On 02.09.2017 18:06, Stefan Fuhrmann wrote:
>
>
> On 02.09.2017 17:17, Branko Čibej wrote:
>> On 02.09.2017 17:12, stef...@apache.org wrote:
>>> +svn_boolean_t
>>> +svn_utf__fuzzy_glob_match(const char *str,
>>> + const apr_array_header_t *patterns,
>>> + svn_membuf_t *buf)
>>> +{
>>> + const char *normalized;
>>> + svn_error_t *err;
>>> + int i;
>>> +
>>> + /* Try to normalize case and accents in STR.
>>> + *
>>> + * If that should fail for some reason, continue with the
>>> original STR.
>>> + * There is still a fair chance that it matches "*.ext" pattern
>>> despite
>>> + * being "broken" UTF8. */
>>
>>
>> What evidence do you have for this statement? It is, quite frankly,
>> completely bonkers.
>>
>> "Broken," as you put in quotes, means it's not UTF-8. What kind of UTF-8
>> do you think there's a fair chance it'll match then?
>
> I've encountered old repositories where some path names
> were apparently not converted properly into UTF8 but
> contained whatever locale-based strings the client sent.
And such repositories were used without problems with current clients? I
can hardly believe that.
> Those would still match "*.someExtension" and similar
> patterns despite having non-UTF8. I would like to cover
> those as well.
But not *.ك
I think you're trying to solve an edge case in an API function by means
of an undocumented implementation detail. My advice is to throw it out.
-- Brane
