grep problem?

[Eric Blake]

On 04/04/2011 04:09 PM, Jim Garrison wrote:
> I'm getting weird behavior from grep. Searching for a bracketed range of characters (i.e. [A-Z]) is doing case-insensitive matching, while an identical but explicit character set match (i.e. [ABCDE...Z]) does not.

Your problem is not with grep, but with your LC_COLLATE settings (which
inherit from LC_ALL).

POSIX states that range expressions (such as [A-Z]) are undefined in any
locale except C; and some locales (like en_US.UTF-8) happen to treat A-B
as AaB, A-b as AaBb, and so forth (that is, they collate
case-insensitively).

> 
> $ grep '[a-b]' test.dat
> abcde
> ABCDE

So, in a case-insensitive collation, this range expression includes at
least one of A or B (but probably not both); and since that matches the
ABCDE line, you get a correct result for the collation locale you requested.

> 
> Contrast with the correctly-working examples below
> 
> $ grep '[ab]' test.dat
> abcde

Here, there's no range, so there's no ambiguity.

Also, try "LC_ALL=C grep '[a-b]' test.dat" to see a difference.

-- 
Eric Blake   eblake@redhat.com    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 619 bytes
Desc: OpenPGP digital signature
URL: <http://cygwin.com/pipermail/cygwin/attachments/20110405/7076fd75/attachment.sig>