Standard does not exactly specify allowed format for versioned tags

[embray]

I'm not sure if we always intended to require that tags have a version associated with them, more that it's a strong recommendation. Nevertheless, this is used pretty universally, and we even allow version wildcards in schemas à la asdf-format/asdf#838 which makes a little more formal the notion of tag versions (see #271).

But nowhere does the standard formalize the format <tag_name>-<version>.

We do specify that the version part should follow Semantic Versioning, but the standard does not actually formally specify the above <tag_name>-<version> format.

Making matters worse, we do say that a pre-release version suffix is allowed, and this is separated from the base version also by a -. The format of the pre-release version is fairly loose and can also contain -.

In effect, it's impossible to know how to split the version part of a tag name because versions can contain hyphens, but we don't actually specify that the tag base name cannot contain hyphens.

We should either:

• Disallow pre-release versions in tag versions (this seems to be rare/unheard of?) and state that the version always comes after the last hyphen in a tag name
• Disallow hyphens in tag names (this is already done by convention it seems, but it's not clear in the standard).

Not having clarity about this turns out to be a headache for implementers wishing to parse tag versions 😉

[braingram]

Thanks for opening this and adding details.

I'm not sure if we always intended to require that tags have a version associated with them, more that it's a strong recommendation. Nevertheless, this is used pretty universally, and we even allow version wildcards in schemas à la asdf-format/asdf#838 which makes a little more formal the notion of tag versions (see #271).

But nowhere does the standard formalize the format <tag_name>-<version>.

We do specify that the version part should follow Semantic Versioning, but the standard does not actually formally specify the above <tag_name>-<version> format.

Making matters worse, we do say that a pre-release version suffix is allowed, and this is separated from the base version also by a -. The format of the pre-release version is fairly loose and can also contain -.

I'm not aware of any pre-release tags.

In effect, it's impossible to know how to split the version part of a tag name because versions can contain hyphens, but we don't actually specify that the tag base name cannot contain hyphens.

We should either:

* Disallow pre-release versions in tag versions (this seems to be rare/unheard of?) and state that the version always comes after the last hyphen in a tag name

* Disallow hyphens in tag names (this is already done by convention it seems, but it's not clear in the standard).

Not having clarity about this turns out to be a headache for implementers wishing to parse tag versions 😉

To look at this in a different way... The previous versions of the specification had verbiage about handling version mismatches which treated minor/major/etc version mismatches differently. This section was changed so that the current version only converts exact matches (see this for more context). Put another way the current description of handling version mismatches is essentially, "treat different versions as different tags".

The wildcard matching is (in practice) also not specific to versions. Instead * matches any non / character and ** matches any.

>> asdf.util.uri_match("asdf://example.com/foo/bar*", "asdf://example.com/foo/barrow")
True
>> asdf.util.uri_match("asdf://example.com/foo/bar**", "asdf://example.com/foo/bar/bam")
True

As far as I can tell the wildcard matching isn't described anywhere in the specification. However wildcards are present in tag validators for some schemas so we should describe this somewhere in the specification.

So I guess one way to look at different tag versions is just that they are different tags. If in the end we require that implementations treat different versions the same as different tags, is there a benefit in formalizing the version format?

[embray]

@braingram Thanks for your comments. I have to think about this a bit more, but I think I understand the principle here--that "versions" in tag names aren't really special necessarily or required by the standard, and that the wildcard support is also independent of this. Seems reasonable actually.