flag_or_value_option type: [TrueClass, String] is too narrow for action-option-with-optional-value commands

[jcouball]

Problem

flag_or_value_option accepts any object with a meaningful #to_s implementation by default — String, Integer, Float, Pathname, Symbol, etc. — and stringifies it automatically during the build phase. However, when type: [TrueClass, String] is added (as required for action-option-with-optional-value commands to prevent false from silently emitting nothing), it over-constrains the input by rejecting valid #to_s-capable values like Symbol.

Example from Git::Commands::Am::ShowCurrentPatch:

flag_or_value_option :show_current_patch, inline: true, type: [TrueClass, String]

This correctly prevents false from passing through silently, but rejects:

• cmd.call(:diff) — Symbol is not in [TrueClass, String], even though :diff.to_s produces the valid CLI argument "diff"
• cmd.call(Pathname.new('diff')) — similarly rejected despite producing correct output

The same pattern exists in Git::Commands::Push (type: [String, FalseClass] on recurse_submodules).

Root cause

The type: modifier validates the Ruby class of the input before #to_s stringification. For action-option-with-optional-value commands, the real need is narrower: reject false (which would silently emit nothing) while allowing anything that stringifies correctly.

Possible solutions

1. Add a reject: modifier — e.g. reject: [FalseClass] — that blacklists specific types instead of whitelisting. The type: constraint would no longer be needed for this pattern.

2. Widen the type list — e.g. type: [TrueClass, String, Symbol] — but this is fragile and doesn't cover all #to_s-capable types.

3. Introduce a default false-rejection for flag_or_value_option — if false is never a valid input for flag_or_value_option, the DSL method could reject it automatically without requiring type:.

Option 3 is probably the cleanest: flag_or_value_option semantically means "flag or value" — false (meaning "don't emit the flag") is arguably not a valid state for this method. If callers don't want the flag, they should omit the keyword entirely (or pass nil).

Affected commands

• lib/git/commands/am/show_current_patch.rb — type: [TrueClass, String]
• lib/git/commands/push.rb — type: [String, FalseClass] on recurse_submodules
• lib/git/commands/commit.rb — type: String on date (less impactful but also too narrow)