Luaver kill the whole batch process on error

[umireon]

The following script will not work as expected:

. ~/.luaver/luaver
if luaver install-luarocks install 5.3.1; then
    echo ok
else
    echo Something wrong
fi

The result was:

$ sh test.sh
No lua version set

$

while printing Something wrong is the expected behavior.

This is caused by luaver#L34, which kills not only the luaver function but also the whole batch process.
This behavior makes difficult to use luaver on the CI.

Since there is no exception handling mechanism on shells (kill -INT $$ is appearently not working), I suggest each command has explicit error handling using && and if.

[DhavalKapil]

Can you elaborate on the issue. I see it is a problem but exactly how is it affecting the CI?

[umireon]

(I am sorry but I cannot fully understand why kill -INT $$ works; I cannot find any spec to describe what is happen when I shoot myself in the foot with shell by SIGINT)

tl;dr luaver behaves like exit on error and this is hard to understand since no one expect that.

Of course, the current luaver is working when there are no errors.
The problem is in debugging, the behavior of kill -INT $$ makes debugging very hard.

For example, travis_retry won't work as expected.
In CI sessions, retrying is essential for stable builds because download failure is one of the common problems.
The following code is a typical pattern to realize the retrying.

install: for i in 1 2 3; do luaver install 5.3.2; done
# essentially equivalent to `travis_retry luaver install 5.3.2`

However, this will never work as expected because luaver kills the ongoing shell process on error, which cause the loop to stop.
If the user is suficiently wise, they will conceive that luaver does something intrusive to handle errors.
The current error handling proccess is exitting the current shell without writing exit.
More concisely, luaver install 5.3.2 is exit 1 under certain conditions.

[umireon]

The behavior of kill -INT $$ in interactive sessions is different among shells and exit code cannot be controlled.
Even worse, zsh masks the error.

bash

$ kill -INT $$

$ echo $?
1

zsh

% kill -INT $$
% echo $?
0

dash

$ kill -INT $$

$ echo $?
130

ksh

$ kill -INT $$
$ echo $?
258

[DhavalKapil]

I get your point. This is indeed a minor yet important bug. An alternate way needs to be found out. I'm ignoring this for the present release though.