On Thu, 2013-06-20 at 21:19 -0400, Rahul Sundaram wrote: > On 06/20/2013 09:04 PM, Adam Williamson wrote: > > > > Frankly I'm not 100% convinced that it's a good idea to try and guess > > language even from a fairly reliable geolocation mechanism like ours. > > It's still not 100% reliable - people behind corporate VPNs can be > > routed through just about anywhere - and even when it gives a correct > > geographic location, what the hell do you do for, say, India? Where > > you'd need to have house-level accuracy to have a snowball's chance in > > hell of guessing what language someone speaks? > It is not that bad despite the recognized thousands of languages, only > about 30 or so are considered major and less than that would have meet > the necessary L10N criteria anyway. Most speak in India speak several > different languages and you can guess the primary language spoken in any > given area quite reliably since the states boundaries were drawn by the > major language spoken in that area. Speaking as someone who has done > L10N work before, I would assume a number of users in India who install > Fedora would want the installation in English and not in their native > language since technical terms often have translations that local > speakers themselves wouldn't recognize. If the installer lets one pick > English quickly, it is not a problem if the default guess is wrong. To me that's just more illustration of the problem. Do things look better to an Indian who just wants to install in English anyway if we: a) default to English? b) default to an Indian language they don't speak at all? Seem pretty clear-cut to me. We have Gujarati, Hindi, Kannada, Malayalam, Marathi, Oriya, Tamil and Telugu in the installer; I've no idea which we default to for "India" (or if we just wave a white flag and go for English), or if langtable actually tries to distinguish by province at all, but I'm kind of pessimistic about our likely success rate in this mechanism. (We also have Bengali for Bangladesh and Punjabi and Urdu for Pakistan; I am going to go out on a limb and guess we really wouldn't want to get those mixed up in any way). For other 'fraught' geopolitical areas we have Traditional and Simplified Chinese, Japanese, and Korean all in the installer - none of which you'd particular want to get wrong. We have Hebrew (Israel); I sure hope our geolocation is right on point there. We have Catalan and 'Spanish' (which some prefer to call 'Castilian'...) both associated with Spain. Oh look! We also have Basque, just to complete the set. We've got Bosnian and Serbian. I haven't looked at langtable to see what actual associations anaconda is going to wind up drawing from the geoloc data, I'm just highlighting language sets in the list which you really wouldn't want to get mixed up via geolocation...yes, this stuff really does get people mad if you get it wrong. Defaulting to English seems rather safer. I suppose, for testing purposes, it'd be rather handy to be able to feed arbitrary results or source data - fake an IP address, etc - into the geoloc processing mechanism via a kernel parameter or kickstart or something... -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | identi.ca: adamwfedora http://www.happyassassin.net -- test mailing list test@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe: https://admin.fedoraproject.org/mailman/listinfo/test
