Fix the implementation of the run helper function to properly parse lines of output

[dimo414]

The existing behavior is broken, as it relies on the shell to word-split unquoted variables. This breaks down in several ways, notably that empty lines are silently dropped, but also because words-splitting also triggers globbing, so any line that happens to match a shell glob will be expanded in the resulting array.

This is a breaking change, but the existing behavior is fundamentally broken and does not work as documented.

Fixes #281 and #284

• I have reviewed the Contributor Guidelines.
• I have reviewed the Code of Conduct and agree to abide by it

[sublimino]

Thanks @dimo414, give me some time to parse and test this. As it's a breaking change we may want to release 1.4.0 first too, but will get that done sooner rather than later (haven't build a roadmap yet).

[dimo414]

Another option, as mentioned in #249, would be to move this behavior to a new function and deprecate run. That would avoid the backwards-incompatibility concerns and allow us to create a more robust utility.

[martin-schulze-vireso]

Closing this in favor of #158

In fact, I am using both. Your code can preserve empty lines while the other can't. However, when we are not interested in that, the other one is about 20x more performant, which becomes noticable on large inputs.