Customize scheme 1

[jeancroy]

Basically I want to get rid of this step

4. The request contains an Accept-Language header with a language
    that matches (exactly or loosely) one of the application languages.

The context is "fr-ca" : a bunch of french speaking people in a sea of English speaking territory.
Official language of business is french.
A lot of computer system are in English. Like 50/50.
Among other things choosing English software help to search on web for settings/feature XYZ.

Basically the language of the os/browser is disconnected from user intent. Most of our target user don't have any idea about advanced settings to adjust asked language.

---

Then I realize Scheme1 without step4 is almost Scheme2 + cookie.
So I tried to implement cookies in DetermineDefaultLanguageFromRequest delegate.
Not complex but also a failure. This is because whatever the ouput of this is hidden as being the default.
I tough this was a bug until I see it documented:

The default URL Localization scheme (Scheme1) will show the language tag in the URL always; 
an alternative scheme, Scheme2, will show the language tag only if it is not the default.

---

So basically the feedback is stuff is glued together.
Url presentation is tied to language detection source.
Cookie language detection is tied to request language detection.

So it would be nice to have separate Boolean for behavior.

DetectLanguageFromCookie
DetectLanguageFromHeader
HideDefaultLanguageInUrl

Sheme1 could map as (true,true,false)
Scheme2 map as (false,false,true)

And after setting the scheme we could refine the behavior

[turquoiseowl]

Yes, I'm with your analysis 100%. It should have been developed like that in the first place TBH.

I'm a bit tied up with other stuff at the moment. Are you able to have a crack at it yourself?

[jeancroy]

I can give it a try. I do have one question.

Here it show GetRequestUserLanguages should choose between the request and the PAL

i18n/src/i18n/Helpers/HttpContextExtensions.cs

Lines 269 to 283 in d3b1983

	public static LanguageTag GetInferredLanguage(this System.Web.HttpContextBase context)
	{
	// langtag = best match between
	// 1. Inferred user languages (cookie and Accept-Language header)
	// 2. App Languages.
	LanguageTag lt = null;
	System.Web.HttpCookie cookie_langtag = context.Request.Cookies.Get("i18n.langtag");
	if (cookie_langtag != null) {
	lt = LanguageHelpers.GetMatchingAppLanguage(cookie_langtag.Value); }
	if (lt == null) {
	lt = LanguageHelpers.GetMatchingAppLanguage(context.GetRequestUserLanguages()); }
	if (lt == null) {
	throw new InvalidOperationException("Expected GetRequestUserLanguages to fall back to default language."); }
	return lt;
	}

But in the body there's this comment

i18n/src/i18n/Helpers/HttpContextExtensions.cs

Lines 201 to 205 in d3b1983

	// NB: originally we passed LocalizedApplication.Current.DefaultLanguageTag
	// here as the second parameter i.e. to specify the PAL. However, this was
	// found to be incorrect when operating i18n with EarlyUrlLocalization disabled,
	// as SetPrincipalAppLanguageForRequest was not being called, that normally
	// overwriting the PAL set erroneously here.

So i'm not sure how LocalizedApplication.Current.DefaultLanguageTag ever surface as the default choice.

[jeancroy]

Also would need some guidance on life cycle of those behavior boolean.

Some options I can think of:

1. Real Boolean. Resolve early. For exam listen to change of scheme.
2. Nullable. Resolve Late. behavior = manualOption ?? getDefautlForScheme(currentScheme) .
3. Don't bother about behavior boolean unless scheme is scheme.Custom.

Option 3 feel a bit like a patch, giving up the idea of base scheme + refine.
But it help ensure we don't introduce subtle behavior change in existing installation.

[turquoiseowl]

There are two sets of language tags: Application Languages and User Languages. The former relates to those supported by your webapp (what's available); the latter are those requested by the browser/user-agent (what's wanted).

The LocalizedApplication.Current.DefaultLanguageTag value is relevant in terms of ApplicationLanguages and is added to that set in addition to any locales/translations you add to your app. It isn't added to User Languages.

