Spawn is a small library exposing only one function:
Spawn.spawn
. Its purpose is to start command in the
background. Spawn aims to provide a few missing features of
Unix.create_process
such as providing a working directory as well as
improving error reporting and performance.
Errors such as directory or program not found are properly reported as
Unix.Unix_error
exceptions, on both Unix and Windows.
On Unix, Spawn uses vfork
by default as it is often a lot faster
than fork. There is a benchmark comparing Spawn.spawn
to
Unix.create_process
in spawn-lib/bench
. If you don't trust
vfork
, you can set the environment variable SPAWN_USE_FORK
to make
Spawn use fork
instead.
Spawn is expected to be fully portable. However, so far it has only been tested on Linux, OSX and Windows.
On Windows, Spawn.spawn
simply uses the
CreateProcess Windows function.
Under Linux, it uses a custom implementation that relies on
fork
/vfork
followed by execve
. Compared to other implementations
such as Unix.create_process
or the posix_spawn C
library call, our implementations supports the following:
- setting the current working directory of the child process
- reporting errors from
execve
to the caller
To report execve
errors, our implementation proceeds as follow: just
before calling fork
or vfork
we create a pipe with the O_CLOEXEC
flag. After forking, the parent reads from this pipe. If the execve
succeeds, the pipe get closed, the read
in the parent returns 0
bytes and the parent concludes that the execve
succceeded. If the
execve
fails, the child process sends the error code to the parent
via the pipe and exits.