Missing plural keys wrong locale displayed, does not respect ignore list

[Gargron]

This is my mistake. For some reason, a missing plural in "it" shows up as missing in "zh-TW", and before that, missing plurals in "cs" also showed up as "zh-TW". I have no idea why, must be some key getting overwritten somewhere.

Also, I discovered I have a weird case where I have a structure like:

platforms:
  other: unknown platform

And unfortunately it shows up as missing "one" even though it's not a plural. Maybe when deciding based on only "other" being present, a condition should be that it contains the "count" variable as well?

[glebm]

For some reason, a missing plural in "it" shows up as missing in "zh-TW", and before that, missing plurals in "cs" also showed up as "zh-TW". I have no idea why, must be some key getting overwritten somewhere.

If you send a PR with a failing test, I'll look into this one.

[glebm]

And unfortunately it shows up as missing "one" even though it's not a plural. Maybe when deciding based on only "other" being present, a condition should be that it contains the "count" variable as well?

Looks like platforms has other children:

https://github.com/tootsuite/mastodon/blob/82e7988afcde5b19b99ad9ecf7973560a8a17f7f/config/locales/en.yml#L727-L739

It shouldn't be detected as a plural key at all. plural_forms? here will return false:

i18n-tasks/lib/i18n/tasks/missing_keys.rb

Line 68 in efcc22c

next if !plural_forms?(children) || present_keys.superset?(required_keys)

Not sure what's happening here but a failing test would help debug the issue.

[glebm]

One issue I see is here:

i18n-tasks/lib/i18n/tasks/missing_keys.rb

Line 69 in efcc22c

node.value = children.to_hash

This changes the node, that shouldn't happen (we should create a new by calling derive). I'll try fixing this one, maybe it'll fix the other issues.

[glebm]

@Gargron Does the commit above fix any issues?

[Gargron]

Looks like platforms has other children:

In en it does, but because it's things like "firefox" they don't need to be translated into other languages, so in cy it would only have the other child

[glebm]

Oh I see. Well, it's a sensible heuristic to add: if a key has only one child and that child does not have the count interpolation, then it's not a plural key.

[Gargron]

This is a CI run from last night (without your fix yet): https://circleci.com/gh/tootsuite/mastodon/26761

Missing translations (1) | i18n-tasks v0.9.27
+--------+--------------------+----------------------------------+
| Locale | Key                | Value in other locales or source |
+--------+--------------------+----------------------------------+
| zh-TW  | sessions.platforms | one                              |
+--------+--------------------+----------------------------------+

If run with -f inspect, I can see it actually refers to the it translation, not zh-TW as it says. I checked all locale files and they all have the correct root key, so it's not a case of it.yml overwriting zh-TW or anything like that. Perhaps your patch already fixes it though, I'll need to test it