On 2021-06-07 16:38:16+0200, Torsten Bögershausen <tboegi@xxxxxx> wrote: > On Mon, Jun 07, 2021 at 10:01:12AM +0900, Junio C Hamano wrote: > > Đoàn Trần Công Danh <congdanhqx@xxxxxxxxx> writes: > > > > > Subject: Re: [PATCH v3] t: use pre-defined utf-8 locale for testing svn > > > > That certainly is better than "user-specific", but a lot worse than > > "user-specified". "pre-defined" leaves it open to question "who > > defines it?", and it is a reasonable interpretation that the locale > > may be hardcoded in the makefile, but that is not what this patch > > does. Saying "user-specified" would not have such a problem. > > > > >> I kind of wonder if trying "C.UTF-8" would be a reasonable fallback so > > >> that people don't even have to set this extra Makefile knob. But I'm not > > >> sure if we have a good way of testing if that locale works (if we can't > > >> find the "locale" binary). > > > > > > I also think we should fallback to "C.UTF-8" instead of not testing those > > > tests. However, I don't know if there're any systems that not have "C.UTF-8" > > > locale. > > > > I do share the feeling, but have a hunch that systems lacking > > "locale -a" may be either superset of, or has at least large overlap > > with, those lacking "C.UTF-8", and the new mechanism introduced here > > will cover both of them, so I think it is OK to stop here, at least > > for now. > > > > Thanks. > > I did a little bit of digging, the first (?) usage of locale was probably > introduced in the i18n patch to use gettext and msgfmt and stuff: > > commit 5e9637c629702e3d41ad01d95956d1835d7338e0 > Author: Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx> > Date: Fri Nov 18 00:14:42 2011 +0100 > > i18n: add infrastructure for translating Git with gettext > > -------------------------- > > I can't get rid of the feeling, that most systems have an UTF-8 > locale enabled by default these days. > > What does the following give us under Linux musl ? > set | grep UTF It's empty. OK, I lied. Distributions include a file in /etc/profile.d to enable LC_ALL/LANG. Without that file, it's empty. However, such use-case is covered by our fallback to LC_ALL and LANG. Technically, Linux with musl always knows those locales C, POSIX, C.UTF-8 I worried more about other systems instead. > MacOs (my box here) has this: > > LANG=en_US.UTF-8 > LC_ALL=en_US.UTF-8 > LC_CTYPE=UTF-8 > XTERM_LOCALE=en_US.UTF-8 > > If, and only if, Linux musl has a similar setup, > then we may be able to skip the Makefile knob for the moment. Yes, this patch allows to skip the Makefile knob. Because LC_ALL and LANG will be the first and second fallback, respectively. And we can skip a (potentially expensive) call to "locale -a". > Just skip locale -a, if locale is not available. > This is just a loose idea, I need to install the git-svn bindings and > test if the git-svn tests pass. -- Danh
