Re: [LIKELY JUNK]Re: [egit-dev] Participation in Helios build?

["Shawn O. Pearce" <spearce@xxxxxxxxxxx>]

"Zivkov, Sasa" <sasa.zivkov@xxxxxxx> wrote:
> > I'd actually prefer to just write our own micro NLS class in JGit.

Chris Aniszczyk and I talked about this yesterday at the board
meeting during a down minute.  What we actually need isn't static
fields for translation, but instance fields we can inject based
on the current client, because JGit can often be used in a server
context where the translation of the error message might need to
differ per-request to suite the user's preference.

So we're thinking a pattern like:

	public class TransportText extends TranslationBundle {
		public static TransportText get() {
			return NLS.get(TransportText.class);
		}

		public String repositoryNotFound;
		public String transportError;
	}

	TransportText_en_US.properties:
		repositoryNotFound=repository {0} not found
		transportError=unknown error talking to {0}

	public abstract class TranslationBundle {
		public void load(String language) {
			... load bundle into instance members via reflection ...
		}
	}

	public class NLS {
		private static final InheritedThreadLocal<NLS> local = new InheritedThreadLocal<NLS>();
		public static <T extends TranslationBundle> T get(Class<T> type) {
			NLS self = local.get();
			if (self == null) {
				self = new NLS(... JVM default ...);
			}
			return self.get(type);
		}

		private Map<Class, TranslationBundle> map = new HashMap();

		public synchronized <T extends TranslationBundle> T get(Class<T> type) {
			TranslationBundle r = map.get(type);
			if (r == null) {
				r = type.newInstance();
				r.load(language);
				map.put(type, r);
			}
			return r;
		}
	}

And code can throw exceptions:

	throw new TransportException(
		uri,
		TransportText.get().transportError);

Client code in applications or in application servers would need
to set the NLS environment before starting JGit.

This might be an issue with the worker thread pool in Eclipse,
how do we ensure the workers get the NLS environment?
 
> Few questions:
> 1. How to decide if a text should or shouldn't be translated? Is it only error message as stated above?
>    I assume (j)git output should stay parsable and some texts mustn't be translated. Right?

IMHO, if the text is an english message that isn't a string required
for a file name or file content or network protocol keyword, we
should translate it.  That's *most* of the strings in the JGit code.

> 2. How to find, at runtime, the set of classes where to perform the injection of translated texts?
>    Is this already limited to particular packages only?

Pretty much all of them.  :-(
 
-- 
Shawn.