Bazaar does not support umlauts in file names on Mac OS X

[Martin (gzlist)]

On 28/11/2009, John Arbash Meinel <john at arbash-meinel.com> wrote:
>
> So in the beginning... we did translate to NFC internally from the Mac
> filesystem. With the idea that most other platforms should only create
> NFC, and thus it was obvious what you would end up with.
>
> Except it turned out that Japanese windows creates wide () characters,
> which turn out not to be NFC. (So the expectation that internally we
> would always use NFC means that you need to translate on more than just
> Mac...)

This sounds wrong, the fullwidth ascii variants are only a problem
with NFKC, they're untouched by NFC.

Isn't the real problem a clash of user expectations? People don't like
filenames being unexpectedly changed, and they also don't like having
the set of valid names reduced. Canonicalisation fixes the mac
problem, but makes unix people unhappy because they can't have a files
named both 'e\xcc\x81' and '\xc3\xa9' (they're still complaining about
not being allowed backslashes), let alone being told a file had been
renamed from one thing to something else visually identical.

An opt-in "only use sane filenames" policy would be nice however.
There was a problem the other week on IRC where a user couldn't branch
a particular repo on windows 7, though it seemed to work fine on xp.
Turned out it was using one of the dos magic names, and actually the
xp repo was being silently broken.

Martin