Re: [PATCH] Add -k kill keyword expansion option to git-cvsimport

From: Martin Langhoff <martin.langhoff@gmail.com>
Date: Aug 15, 2005, 07:37 UTC
Message-ID: <46a038f90508150037f128d6@mail.gmail.com>
In-Reply-To: <7vk6in65dp.fsf@assigned-by-dhcp.cox.net>

On 8/15/05, Junio C Hamano <junkio@cox.net> wrote:

> The discussion between you and Linus since you brought this up
> has kept me wondering if -ko is the only thing people may want
> to do, or sometimes -kk or even -kb or -kv make sense for some

The git-cvsimport script requests the full file at a given revision straight from the cvs server daemon it has instantiated. So I suspect we are mixing up cvs client flags with protocol flags. Hmm, reading up on it ( http://www.elegosoft.com/cvs/cvsclient.html#SEC9 ) the flags are carried through, and it sounds like it's pretty broken.

Still, it clearly leaves handling of newlines to the client. The difference between -kb and -ko (we are using -ko ATM) is the newline handling if you are using standard cvs clients. With git-cvsimport, we are doing the unix thing, which happens to be right thing to do.

Perhaps repos created with early versions of CVSNT are broken in this regard, but tough.

I think -kv is just the wrong thing to do if you are migrating to git. Anyway, this script has so far followed cvs's own default... which is -kv, and I am generally unwilling to break backwards compatibility. Though we could make it default to 'on' and provide '-K' for those masochistic enough to want it.

People with repos where cvswrappers was set to mark files as -kb or -ko are safe from all this pain and tears.

cheers,

martin