Empty env variables are not preserved when S6_KEEP_ENV is disabled using with-contenv

[ximarx]

When S6_KEEP_ENV is disabled, empty but defined environment variables are not preserved when executing commands via with-contenv.

---

Steps

Using the following docker-compose.yaml:

# compose.yaml
x-service-def:
  &service-def
  build:
    context: .
    dockerfile_inline: |
      # syntax=docker/dockerfile:1.17.0
      FROM alpine
      ADD https://github.com/just-containers/s6-overlay/releases/download/v3.2.1.0/s6-overlay-noarch.tar.xz /tmp
      RUN ls -al /tmp
      RUN tar -v -C / -Jxpf /tmp/s6-overlay-noarch.tar.xz
      ADD https://github.com/just-containers/s6-overlay/releases/download/v3.2.1.0/s6-overlay-x86_64.tar.xz /tmp
      RUN tar -v -C / -Jxpf /tmp/s6-overlay-x86_64.tar.xz
      ENTRYPOINT ["/init"]
  platform: linux/amd64
  command:
    - /command/with-contenv
    - printenv

# Set of environment variables
x-env-def:
  &environments
  S6_VERBOSITY: 0
  NOT_EMPTY_VARIABLE: 'not-empty'
  EMPTY_BUT_DEFINED_VARIABLE: ''
  MULTI_LINE: |-
    first line
    second line

services:
  with_keep_env_disabled:
    <<: *service-def
    environment:
      <<: *environments

  with_keep_env_enabled:
    <<: *service-def
    environment:
      <<: *environments
      S6_KEEP_ENV: 1

1. Run docker compose run --rm with_keep_env_disabled.
2. Observe lack of env EMPTY_BUT_DEFINED_VARIABLE in output
   E.g.

NOT_EMPTY_VARIABLE=not-empty
MULTI_LINE=first line
second line
HOME=/root
S6_VERBOSITY=0
SHLVL=2
PATH=/command:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
TERM=xterm
HOSTNAME=8869c1e90a28
PWD=/

3. Execute docker compose run --rm with_keep_env_enabled
4. Observe presence of env EMPTY_BUT_DEFINED_VARIABLE

HOSTNAME=419b37a708f4
SHLVL=3
HOME=/root
OLDPWD=/
NOT_EMPTY_VARIABLE=not-empty
S6_VERBOSITY=0
TERM=xterm
EMPTY_BUT_DEFINED_VARIABLE=
PATH=/command:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
MULTI_LINE=first line
second line
PWD=/
S6_KEEP_ENV=1

The expectation would be to have same behavior with and without S6_KEEP_ENV

---

If you've identified a fix, please open a pull request.

Maybe I identified the cause of the issue. I can open a pull request with some minimal guidance on the best resolution possible

From my analysis the issue relies on the fact that env dir /run/s6/container_environment is created using s6-linux-init-maker -s /run/s6/container_environment when S6_KEEP_ENV is disabled ref at stage0.

Doing so s6-linux-init is instructed to create an env dir which should be (quoting doc) compatible with s6-envdir -fn (see https://skarnet.org/software/s6-linux-init/s6-linux-init-maker.html).

• Envdir env_dump documentation function https://skarnet.org/software/skalibs/libstddjb/env.html
• Place where s6-linux-init env_dump is performed: https://github.com/skarnet/s6-linux-init/blob/main/src/init/s6-linux-init.c#L283
• env_dump implementation https://github.com/skarnet/skalibs/blob/main/src/libenvexec/env_dump.c

with-contenv loads the env dir s6-envdir -Lfn (https://github.com/just-containers/s6-overlay/blob/master/layout/rootfs-overlay/package/admin/s6-overlay-%40VERSION%40/command/with-contenv#L10C3-L10C17), so whole file, no chomp

Possible resolutions I can think of:

1. Modify env_dump behavior to match s6-dumpenv -N and update with-contenv to use s6-envdir -Lf
2. Avoid relying on s6-linux-init-maker for envdir creation when S6_KEEP_ENV=0. Instead, use: s6-dumpenv -N and update with-contenv to use s6-envdir -Lf.

[skarnet]

That was an amazingly complete bug-report, thank you for this. I'll find a way to make the behaviour consistent.

[skarnet]

I aligned everything on the s6-envdir -f behaviour (verbatim but with chomping a terminating newline) because it was the only way to make everything consistent. That requires an API change to s6-linux-init, so the change will have to wait for a full s6-rc release, sorry.

[ximarx]

For my personal knowledge, may I ask why -f (limiting to 4k env content) and not -Lf?

so the change will have to wait for a full s6-rc release, sorry.

Do you have a rough idea of when it will happen? (no pressure, just to know)

[skarnet]

Not sure: the code is very much in flux at the moment. Could be as soon as next month, or as late as Christmas. I'm sorry I can't narrow it down.

[GreyXor]

Not sure: the code is very much in flux at the moment. Could be as soon as next month, or as late as Christmas. I'm sorry I can't narrow it down.

Hello gentle ping i'm curious if there's already some progress on this ? no pressure! Thanks a lot skarnet, s6 (and s6-overlay) is very valuable for us!

[skarnet]

Yes! I released the version of s6-linux-init that fixes the problem last week. In order to cut a new s6-overlay release, though, I need to also cut an s6-rc release first - and it is almost ready but I still have some documentation to write, and it's going slow because I'm sick 🤧

Expect something around early January. Happy holidays!

[GreyXor]

Yes! I released the version of s6-linux-init that fixes the problem last week. In order to cut a new s6-overlay release, though, I need to also cut an s6-rc release first - and it is almost ready but I still have some documentation to write, and it's going slow because I'm sick 🤧

Expect something around early January. Happy holidays!

Just, awesome! Hey take care

[skarnet]

Version 3.2.2.0 is released! Can you please test it and check that it fixes the S6_KEEP_ENV bug as intended?