Re: [PATCH v3] wildmatch: properly fold case everywhere

[Anthony Ramine <n.oxyde@gmail.com>]

Replied inline.

-- 
Anthony Ramine

Le 29 mai 2013 à 15:52, Duy Nguyen a écrit :

> On Wed, May 29, 2013 at 8:37 PM, Anthony Ramine <n.oxyde@gmail.com> wrote:
>> Le 29 mai 2013 à 15:22, Duy Nguyen a écrit :
>> 
>>> On Tue, May 28, 2013 at 8:58 PM, Anthony Ramine <n.oxyde@gmail.com> wrote:
>>>> Case folding is not done correctly when matching against the [:upper:]
>>>> character class and uppercased character ranges (e.g. A-Z).
>>>> Specifically, an uppercase letter fails to match against any of them
>>>> when case folding is requested because plain characters in the pattern
>>>> and the whole string and preemptively lowercased to handle the base case
>>>> fast.
>>> 
>>> I did a little test with glibc fnmatch and also checked the source
>>> code. I don't think 'a' matches [:upper:]. So I'm not sure if that's a
>>> correct behavior or a bug in glibc. The spec is not clear (I think) on
>>> this. I guess we should just assume that 'a' should match '[:upper:]'?
>> 
>> I don't know, in my opinion if case folding is enabled we should say [:upper:], [:lower:] and [:alpha:] are equivalent.
>> 
>> This opinion is shared by GNU Flex [1]:
>> 
>>>      • If your scanner is case-insensitive (the ‘-i’ flag), then ‘[:upper:]’ and ‘[:lower:]’ are equivalent to ‘[:alpha:]’.
>> 
>> [1] http://flex.sourceforge.net/manual/Patterns.html
> 
> Then we should do it too because of this precedent, I think.
> 
>>>> @@ -196,6 +196,11 @@ static int dowild(const uchar *p, const uchar *text, unsigned int flags)
>>>>                                       }
>>>>                                       if (t_ch <= p_ch && t_ch >= prev_ch)
>>>>                                               matched = 1;
>>>> +                                       else if ((flags & WM_CASEFOLD) && ISLOWER(t_ch)) {
>>>> +                                               uchar t_ch_upper = toupper(t_ch);
>>>> +                                               if (t_ch_upper <= p_ch && t_ch_upper >= prev_ch)
>>>> +                                                       matched = 1;
>>>> +                                       }
>>> 
>>> Or we could stick with to tolower. Something like this
>>> 
>>> if ((t_ch <= p_ch && t_ch >= prev_ch) ||
>>>  ((flags & WM_CASEFOLD) &&
>>>     t_ch <= tolower(p_ch) && t_ch >= tolower(prev_ch)))
>>>  match = 1;
>>> 
>>> I think it's easier to read if we either downcase all, or upcase all, not both.
>> 
>> If the range to match against is [A-_], it will become [a-_] which is an empty range, ord('a') > ord('_'). I think it is simpler to reuse toupper() after the fact as I did.
>> 
>> Anyway maybe I should add a test for that corner case?
> 
> Yeah I was thinking about such a case, but I saw glibc do it... I
> guess we just found another bug, at least in compat/fnmatch.c. Yes a
> test for it would be great, in case I change my mind 2 years from now
> and decide to turn it the other way ;)

Should I patch compat/fnmatch.c too? That would make it different from the glibc's one.

>> 
>>>>                                       p_ch = 0; /* This makes "prev_ch" get set to 0. */
>>>>                               } else if (p_ch == '[' && p[1] == ':') {
>>>>                                       const uchar *s;
>>>> @@ -245,6 +250,8 @@ static int dowild(const uchar *p, const uchar *text, unsigned int flags)
>>>>                                       } else if (CC_EQ(s,i, "upper")) {
>>>>                                               if (ISUPPER(t_ch))
>>>>                                                       matched = 1;
>>>> +                                               else if ((flags & WM_CASEFOLD) && ISLOWER(t_ch))
>>>> +                                                       matched = 1;
>>>>                                       } else if (CC_EQ(s,i, "xdigit")) {
>>>>                                               if (ISXDIGIT(t_ch))
>>>>                                                       matched = 1;
>>> 
>>> If WM_CASEFOLD is set, maybe isalpha(t_ch) is enough then?
>> 
>> Yes isalpha() is enought but I wanted to keep the two cases separated, I can amend that if you want.
> 
> Either way is fine. I don't think this code is performance critical. Your call.
> --
> Duy