The job of i18n is then to make a best match between those two sets of languages.

[turquoiseowl]

WRT booleans, how about somehting like:

    [Flags]
    public enum UrlLocalizationScheme
    {
        _DetectLanguageFromCookieFlag = 0x01,
        _DetectLanguageFromHeaderFlag = 0x02,
        _HideDefaultLanguageInUrlFlag = 0x04,

        Void = 0,
        Scheme1 = _DetectLanguageFromCookieFlag | _DetectLanguageFromHeaderFlag,
        Scheme2 = _HideDefaultLanguageInUrlFlag,
        Scheme3 = 0,
    }

[turquoiseowl]

Thinking about this a bit more, are you sure Scheme2 excludes Accept-Language languages from consideration? IIRC the only thing it modifies is whether or not the langtag is shown in the URL when the default app language was current. That was its original purpose anyhow.

[turquoiseowl]

And some more: the UrlLocalization and TextLocalization are somewhat distinct parts of i18n. Perhaps the DetectLanguageFromHeaderFlag settings etc. are not strictly related to UrlLocalization and can best be implemented as a simple booleans in the app settings or static flag properties on the LocalizedApplication class.

[jeancroy]

Thinking about this a bit more, are you sure Scheme2 excludes Accept-Language languages from consideration?

My default browser send:
Accept-Language:en-US,en;q=0.8

The default language is
i18n.LocalizedApplication.Current.DefaultLanguage = "fr";

Using Scheme 2 french is applied,
Using Scheme1 english is applied.

So current behavior is scheme2 exclude Accept-Language.

---

As far as design perspective, I'm not sure it should be linked either. Scheme1/2 could control only display, _DetectLanguageFromCookieFlag , _DetectLanguageFromRequestFlag could be set independently

Then this package have tons of download on nugget so I'm not sure it's that good to change default

[turquoiseowl]

Yes, you're right: a side effect of Scheme2 is that a langtag will be derived from the URL, whether or not the URL contains a langtag. Therefore, the cookie and Accept-Language header never get a look in.

The simplest thing looks like two corresponding flags in the HttpContextExtensions.GetInferredLanguage method.

[turquoiseowl]

HttpContextExtensions.GetInferredLanguage then needs to fallback explicitly on LocalizedApplication.Current.DefaultLanguageTag if/when UserLanguages (i.e. headers) are disabled.

[Jandev]

It appears I'm stumbling across the mentioned issue also.
We also want/need to control if the Headers will be parsed for a language or not.

Can you enlighten us if there has been any progress in the development? If not I'll just have to create it myself :-)

[turquoiseowl]

I defer to @jeancroy on this.

[jeancroy]

It kinda depend how universal we want to fix this.
Do we agree on a minimum fix that only skip the header ?

[turquoiseowl]

Do we agree on a minimum fix that only skip the header ?

Yes. Single-purpose flag on LocalizedApplication maybe.

[Jandev]

Sounds good! I'll discuss internally if we can wait for this feature or need to develop something similar in the meantime.

[Jandev]

I've forked the project and made a small change concerning the Accept-Language header by creating a delegate which a developer can override if necessary. Similar to the UrlLocalizer code.
Link: LasaulecBV@4d4e639

It's not based on the flags, like discussed over here. For now our team can work with this change I made. If you want I can create a PR (with comments and stuff added of course), but perhaps the flags might be a bit easier (for now).

[turquoiseowl]

Even better than a flag! I would be very pleased to receive a PR.

If you've time, a note on this feature in the README would be cool.

[Jandev]

Sure, no problem. I'll get to it tomorrow or the day after.
In my current branch I've also upgraded the project to VS2015. Is this a problem?
Otherwise I can make the changes by not converting it to VS2015 if necessary.

[turquoiseowl]

Yes, if we can keep the project open to VS2013 users for now that would be great. Much appreciated.