if that was true I would have assumed that the basic C++ program would of also failed with make -jn; n > 1?

Any help at all would be really really appreciated as this seems to be the last hangup in porting a lot of CMake magic over to corrosion. Thank you,

Additional Info:

Below is a table of test builds and their results

Scenario	Success / Fail build
Regular cargo build	Works in and out of Docker image
make -j8	Works out of Docker image fails inside
make -j1	Works in and out of Docker image
Regular c++ project running make -j8 (parallel in my repo I posted)	Works in and out of Docker image
Building libffi C library directly	Works in and out of Docker image
Corrosion make -j8 without libffi-* crates	Works in and out of Docker image

I have also tried setting cargo build flag if -j to 1 and codegen-units to 1.

[jschwe]

make[5]: *** read jobs pipe: Resource temporarily unavailable. Stop.

You use a top-level make (generated from cmake), which spawns a make job-server.
Which make version do you have inside and outside of docker? I know something changed in make 4.4 with regards to the job server.
I believe the child make process from lib-ffi-sys will try to respect the make jobserver from the parent make, but it seems there is a problem there. This can be due to buggy make code, but I'm not a make expert, so you'll have to due some digging yourself to find out whats going wrong with the jobserver.

[Christian-Prather]

(Sorry was out of town) Ugh that was it, was on a version of make < 4.4 that worked a treat thank you once again!