There are a few remaining tasks that I'll be working on in the next few days before merging this PR, that includes:

• Open a PR for react-i18next repo supporting the new types.
• Open a PR for next-i18next repo supporting the new types.
• Create new tests and update current ones.
• Update i18next documentation.
• Include benchmark results here.

After merging this PR:

• Update examples.

(this comment is somewhat tangential to this PR, but I think being aware of this can help shape this new typing approach)

Hi @pedrodurek,

earlier this year, I tried to come up with some type fixes for getFixedT() in #1898.

Maybe we can incorporate some of those ideas into your new typing approach. Also, that PR had some new test cases that might make sense to incorporate here/add later.

getFixedT()

To me, it seems the types for getFixedT() should work somewhat differently to the useTranslation() ones:

Adapting your first example from the PR description, getFixedT(null, ['ns1', 'ns2']) should allow keys from both namespaces both without a ns prefix and with a prefix, i.e. 'key1' | 'key2' | 'key3' | 'key4' | 'ns1:key1' | 'ns1:key2' | 'ns2:key3' | 'ns2:key4'.

getFixedT(null, ['ns1']) should allow 'key1' | 'key2' | 'ns1:key1' | 'ns1:key2'. This seems to work fine with this PR.

getFixedT(null, 'ns1') (with a string instead of an array) should allow the same keys, but with this PR only allows 'key1' | 'key2'.
To change this, one could change the ParseKeysByNamespaces type to this:

type $Flatten<A> = A extends ReadonlyArray<infer E> ? E : A;
type ParseKeysByNamespaces<
  Ns extends Namespace,
  Keys,
  UnionNsps = $Flatten<Ns>,
> = UnionNsps extends keyof Keys ? AppendNamespace<UnionNsps, Keys[UnionNsps]> : never;

As outlined in #1889 (the bug corresponding to the PR mentioned above), getFixedT() - like i18next.t() - should allow the keys of all other namespaces available in i18next with the respective namespace prefix - even if they are not mentioned via the namespace parameter to getFixedT().

So in fact getFixedT(null, ['ns1']) and getFixedT(null, 'ns1') should allow 'key1' | 'key2' | 'ns1:key1' | 'ns1:key2' | 'ns2:key3' | 'ns2:key4'. This is quite different to how useTranslation() seems to work.

Lots of TS key suggestions

These are the same keys that are valid for your i18next.t() example. In the real world, there are much more translation keys. Also, the non-namespaced keys don't usually happen to be alphabetically before the namespaced keys like in the example. Due to this, TS will suggest a lot of keys and will happily mix ns-prefixed and unprefixed keys. I think this can be quite a mess to look at, making the key suggestions less helpful.

To cope with that I came up with the (hack-ish) type DisableSuggestion (see https://github.com/kkuegler/i18next/blob/18c6274451d133f8df98c91b53b8b98ba4db15d5/index.d.ts#L10). This can allow some keys, but keep them out of the TS suggestion list. I hoped to find more "official" ways to do this, but couldn't find any.

The end :)

I'm not sure if you can incorporate some of these aspects into this PR or not, but in the long run I think we should address these, so the runtime behavior and the types for getFixedT() match.

Perhaps we should make clear what useTranslation is...it's only reason to exist is to assert the namespaces provided are loaded (nothing more) before rendering (and is only needed in combination with using backends).

the getFixedT is just a helper to get the t function, but not bound to the default namespace and bound to namespace(s) passed in (bound means just not needing the namespace prefix ns1:key).

t function itself does not care about translations being loaded or not. It just looks up translations based on given namespace and key (and does this by using default namespace or given namespace as prefix to the key)

Under the hood useTranslation calls the getFixedT function and returns the provided bound t function (beside other stuff). But the important thing is calling loadNamespaces and throwing the promise if not yet loaded - and triggering Suspense (or returning the ready flag with value false)

This looks like a fantastic improvement.
We're using these updated types at work, and have some feedback. I know it's a W.I.P, but could be useful nonetheless.

• The types are faster, but still very slow when we have many keys. We have about 1500 keys spread out in ~60 namespaces in our project, and strictly typing them makes the tsc compilation time go from ~120s to ~250s.

• Autocomplete of the interpolated values is cool, but seems to break when using multiple namespaces and accessing the key without the namespace prefix. It could be very useful be able to enforce that you can't use interpolated values that are not defined in the types.

• Typescript doesn't seem to be able to keep track of the string references, so we can't "go to definition", and "go to references". It would be fantastic if there was some way to accomplish this, as it would make finding unused translations very easy.

