From: Jakub Narebski Date: Fri, 02 Mar 2012 14:56:12 GMT Subject: Re: [PATCH] Update l10n guide Message-ID: In-Reply-To: Jiang Xin writes: > 2012/3/2 Johannes Sixt : > > It does not convert, but it records which encoding the text has. If you > > don't specify anything, UTF-8 is assumed, and if your text is actually not > > UTF-8, the result is necessarily garbage. > > > > Then you haven't set i18n.commitEncoding. Try this: > > > >   git config i18n.commitEncoding CP936 > > I know there are two config variables. i18n.commitEncoding will insert > a "encoding XX" line to the commit object, while i18n.logOutputEncoding > will set the default output encoding. Note that according to documentation 'git commit' issues a warning if the commit log message given to it does not look like a valid UTF-8 string, unless you explicitly say your project uses a legacy encoding. Modern git would also warn if you have NUL ("\0") character in your commit message, e.g. when using UCS-2 / UTF-16 encoding. > But this implementation seems like a workaround. > > * Tree objects do not have such implementation, so multibyte characters > can not be used as filenames. And there is no place on pathnameEncoding in 'tree' object, unfortunately. One proposed solution was to convert filenames from filesystem encoding to normal-form composed UTF-8 when creating tree objects, but this would have to be optional. Anyway at least for source code using characters outside US-ASCII is really discouraged anyway. Note also that sample 'pre-commit' hook prevents adding files with non-ascii filenames. > * Commit object without "encoding" instruction will be used as it is. So > people under the same non-utf8 locale may not notice that they > have not set the proper i18n.commitEncoding, until one day they > need accross platform development. But see above. > * As raw commit log alway used as fallback, sometimes it is hard to find > commit objects with wrong encoding instructions. > > I think save commit object, tree object, packed-refs in UTF-8 is > a better implementation. Backward compatibiltiyt and performance. -- Jakub Narebski