Example robust to NULL osVersion

[MichaelChirico]

Recently ran into some packages having trouble with utils::osVersion being NULL (as is a documented possibility).

Now I am searching around GitHub a bit to see if I can't improve robustness in some other places too.

That led me to this change: if osVersion is NULL, startsWith() will return logical() and trip up if(). isTRUE() as done here is the simplest fix.

• This is the only usage of osVersion in the package
• I also note just above the condition exists("osVersion"). As osVersion was introduced in R 3.5.0, it wouldn't be so "weird" for it to be absent there 😉. I am not sure if you care about this dotted-dotted-line implicit R version dependency.

[eddelbuettel]

That's an interesting one! As you can imagine, installRub.r only works on Ubuntu so this check is relatively new and rare. Generalising it is easy so thanks for the PR.

[ Wasn't the "public word" that your in-house distro switched from Ubuntu-based to Debian-based so maybe you should stay away from Ubuntu binaries ... Anyway. ]

[ Also, I am happy to accommodate an endless stream of patches for your in-house special situation but at some point you guys may look into in-house firepower to take care of your needs. Just saying ... (as a share holder to boot ...) relying on unpaid external volunteers is a bit weird. ]

[MichaelChirico]

I didn't actually run into any issue from littler in this case. This one comes from searching CRAN & glancing at what I found:

https://github.com/search?q=author%3Amichaelchirico+is%3Apr+osVersion&type=issues

if you have a threshold below which you'd rather not be bothered with this type of janitorial PR, let me know & I can try and calibrate what to just let slide.

[eddelbuettel]

No I love'em when they are clean and focussed. So thank you for those, and thank you for indulging in my habit of maintaining a ChangeLog file.

That said, we're both economists by training and it's hard to let go of comparing marginal cost and marginal benefit. My cost is here is low, but was there a benefit?

[MichaelChirico]

There's an interesting discussion to be had there for sure. I know it passes my own cost/benefit test, especially being able to send the same fix to 9 packages in ~20 minutes (scaling up the very minimal benefit at very little marginal cost). I am cognizant of the overhead on review time though -- it's harder to know other maintainers' preferences at scale.