toUpperCase on Android is incorrect for two-argument and default Greek and Turkish Locales

Question

I am having trouble with Greek and Turkish when using toUpperCase() with the default locale or more interestingly the two argument Locale constructor.

Issue happens on Galaxy Tab S2 Android 5.0.2 (also reproduced on 5.1.1)
The Issue is reproducible via BOTH the Settings App AND MoreLocale 2

Considering this value for title: Τέλος συνεδρίας

These calls work fine.

title.toUpperCase(new Locale("el_GR")) 
title.toUpperCase(new Locale("el-GR"))

Both generate the correct result. If you look closely there are tick marks after the T and the P.

ΤΈΛΟΣ ΣΥΝΕΔΡΊΑΣ

However, I get a different result for the default locale and the two-argument Locale constructor.

This is the default Locale on my tablet:

Locale.getDefault() ==  el_GR

Which is used in the generic toUpperCase()

public String toUpperCase() {
    return CaseMapper.toUpperCase(Locale.getDefault(), this, value, offset, count);
}

When I call this it returns the incorrect result.

title.toUpperCase()

Note the missing tick marks on the T and P.

ΤΕΛΟΣ ΣΥΝΕΔΡΙΑΣ

I get the same result if I use the two argument constructor for a new locale:

title.toUpperCase(new Locale("el","GR"))

ΤΕΛΟΣ ΣΥΝΕΔΡΙΑΣ

Adding these lines to application startup resolves the issues but is rather hackish.

String language = Locale.getDefault().getLanguage();
String country = Locale.getDefault().getCountry();
Locale.setDefault(new Locale(language + "-" + country));

I would prefer to simply defer to the default locale.

Why would there be ticks there? That doesn't seem right to me. — Sotirios Delimanolis, Aug 25 '16 at 17:06
Not a linguist but they somehow provide stress indications. See http://www.foundalis.com/lan/grstress.htm Turkish uses dots similarly. Because they are applied differently to capitals Turkish and Greek have unique case-conversion challenges. — amhest, Aug 25 '16 at 17:19
Oh, you're trying to uppercase `"τέλως συνεδρίασης"`. Sorry, I was trying this without the accents. — Sotirios Delimanolis, Aug 25 '16 at 17:36
As far as I know, though my Greek is rusty, capitalization should not contain the accents. — Sotirios Delimanolis, Aug 25 '16 at 17:42
See [this](http://stackoverflow.com/questions/24621265/converting-greek-to-uppercase-in-java) for example, though Android might behave differently. — Sotirios Delimanolis, Aug 25 '16 at 17:45
It seems that the single argument constructor probably sets a generic Greek rather than Greek (Greece) Generic Greek on Android seems to keep accents on upper case while Generic Greek on JDK 1.8 for windows and Greek (Greece) both keep the accents. Greek (Greece) on Android does not. — amhest, Aug 25 '16 at 19:53
you should not use `getDefault()` method, this method is for when you want to change your app's language based on changing the device's language. you should explicitly use `Locale` like the way you used at first. because you are focusing only on Greek language. — M D P, Aug 30 '16 at 09:45

score 4 · Accepted Answer · answered Aug 30 '16 at 16:33

As you can see in CaseMapper.java the method toUpperCase have a special behaviour for the Azeri, Lithuanian, Turkish and Greek locales:

public static String toUpperCase(Locale locale, String s, int count) {
    String languageCode = locale.getLanguage();
    if (languageCode.equals("tr") || languageCode.equals("az") || languageCode.equals("lt")) {
        return ICU.toUpperCase(s, locale);
    }
    if (languageCode.equals("el")) {
        return EL_UPPER.get().transliterate(s);
    }

    ...
}

Why you got different results depending on the type of Locale used?

Let's try to analyze the different kind of Locale created:

locale = new Locale("el_GR");
Log.d(TAG, String.format("[el_GR] Language: '%s' ~ Country: '%s'", locale.getLanguage(), locale.getCountry()));

locale = new Locale("el", "GR");
Log.d(TAG, String.format("[el, GR] Language: '%s' ~ Country: '%s'", locale.getLanguage(), locale.getCountry()));

locale = Locale.getDefault(); //Device with el_GR language
Log.d(TAG, String.format("[default] Language: '%s' ~ Country: '%s'", locale.getLanguage(), locale.getCountry()));

These are the results obtained:

[el_GR] Language: 'el_gr' ~ Country: ''
[el, GR] Language: 'el' ~ Country: 'GR'
[default] Language: 'el' ~ Country: 'GR'

As you can see in the manually created Locale with only one parameter the language is el_gr, for the other two cases instead the language is el. The toUpperCase has a different behaviour only when the language code is tr, az, lt or el, otherwise it behave normally. For this reason the output is different.

Unfortunately if you want the tick marks the only viable solution is to use the Locale constructor with only one parameter.

Almost perfect answer, but for the end: "the only viable solution is to use the Locale constructor with only one parameter". That is not correct. You can use new Locale("el", "GR",). Or Locale.forLanguageTag("el-GR") for API level ≥ 21. new Locale("el_GR") is not Greek. — Mihai Nita, Sep 02 '16 at 07:24
As explained in the answer if you use `new Locale("el", "GR",)` (and also `Locale.forLanguageTag("el-GR")` since it generate the same `Locale`) you don't get the tick marks as requested by the OP. The only method to obtain the tick marks is to use a `Locale` with a language different from `el`, so `toUpper` method will use the standard behaviour to uppercasing the string. — Mattia Maestrini, Sep 02 '16 at 07:53

toUpperCase on Android is incorrect for two-argument and default Greek and Turkish Locales

1 Answers1