Thanks for the thourough description of what the PR does @pedrodurek ! I am eager to see this merged, and I am specially interested in typed interpolation params.

Until this is merged, I have a quick question on how useTranslation would be better used:

With i18next@22.4.11 and react-i18next@12.2.0, and having resources defined as

const ns1 = {
  key1: '',
}

const ns2 = {
  key2: '',
}

const resources = {
  en: {
    ns1,
    ns2,
  },
}

Depending if I call useTranslation(nsArray) or useTranslation<NSType>() I get different allowed keys for t.

With the first one:

  const { t } = useTranslation(['ns1', 'ns2'])
  t('key1') // no error
  t('ns1:key1') // no error
  t('key2') // TSC Error: No overload matches
  t('ns2:key2') // no error

While the later:

  const { t } = useTranslation<'ns1'|'ns2'>()
  t('key1') // no error
  t('ns1:key1') // TSC Error: No overload matches
  t('key2') // no error
  t('ns2:key2') // TSC Error: No overload matches

As I understand, using one or the other might be subject to preference or use cases, but curious if there is any reason to use one or the other or if the PR proposal is to deprecated one of these.

Thanks!

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

Can this resolve #1894?

Benchmark results in Mac M1:

Num keys: 11,429
Num namespaces: 55
Compilation time with the types: ~6.27s
Compilation time without the types: ~1.49s
VSCode autocomplete keys (without cache): ~2.477s

I won’t share the results of the current types because it doesn’t handle these many keys/namespaces. 😅

I’ll share the project that I used for the benchmark once we merge this PR.

I do believe that I still can improve the compilation in 30-40%, but I'll work on it in another PR.

Hey @falsandtru, I'm not sure, but we can work something out once it's merged

I see

We found another issue in the new types.
When the namespace can be one of several ones, it doesn't complain if we access a key that only exists on one of them, but not the other.

In the above example, choose-amount.heading is available only on deposit, but we get no error.

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

fyi: ts tests are failing: https://app.circleci.com/pipelines/github/i18next/i18next/1104/workflows/55564641-6a5f-462f-af39-139d8e209ba8/jobs/908/parallel-runs/0/steps/0-105

fyi: goal is to release a new major version with #1885 and #1945 and also addressing some other stuff: #1884
but first we need to address:
#1885 (comment)
#1911 (review)

Do you mean this?

@pedrodurek I mean because of this

@pedrodurek now all but this is ready to release the major version...
Can you check the example and describe how the new typesafe translation approach should be set up in a project?

@pedrodurek does this example also need to be updated? https://github.com/i18next/i18next/tree/master/examples/typescript it seems to not work with the new index.d.ts

seems the compilerOptions needs to be strict: b7075f2#diff-603812bc45f8ec4c682a4be24c9423d325e92c7ea7f10afe1bd84b6c46077164R10

stuff to check before releasing this:

• why is this interpolation check working in tests, but not in example
• why is strict mode necessary for this to work? is there a workaround?

• why is this interpolation check working in tests, but not in example

seems because typeof here is just using string type for the translation values, so there is no info of the interpolation anymore...

@pedrodurek a possible workaround:

• why is strict mode necessary for this to work? is there a workaround?

"strictNullChecks": true is sufficient

Hi guys. This update reintroduced an issue I helped fix in the previous version:
For details see here: #1867

Essentially, when the t function receives a default value (either as a string, or as part of the defaultValue prop in the options), typescript should not enforce "known keys only".

That's the whole point of the default value option. If we're forcing people to always have a value for every key in their main namespace, then this entire option is completely useless, as it will always be overwritten anyway.

The same is true by the way for the Trans component of react-i18next, but this is probably less of your concern.

Hi guys. This update reintroduced an Issue I had fixed in the previous version: For details see here: #1867

Essentially, when the t function receives a default value (either as a string, or as part of the defaultValue prop in the options), typescript should not enforce "known keys only".

That's the whole point of the default value option. If we're forcing people to always have in their main namespace, then this entire option is completely useless, as it will always be overwritten anyway.

The same is true by the way for the Trans component of react-i18next, but this is probably less of your concern.

