bug#16919: [PATCH] fix mismatch between dfa and regex for treatment of titlecase

[Paul Eggert]

On 03/05/2014 07:11 AM, Norihiro Tanaka wrote:
> I still believe that upper or lower case of a character should
> also match title case

The (soon-to-be-fixed) gnulib regex code agrees with you, assuming that 
towupper (X) agrees for all three values of X, because it uses (towupper 
(input) == towupper (pattern)). However, the most-plausible reading of 
POSIX does not agree with you, as it would require (input == pattern || 
towlower (input) == pattern || towupper (input) == pattern), which means 
a titlecase pattern will match only itself.

It seems pretty clear to me that the most-plausible reading of POSIX is 
buggy, for this reason.  No wonder so many implementations fail to 
conform to it.

I thought of a different way where gnulib/glibc regex does not conform 
to POSIX, and here there doesn't seem to be any ambiguity about it.  In 
the POSIX locale when ignoring case, the pattern '[Z-a]' matches the 
data 'Z', 'z', 'A', 'a', and the nonalphabetic characters like '^' that 
collate between 'Z' and 'a'.  But the glibc regex code rejects that 
pattern entirely.  Conversely, in the same situation the glibc regex 
code says '[A-z]' matches only alphabetic characters, whereas POSIX says 
it should also match the nonalphabetic characters like '^' that collate 
between 'Z' and 'a'.  It appears that nobody cares, as this 
incompatibility has been present for years and I don't recall anyone 
complaining.  Though it is weird that this means "grep PAT" can match 
some lines that "grep -i PAT" doesn't.

Here POSIX is not merely ambiguous, it's clearly disagreeing with common 
practice.  It's not clear whether the bug is in POSIX or in the 
implementation.