I guess I find writing portable shell scripts to be tedious to get right (See https://www.gnu.org/software/autoconf/manual/autoconf-2.64/html_node/Portable-Shell.html).  But, if someone can do it I have no objection.

The tests I was proposing could go in their own .tst file under the test/ directory or within their own directory under the test/ directory so I think it is easily extensible.

I hadn't been targeting running in existing installations, but it seems like it wouldn't be that hard to distinguish between run-octave and a regular octave run.

When using run-octave,

program_invocation_name
ans = /home/rik/wip/Projects_Mine/octave-dev/src/.libs/octave-gui

When using an installed version of Octave

program_invocation_name
ans = /usr/local/libexec/octave/5.1.0/exec/x86_64-pc-linux-gnu/octave-gui

@Mike: Testing for the exit status seems like a useful behavior.  However, I'd rather avoid a shell script (which can be different between different Unix flavors and certainly different on Windows).  I wonder if the 4 tests can be re-written to occur within Octave by using the system() call?

For example, I re-wrote the third test (non-zero exit status) as

[st,out] = system ("./run-octave -f --eval 'exit(2)'")
st = 2
out =

What do you think?

(Apologies: the original submission had some formatting problems.)

I use Octave primarily as a command-line script interpreter (i.e. I run executable scripts with #!/usr/bin/octave to invoke Octave) for scientific applications on computer clusters. It is important for this application that Octave conform to standard exit status behaviour:

• zero exit status is returned on success, either through normal completion or by calling exit(0), and

• non-zero exit status is returned on failure, either through raising an error() or by calling exit(n) with n != 0.

In Octave 4.2.2 the standard exit status behaviour was broken:

> $ octave --version
> GNU Octave, version 4.2.2
> ...
>
> $ echo '0;' > script.m; octave script.m; echo $?
> 0   # ok
>
> $ echo 'error("msg");' > script.m; octave script.m; echo $?
> error: msg
> error: called from
>     script at line 1 column 1
> 0   # should return not 0
>
> $ echo 'exit(0);' > script.m; octave script.m; echo $?
> terminate called after throwing an instance of 'octave::exit_exception'
> panic: Aborted -- stopping myself...
> Aborted
> 134   # should return 0
>
> $ echo 'exit(1);' > script.m; octave script.m; echo $?
> terminate called after throwing an instance of 'octave::exit_exception'
> panic: Aborted -- stopping myself...
> Aborted
> 134   # should return 1

However it seems to have been fixed recently in development:

> $ cat ./HG-ID
> 3d5f953e2ef6
>
> $ echo '0;' > script.m; ./run-octave script.m; echo $?
> 0   # ok
>
> $ echo 'error("msg");' > script.m; ./run-octave script.m; echo $?
> error: msg
> error: called from
>     script at line 1 column 1
> 1   # ok
>
> $ echo 'exit(0);' > script.m; ./run-octave script.m; echo $?
> 0   # ok
>
> $ echo 'exit(1);' > script.m; ./run-octave script.m; echo $?
> 1   # ok

I was unable to find a unit test for Octave's exit status behaviour, so the attached patch adds such a unit test to ensure the exit status behaviour is not broken in future. The patch adds a script ./test-octave-exit-status.sh which runs Octave code snippets either as a script or using --eval, and then tests Octave's exit status for the standard behaviour given above.

The script ./test-octave-exit-status.sh is called from "make check" before the rest of the Octave test suite; a failed test in ./test-octave-exit-status.sh will cause "make check" to fail; since the exit status tests are separate from the rest of the Octave test suite, a silent test failure might otherwise go unnoticed.