environment-to-ini tracking bug

[n9]

Description

There are known issues with environment-to-ini: https://gitea.com/gitea/helm-chart/issues/356

However, I have not found an issue this repo, where the implementation of environment-to-ini is located.

Gitea Version

1.20.5

Can you reproduce the bug on the Gitea demo site?

No

Log Gist

No response

Screenshots

No response

Git Version

No response

Operating System

No response

How are you running Gitea?

Currently, the issue only occurs when environment-to-ini is explicitly used.

However, @silverwind mentioned in #24804 (comment) that there is an intention to merge enviornment-to-ini into the codebase. (I am not sure, whether this may cause then also issues in other deployments.)

Database

None

Further notes

In my case, the issue occurs during the configuration of "custom logger modes":

Failed to load writer mode "router" for logger router: invalid log writer type (mode): router, maybe it needs something like 'MODE=file' in [log.router] section
Failed to load writer mode "xorm" for logger xorm: invalid log writer type (mode): xorm, maybe it needs something like 'MODE=file' in [log.xorm] section

This is because environment-to-ini is unable to write [log.router] MODE=console to app.ini, as it is a default value.

(In this case, however, I am not sure what should be fixed, whether the environment-to-ini or "custom logger modes".)

[wxiaoguang]

It was designed so. environment-to-ini only adds/updates, never deletes.

[n9]

@wxiaoguang This is not about deletes, but is about "adds".

[n9]

Try to run environment-to-ini with e.g.

GITEA__log_0X2E_router__MODE=console
GITEA__log_0X2E_xorm__MODE=console

These values will not be added to app.ini if the [log] has MODE already set to console in the app.ini file.

[n9]

I have added a failing test to demonstrate the issue: n9@c708961

[wxiaoguang]

It's the legacy/unmaintained INI package's bug (or quirk behavior).

The section values are inherited. So [sec].key=SOME also means [sec.sub].key=SOME

[n9]

Yes, but this does not work in loadLogModeByName and fails here:

gitea/modules/setting/log.go

Line 168 in 4126aad

return "", "", writerMode, fmt.Errorf("invalid log writer type (mode): %s, maybe it needs something like 'MODE=file' in [log.%s] section", writerType, modeName)

As mode is resolved using ConfigSectionKeyString:

gitea/modules/setting/log.go

Line 116 in 4126aad

writerType = ConfigSectionKeyString(sec, "MODE")

which calls ConfigSectionString that is not using inheritance:

gitea/modules/setting/config_provider.go

Lines 118 to 130 in 4126aad

	// ConfigInheritedKey works like ini.Section.Key(), but it always returns a new key instance, it is O(n) because NewKey is O(n)
	// and the returned key is safe to be used with "MustXxx", it doesn't change the parent's values.
	// Otherwise, ini.Section.Key().MustXxx would pollute the parent section's keys.
	// It never returns nil.
	func ConfigInheritedKey(sec ConfigSection, key string) ConfigKey {
	k := sec.Key(key)
	if k != nil && k.String() != "" {
	newKey, _ := sec.NewKey(k.Name(), k.String())
	return newKey
	}
	newKey, _ := sec.NewKey(key, "")
	return newKey
	}

[wxiaoguang]

Thank you for your detailed report.

I think the fix could be this: Fix environment-to-ini inherited key bug #27543

[n9]

Thank you for the fix.