Note that if you remove the leading 'r' from the version, vercmp behaves like the exact opposite, which seems ok for me since I've never seen a version not starting with a number.

Many VCS packages use pkgver starting with a letter:

https://wiki.archlinux.org/index.php/VCS_package_guidelines

Could you explain why this is a meaningful/helpful distinction? A PKGBUILD without a pkgver will be rejected by makepkg. If you circumvent this, a package without a version will be rejected by pacman on installation.

Well, vercmp is useful in other cases, such as comparing sync/local versions (with pacman --print or expac) against the AUR RPC (with curl/json). If either string is empty, you'd have to be aware of the limitation, and workaround by e.g. setting -1 yourself if the variable is empty.

Note that vercmp also behaves inconsistently if you supply it a single argument:

% vercmp 123
1
% vercmp r123
-1

I'm not sure on the logic behind this (if any), but at least it should be documented.

I am confused by the references to calling vercmp with a single argument. I'm really not even sure why we allow that. vercmp is for comparing two version strings; calling it with just one doesn't make sense. So, I don't think its output with one argument is relevant here.

Treating the empty string as a "non-existent" version also doesn't make sense. Something that does not exist cannot be newer or older than something else. Alphabetic components of version strings are older than any non-alphabetic component, including the empty string. That's why 1.0alpha is older than 1.0 and why you see the difference between `vercmp 123` and `vercmp r123`.

In that case, I would suggest to:

1. Only allow calling vercmp with two arguments (3+ are also allowed with v5.0.1)

2. Add "Alphabetic components of version strings are older than any non-alphabetic component, including the empty string." to vercmp(8), clarifying the "Version comparison operates as follows:" section.

I thoroughly agree that vercmp should not allow anything other than exactly two arguments (unless we exit early after detecting --help as the first argument).

This should be really simple to fix, e.g. the attached patch. I'm not sure whether to provide an error message though -- and if so, do we need to drag in gettext o_O as well?
(That being said, we don't translate the current help text so probably not...)

Sent to pacman-dev.

https://lists.archlinux.org/pipermail/pacman-dev/2017-December/022211.html