[cvsnt] Re: Incompatibility with "plain" CVS server?

[Johan Holmberg]

Tony Hoyle <tmh at nodomain.org> writes:
>
> > - What role does the "_H_" play in CVSNT?
> 
> It's used mostly server-side to handle directory versioning... it
> exists for a fraction of a second between two operations (predirectory
> and directory callbacks).  The client also uses it in some
> circumstances but for much less.  It does not exist on files and is
> never written to the RCS files, and is never sent over the protocol.
> 

I think I showed in my earlier mail that the '_H_' characters can be
"sent over the protocol" if it is part of the content of a "CVS/Tag"
file (when a file is added on a branch).

The last days I have been trying to understand how the '_H_' appeared
in our CVS/Tag files (about 150 files, and we have NOT been
"hand-editing" all those files :). I've tried to look in the CVSNT
source but I haven't understood enough to be able to re-create the
same situation with '_H_' in a new sandbox.

You wrote earlier (about the '_H_'):

> The client also uses it in some circumstances but for much less.

Could you be more specific about what this "much less" is?  I
understand that it is not possible to explain all of how CVSNT works
in a few paragraphs, but with a little more knowledge I might perhaps
be able to investigate things further myself.

Could it be possible that that the CVSNT client had rewritten the
"CVS/Tag" files from "TB041122" to "TB041122:_H_", and then later
"crashed" (for some unrelated reason), leaving the "CVS/Tag" files in
this state?

I'm of course just guessing, since I don't understand the source well
enough. But it makes me very uncomfortable not knowing the reason for
our current problems. As far as I can tell, they could occur again
tomorrow.

/Johan Holmberg