@pedrodurek what do you think?

addendum

My gut instinct would be to solve this via overloads:

/**************************
 * T function declaration *
 **************************/
export interface TFunction<Ns extends Namespace = _DefaultNamespace, KPrefix = undefined> {
  // known keys
  <
    Key extends ParseKeys<Ns, TOpt, KPrefix> | TemplateStringsArray,
    TOpt extends TOptions,
    Ret extends TFunctionReturn<Ns, AppendKeyPrefix<Key, KPrefix>, TOpt>,
  >(
    ...args:
      | [key: Key | Key[], options?: TOpt & InterpolationMap<Ret>]
      | [key: Key | Key[], defaultValue: string, options?: TOpt & InterpolationMap<Ret>]
  ): TFunctionReturnOptionalDetails<Ret, TOpt>;
  // unknown keys
  <
    TOpt extends TOptions,
    Ret extends TFunctionReturn<Ns, AppendKeyPrefix<Key, KPrefix>, TOpt>,
  >(
    ...args:
      | [key: string | string[], options?: TOpt & InterpolationMap<Ret> & { defaultValue: string }]
      | [key: string | string[], defaultValue: string, options?: TOpt & InterpolationMap<Ret>]
  ): TFunctionReturnOptionalDetails<Ret, TOpt>;
}

But you likely had a reason for eliminating them (probably speed?).
I haven't tested or benchmarked the above, but it might help as a starting point.

@janpaepke can you try with v23.2.6 ?

I did and it works. 👍
Thanks for your quick reaction!

It let's me set any string as the key, when a default value is provided and complains if there's not.
Is it possible to implement a similar solution for the Trans component? (default value is children in that case).
Or should I post this in a different repo?

There's another thing I noticed, but I am not entirely sure if it's related to my environment (IDE/ts version).

Key-Autocompletion for the t function doesn't seem to work for me any more.
Interestingly it does still work for the Trans Component.
And yes, this was also the case without this latest change.

Should I create a separate issue for that?

@janpaepke intellisense is tracked here: microsoft/TypeScript#54708 (probably fixed in next ts version and/or ide version)
Regarding Trans component, yes open a new issue in the react-i18next repo. btw: alternatively, just omit the i18nKey for the Trans component.

@adrai thanks a lot!
Trans Component -> will add an issue.
re your workaround: How would leaving out the key work? It couldn't relate the respective translation vars, if available, right?
Plus i18next-parser can't extract it anymore... Sorry, if off topic.

In a React Component, when I'm loading a namespace foo (not configured to be the default ns !), I got an TS error by not specifying the namespace in the t function :

const { t } = useTranslation('foo');
...
<div>
  // TS error
  {t('key')}
  // No TS error
  {t('foo:key')}
</div>

The TS error says Argument of type '["key"]' is not assignable to parameter of type '[key: TemplateStringsArray |...... I thought it was possible, not to specify the namespace but seems, it's requiered with v23.

Should you update this documentation of loading namespace in react-i18next ?

Thanks in advance for your answer

In a React Component, when I'm loading a namespace foo (not configured to be the default ns !), I got an TS error by not specifying the namespace in the t function :

const { t } = useTranslation('foo');
...
<div>
  // TS error
  {t('key')}
  // No TS error
  {t('foo:key')}
</div>

The TS error says Argument of type '["key"]' is not assignable to parameter of type '[key: TemplateStringsArray |...... I thought it was possible, not to specify the namespace but seems, it's requiered with v23.

Should you update this documentation of loading namespace in react-i18next ?

Thanks in advance for your answer

@MiryksV please provide a complete but minimal reproducible example repository or similar...

I couldn't reproduce on my sandbox, I'll take a deeper look in my project. Sry to bothering you for nothing.
Thanks for your answer.

I just adopted this, and the typescript compiler needs about 8.5GB ram now in order to type check our code.
It takes about 1.7min (with a few errors left to fix).

We use the interface generation, because we have our translations in .json files.
The interface file is 8888 lines long (I swear, I didn't make this up xD), with 26 namespaces.

Without the generated interfaces included, typescript needs about ~3s to run.

If you have any ideas on why it might take so long, I would really appreciate it!

Edit: after getting rid of all TS errors, it went down to 3s runtime, and about 400Mb ram.

Seems like, once I had a no-error run the tsc cache really speeds up the compiler.
Every time I bust the cache (manual deletion, or different branches) it will take up to ~2.5 minutes and over 8GB of ram.
After a successful run it will be in the low seconds again.

Hey, before i can use generic type to specify which type would return
i18n.t<string, string[]>('i18nkey', { returnObjects: true })

How can i do the same with new version, when passing third param ts complain that 'Type  string[]  does not satisfy the constraint  string'?
i18n.t<string, any, string[]>('i18nkey', { returnObjects: true })

I have the same issue as @reckter with very long tsc run times because of this, any way to solve the issue ?