aura misses some package updates

[mkoloberdin]

yaourt -Sua offers to upgrade the following packages among others:

==> Software upgrade (new version) :
aur/coin                  4.0.0a+11927+me74da184f75b-1 -> 4.0.0-1
aur/coin-docs             4.0.0a+11927+me74da184f75b-1 -> 4.0.0-1
aur/soqt                  1.6.0a+2014+m872b87e73dfb-1  -> 1.6.0-1

...while aura -Akxua misses these upgrades.

[fosskers]

It likely due to the version numbers. 4.0.0-1 is likely being considered "less than" 4.0.0xyz123.

[fosskers]

Those packages don't really conform to any versioning scheme, so it's hard to tell which of the two versions should be considered newer. Do you have thoughts on that? I wrote a library a few years ago (which Aura uses) that tackles this problem in general.

[mkoloberdin]

I've looked into how yaourt compares version numbers. Turns out it just uses an external utility that comes with pacman called vercmp:

$ vercmp
vercmp (pacman) v5.2.1

Compare package version numbers using pacman's version comparison logic.

Usage: vercmp <ver1> <ver2>

Output values:
  < 0 : if ver1 < ver2
    0 : if ver1 == ver2
  > 0 : if ver1 > ver2

$ vercmp 4.0.0-1 4.0.0a+11927+me74da184f75b-1
1

[mkoloberdin]

According to the comments in pacman source the comparison logic has been copied from rpm. So it's not even unique to Arch.

[fosskers]

I've managed to extend versions to account for this particular case, where a complex version value is close enough to SemVer to allow for more meaningful comparisons. The change will be reflected in Aura in the upcoming release.

[mkoloberdin]

Here is one more package/version update aura misses:

(part of yaourt's output):

aur/megatools                    1.11.0.git.20200404-1 -> 1.11.0+20200830-1

$ vercmp 1.11.0+20200830-1 1.11.0.git.20200404-1
1

[fosskers]

Hm, that should have been detected, since we expect the first to parse as a Version and the second to parse as a SemVer. Comparison between those two is generally well-defined. I'll add a test case to the versions library.

[fosskers]

Yup, versions thinks the "git" variant is earlier.

[fosskers]

My Rust port of the same library gives a bit more information:

Attempted: 1.11.0.git.20200404-1 < 1.11.0+20200830-1
Lesser: General(Version { epoch: None, chunks: Chunks([Chunk([Digits(1)]), Chunk([Digits(11)]), Chunk([Digits(0)]), Chunk([Letters("git")]), Chunk([Digits(20200404)])]), release: Some(Chunks([Chunk([Digits(1)])])) })
Greater: Complex(Mess { chunk: ["1", "11", "0"], next: Some((Plus, Mess { chunk: ["20200830"], next: Some((Hyphen, Mess { chunk: ["1"], next: None })) })) })

So it's likely that the 20200830 was thrown up against git (as a string, by the looks of it) and the letters took precedence. Quick tests show that "git" is considered greater, at least when comparing raw strings.

[fosskers]

The solution:

• 1.11.0+20200830-1 now actually parses as a Version, not a Mess.
• If "lettered" sections are encountered, they are considered "beta" or "prerelease". So here, 1.11.0 is considered greater than 1.11.0.git.whatever.

[mkoloberdin]

Now aura acts as if this package upgrade does not exist (yay output): aur/megatools 1.11.0+20210505-1 -> 1.11.0+20211030-1