[pandoc_crossref] Compat change requires version bump

[omus]

Follow up to: #8892 (comment)

[omus]

One alternative I could see is to translate the pandoc-crossref version and include all of the 4th digit in the patch version:

• v0.3.17.1a -> v0.3.1710
• v0.3.16.0e -> v0.3.1604

I don't think the registry tests will like this though as the version number is non-incrementing. There may be a special case for JLLs though.

[omus]

Hold off on merging this until we settle on a convention

[omus]

I couldn't remember if we could use non-sequential version numbers with JLL packages and the answer is yes:

• https://github.com/JuliaRegistries/RegistryCI.jl/blob/e4bb3ad236f66b626eb6b4ec8bcaf2b3b3e0246d/src/AutoMerge/guidelines.jl#L1105
• https://github.com/JuliaRegistries/RegistryCI.jl/blob/e4bb3ad236f66b626eb6b4ec8bcaf2b3b3e0246d/src/AutoMerge/pull_requests.jl#L191-L192

[palday]

Looking at the version history of pandoc-crossref, it looks like it uses zerover with the schema 0.major.minor.patch[build] where build is a letter. So we could promote everything one level and and combine patch and build into a single entity: major.minor.julia_patch where julia_patch = patch * 100 + sequence_number(build).

Here are a few examples:

• 0.3.17.0 -> 3.17.0
• 0.3.17.0f -> 3.17.006 = 3.17.6
• 0.3.17.1 -> 3.17.100
• 0.3.17.1a -> 3.17.101

[omus]

Do we have evidence that pandoc-crossref is using 0ver? From this comment it seems like the package may be using Haskell PvP

I also found this Haskell release of pandoc-crossref which seems to drop the letter from the version: https://hackage.haskell.org/package/pandoc-crossref

[palday]

The letter is the build -- no real source change, it just reflects which version of pandoc it was compiled against, which is why for haskell sources it doesn't show up.

I think Haskell PvP is perhaps the better description. Haskell PvP is a lot like SemVer, but one key difference is that it has MAJOR.MAJOR.MINOR.PATCH versioning. It looks pandoc-crossref has locked the first MAJOR to 0, which is the core detail that I was assuming.

[palday]

To full incorporate the two MAJORs in PvP, the SemVer major version could be MAJOR[1]*100 + MAJOR[2] so that even if pandoc-cross ref increments the first major, the scheme still works.

[omus]

Some examples of the current JLL version created from the pandoc-crossref version:

julia> pandoc_crossref_jll_version("0.3.17.0")
v"3.17.0"

julia> pandoc_crossref_jll_version("0.3.17.0f")
v"3.17.6"

julia> pandoc_crossref_jll_version("0.3.17.1")
v"3.17.10"

julia> pandoc_crossref_jll_version("0.3.17.1a")
v"3.17.11"

julia> pandoc_crossref_jll_version("0.3.16.0")
v"3.16.0"

julia> pandoc_crossref_jll_version("0.3.16.0a")
v"3.16.1"

julia> pandoc_crossref_jll_version("1.3.4.5")
v"103.4.50"

[omus]

Updated examples:

julia> pandoc_crossref_jll_version("0.3.17.0")
v"3.17.0"

julia> pandoc_crossref_jll_version("0.3.17.0f")
v"3.17.6"

julia> pandoc_crossref_jll_version("0.3.17.1")
v"3.17.100"

julia> pandoc_crossref_jll_version("0.3.17.1a")
v"3.17.101"

julia> pandoc_crossref_jll_version("0.3.16.0")
v"3.16.0"

julia> pandoc_crossref_jll_version("0.3.16.0a")
v"3.16.1"

julia> pandoc_crossref_jll_version("1.3.4.5")
v"103.4.500"