Honor env sources on duplicated flags

[ibobgunardi]

What type of PR is this?

• bug

What this PR does / why we need it:

• Preserves env-source values when the same persistent flag instance is registered on an ancestor command and a subcommand.
• Skips re-preparing a shared persistent flag instance on descendant commands so the value loaded during the parent parse is not reset.
• Adds regression coverage for duplicated persistent flag env sources.

Which issue(s) this PR fixes:

Fixes #2217

Special notes for your reviewer:

The main behavior to review is the exact flag-instance check for inherited persistent flags.

Testing

• go test ./... -run TestDuplicatePersistentFlagUsesEnvSource -count=1 -v
• go test ./... -run "TestPersistentFlag|TestDuplicatePersistentFlagUsesEnvSource|TestRequiredPersistentFlag|TestCommand_IsSet_fromEnv|TestCheckRequiredFlags" -count=1 -v
• go test ./... -count=1

Release Notes

Fixed env-source values being reset when the same persistent flag instance is registered on a parent command and subcommand.

[dearchap]

@ibobgunardi Can you follow the PR template when submitting a PR ? Thanks

[dearchap]

Root Cause

When the same flag instance is registered on both a parent and subcommand, FlagBase.PreParse() re-initializes f.value to the default (f.Value) without checking f.applied. This overwrites the env-sourced value set by the parents PostParse()`.

Meanwhile, FlagBase.PostParse() guards with f.hasBeenSet, which is already true (set by the parent`s PostParse), so it skips re-reading the env var. End result: the subcommand gets the zero/default value instead of the env value.

Fix Validation

hasPersistentFlagOnAncestor() uses pointer equality to detect the shared instance, and the PreParse skip in command_run.go:144-146 correctly:

1. Prevents the destructive re-initialization on the subcommand
2. Preserves the env (or CLI-arg) value already set on the parent
3. PostParse respects hasBeenSet and does not overwrite

Test Coverage

The test TestDuplicatePersistentFlagUsesEnvSource covers:

• Env source read by the parentsBefore` handler ✓
• Env source value propagated to the subcommands Action` ✓
• IsSet() returning true on both parent and subcommand ✓

LGTM.

[ibobgunardi]

Updated the description to follow the PR template. Thanks for the reminder.

Undoing approval - user requested

[dearchap]

Good question about f.applied in PreParse. I tested it — two regressions:

1. TestSliceStringFlagParsing — PreParse re-creates f.value from scratch (via f.creator.Create). Without it, Set() appends to the existing slice, causing cross-sub-test value leakage. e.g. expected ["x", "y"], got ["x", "x", "y"].

2. TestRequiredFlagDelayed — applied being true on a shared flag instance prevents PreParse from running on the leaf command, altering required-flag behavior that non-shared flags depend on.

applied serves dual roles (prevent recursion + track initialization), so adding it to PreParse is too broad. The pointer-equality/ancestor check is correctly scoped. LGTM.