• Change to the root directory in the parent (dumb-init) process after fork()ing (#210).

  Thanks to @Villemoes for the patch!

• Actually fix the bug that can cause --help or --version to crash in some scenarios (#215).

  Thanks to @suve for the patch!

Note on binary naming: In #202, we switched from using Debian naming for architectures to the Linux kernel's naming for binaries. The Debian package names are unchanged. That means binaries now have these architectures in their names:

• x86_64 (previously amd64)
• aarch64 (previously arm64)
• ppc64le (previously ppc64el)

• Fix a bug that can cause --help or --version to crash in some scenarios (#213).

  Thanks to @suve for the patch!

Note on binary naming: In #202, we switched from using Debian naming for architectures to the Linux kernel's naming for binaries. The Debian package names are unchanged. That means binaries now have these architectures in their names:

• x86_64 (previously amd64)
• aarch64 (previously arm64)
• ppc64le (previously ppc64el)

• Fix a race condition which can cause the child to receive SIGHUP and SIGCONT very shortly after start (#174).

  In general this was very rare, but some environments (especially some container and virtualization environments) appear to encounter it at a much higher rate, possibly due to scheduler quirks.

Note on binary naming: In #202, we switched from using Debian naming for architectures to the Linux kernel's naming for binaries. For compatibility, we've kept the old names around in this release. This means that these files are identical:

• dumb-init_1.2.2_x86_64 (old name dumb-init_1.2.2_amd64)
• dumb-init_1.2.2_aarch64 (old name dumb-init_1.2.2_arm64)
• dumb-init_1.2.2_ppc64le (old name dumb-init_1.2.2_ppc64el)

In the next release, only the binaries with the updated architecture naming will be uploaded. The Debian packages will continue to be named as-is.

• Fix verbose debug logging for ignored signals.

  Before this patch, they were reported in the verbose log as "forwarded
  signal 0 to children" instead of "not forwarding signal to children".

  Since signal 0 is a noop, there is no actual behavior change here.

  Thanks @kpengboy for the patch!

• Fix Python package installation on platforms without -static support (e.g. Mac OS X)

  Thanks @NicolasLM for the issue and @asottile for the patch!

• Hand the controlling TTY to the child process, if we have one (#122).

  This fixes warnings that are printed when running a typical command like:
  docker run -ti dumb-init bash
  ...as well as allowing you to use job control.

  Thanks to @Ehlers for the patch, and @alhafoudh (and several others) for
  reporting the issue and providing details!

• Add support for FreeBSD kernel (#109). Thanks @onlyjob for bringing this to our attention!

• Fix race when the child exits very quickly that leads to dumb-init not reaping the child. This should be pretty rare and most likely to happen when dumb-init's child fails to exec (such as when you try to run a file that doesn't exist). (#104)

• Fix segfault when passing unknown arguments (thanks @asottile for noticing!) (#88).

• Add ability to rewrite incoming signals before proxying via the --rewrite flag. Thanks @mcclurmc for the PR (#83)!
• Add ability to not rewrite incoming "suspend" job control signals in setsid mode (#85).
• Add ability to ignore (not proxy) incoming signals (#86).