[cvsnt] Re: codepages on unix

[Tony Hoyle]

Richard Wirth wrote:
> Hello Tony,
> 
>   I have a small problem concerning codepage translation:
> 
>   1. CVSNT servers on unix try to run in UTF-8 if the system supports
>      this.

Yes this has been true for a while.  It's the only way to ensure a 
common protocol - cvsnt clients then translate back to their local 
codepage as required (2.6 or possibly 2.5.04 will probably have a 
solution for older clients but I've not really looked at how hard 
that'll be yet... mostly it's just a matter of seeing if the client has 
sent a server-codepage request and if not translating outgoing/incoming 
messages to/from something user defined (on Win32 that'd be the ANSI 
codepage normally).  Whether that is easy to drop into the code 
somewhere is something that needs looking at).

>   2. If one moves a repository from WIN32 or cvshome server, the
>      codepage of the server has to be overwritten, or the repository
>      has to be converted.

You'd have to do that anyway for a Win32->Unix or Unix->Win32 change, 
since the character sets are often different.  It's not been an issue 
for people normally - really only affects the comments.

RCS files don't contain codepage information so it's not trivial to 
convert automatically -  you can have comments in multiple languages 
etc. with no indication of their original character set.

>   3. If one overwrites the servers codepage via /etc/cvsnt/PServer
>      clients running on the same machine as the server also get this
>      codepage forced.

In local mode you have to, so everything gets written to the repository 
correctly.. in client mode we don't need to but it's setup long before 
we parse the cvs root string.  Shouldn't matter normally though - unix 
is quite good at handling this stuff.

Tony