Invalid memory access on Alpine Linux

[snacks02]

Sometimes Ameba randomly fails in GitHub Actions with:

ci.yml:

name: GitHub Actions

on: push

jobs:
  check:
    name: Check
    runs-on: ubuntu-latest
    container:
      image: crystallang/crystal:latest-alpine
    steps:
      - uses: actions/checkout@v3
      - run: shards install
      - run: bin/ameba
      - run: crystal tool format --check
      - run: crystal spec

shard.yml:

development_dependencies:
  ameba:
    github: crystal-ameba/ameba
    version: ~> 1.4.0

Rerunning fixes the issue, but would be great to fix this this in Ameba.

Happy to provide additional information if needed.

[Sija]

Perhaps using alpine (musl) might have sth to do with it?

[snacks02]

Probably so. Never had it happen on Arch with glibc.

[Sija]

I'd suggest trying with a regular crystal:latest image and checking whether it exhibits the same behaviour to begin with.
Regardless, this is most likely Crystal bug, not ameba's.

[Sija]

In any case it would be good to debug it further and eventually report the buggy behaviour to the crystal's issue tracker instead.

[jwoertink]

Just got this issue on Cable https://github.com/cable-cr/cable/actions/runs/4774224693/jobs/8487769812?pr=69

 Postinstall of ameba: make bin && make run_file
Failed postinstall of ameba on make bin && make run_file:
shards build -Dpreview_mt
Dependencies are satisfied
Building: ameba
Error target ameba failed to compile:
Showing last frame. Use --error-trace for full trace.

In src/ameba/rule/base.cr:142:10

 142 | {{ run("../../contrib/read_type_doc",
          ^--
Error: Error executing run (exit code: 11): ../../contrib/read_type_doc QueryBoolMethods /__w/cable/cable/lib/ameba/src/ameba/rule/style/query_bool_methods.cr


stderr:

    Error while trying to determine if a stack overflow has occurred. Probable memory corruption
    Invalid memory access (signal 11) at address 0x38
    [0x55caacd58e56] ?? +94328971431510 in /github/home/.cache/crystal/__w-cable-cable-lib-ameba-src-contrib-read_type_doc.cr/macro_run
    [0x55caacd58e1a] ?? +94328971431450 in /github/home/.cache/crystal/__w-cable-cable-lib-ameba-src-contrib-read_type_doc.cr/macro_run
    [0x7ff58ba73c9f] ?? +140692586708127 in /lib/ld-musl-x86_64.so.1


make: *** [Makefile:9: build] Error 1
Error: Process completed with exit code 1.

Seems it only happened on Crystal 1.6. I'm also using crystallang/crystal:${{ matrix.crystal_version }}-alpine, so I'll give that a shot to try not using alpine.

[jwoertink]

Confirmed, taking off the -alpine fixed it 👍

[daliborfilus]

Using ameba (alpine base) build crashes on our CI too, this is our latest crash:

$ ameba
Error while trying to determine if a stack overflow has occurred. Probable memory corruption
Invalid memory access (signal 11) at address 0x38
[0x55c44399fd36] *Exception::CallStack::print_backtrace:Nil +118 in /usr/bin/ameba
[0x55c4439845f6] ~procProc(Int32, Pointer(LibC::SiginfoT), Pointer(Void), Nil) +310 in /usr/bin/ameba
[0x7f10416735f0] ?? +139707793487344 in /lib/ld-musl-x86_64.so.1

but only once in a week. 99% of the time it works.
We have our own ameba docker alpine image built using:

FROM alpine:edge as builder
RUN apk add --update crystal shards yaml-dev musl-dev make
RUN mkdir /ameba
WORKDIR /ameba
COPY . /ameba/
RUN make clean && make

FROM alpine:latest
RUN apk add --update yaml pcre2 gc libevent libgcc
RUN mkdir /src
WORKDIR /src
COPY --from=builder /ameba/bin/ameba /usr/bin/
RUN ameba -v
ENTRYPOINT [ "/usr/bin/ameba" ]

The image is rebuilt every day using these:

build:
  stage: build
  script:
    # clone to subdirectory "ameba"
    - git clone --depth=1 https://github.com/crystal-ameba/ameba.git ameba
    # add our patched Dockerfile
    - cp Dockerfile ameba/Dockerfile
    - cd ameba
    # build
    - docker build --pull -t "${CI_REGISTRY_IMAGE}:latest" .
    # push
    - docker push "${CI_REGISTRY_IMAGE}:latest"

(The Dockerfile is based of ameba's Dockerfile.)