.conflicts_with error messages not symmetric

[joshtriplett]

The documentation for conflicts_with says:

Conflict rules only need to be set for one of the two arguments, they do not need to be set for each.

NOTE: Defining a conflict is two-way, but does not need to defined for both arguments (i.e. if A conflicts with B, defining A.conflicts_with(B) is sufficient. You do not need need to also do B.conflicts_with(A))

However, the error messages for the conflict differ based on whether I provide one conflicts_with or both.

If I mark the arguments --rfc and --subject-prefix as each conflicting with each other, then passing both will give an error message that treats the second such argument as the error and suggest removing it, which makes sense:

(1) ~/src/git-series$ git series format --rfc --subject-prefix x
error: The argument '--subject-prefix <Subject-Prefix>' cannot be used with '--rfc'

USAGE:
    git series format --rfc

For more information try --help
(1) ~/src/git-series$ git series format --subject-prefix x --rfc 
error: The argument '--rfc' cannot be used with '--subject-prefix <Subject-Prefix>'

USAGE:
    git series format --subject-prefix <Subject-Prefix>

For more information try --help

However, if I only call .conflicts_with("subject-prefix") on --rfc, but not the other way around, the error message always marks --subject-prefix as the error and always suggests dropping --subject-prefix:

(1) ~/src/git-series$ git series format --rfc --subject-prefix x
error: The argument '--subject-prefix <Subject-Prefix>' cannot be used with '--rfc'

USAGE:
    git series format --rfc

For more information try --help
(1) ~/src/git-series$ git series format --subject-prefix x --rfc 
error: The argument '--subject-prefix <Subject-Prefix>' cannot be used with '--rfc'

USAGE:
    git series format --rfc

For more information try --help

[kbknapp]

Thanks for posting this. Yeah it has to do with the order in which clap detects the conflict, and then searches for the root cause of the error. I think it's because clap uses an early exit strategy, and returns an error as soon as it comes to it, instead of parsing the whole line.

For your first example, the way it works is:

• clap detects --rfc and adds --subject-prefix to list of known conflicts
• Then detects --subject-prefix and exits with a known conflict error

For the second example it goes a little differently:

• clap detects --rfc
• clap detects --subject-prefix and adds --rfc to a list of known conflicts`
• clap finishes parsing the rest of the arguments
• Prior to returning a success, it does a once over all the matches, and ensures there are no conflicts
• clap finds --rfc is on the known conflict list and searches for the a used argument that conflicts with it
• clap finds --subject-prefix and exits with a conflict error, but doesn't actually know which one came first

But since what you're posing makes 100% sense, I do need to look at a way to make this more symmetric and less confusing.

[joshtriplett]

@kbknapp You could store all conflicts symmetrically, manufacturing and adding to a conflict group. For instance, when you call conflicts_with from one arg to another, it'd look up whether that arg already has a conflict group, add itself to that conflict group if so, or create a new conflict group with both arguments if not.

[kbknapp]

Yeah, I'm going to look at some of those possibilities. I'll first have to get re-aquainted with the code since it's been a while since I looked at that piece 😜 I know some easy ways to do it, but I'd also like to keep from allocating more than I absolutely need to.

[kbknapp]

@joshtriplett I was able to get this working. Once #720 merges let me know if v2.16.4 is more along the lines of what you're thinking.

[joshtriplett]

Looks promising. I look forward to testing it.