[cvsnt] Core dumps on Solaris 10

[Arthur Barrett]

Tom,

> /usr/local/bin/sparcv9/cvsnt -f --allow-root=/export/lo\
>     cal/home/toolsa

OK - first of all - forget everything you know about CVS - CVSNT is NOT
CVS.  Get rid of all the options to cvsnt in inetd.conf like -f and
--allow-root.  CVSNT doesn't have all the security headaches of CVS and
needs to have all the repositories configured in /etc/cvsnt/PServer for
ALL protocols (except for :local:) - unfortunately the name of the file
is a little misleading.

Second - unless you are going to be working with very large revisions
(eg: > 500MB) or lots of revisions of largish files (eg: 500 revisions
of a 100Mb file) then use the 32 bit CVSNT on Solaris.  On the
commercial side the minimum price tag for x64 installs is significantly
higher than the minimum price of a x32 install because the x64 builds
are a lot of work to get right and the extra time consumes the extra
money.

>      ffffffff7f17c1fc _ZN8CRunFile4waitERii (ffffffff7fffd980, 
> ffffffff7fffd974, ffffffffffffffff, ffffffff7fffd090, 0, 0) \
>     + 610
>      ffffffff7bb09fb4 

Do you have cvslockd running on the server?  You should.  Check you can
connect to it ok on the server using 'telnet localhost 2402'.  Some
'external' command is being ran (which is either a trigger that you have
defined or the server manually trying to start the cvs lock daemon
because it's not already running).


> looks like the postcommand trigger has not yet officially 
> completed and the error is occurring in the socket 
> communications back to the client.

Do you have anything defined in the postcommand, ie:
CVSROOT/postcommand?


> Unfortunately, when I enable tracing in PServer via:
> 
>     AllowTrace=1
>     ServerTraceFile=/tmp/cvsnt.log
> 
> I get a client-side error of:
>     15:39:16: S -> tag_fileproc - finally RCS_settag
>     15:39:16: S -> tag_fileproc - finally RCS_settag ok
>     15:39:16: S -> tag_fileproc(2) rewrite 
> rcsfile="/export/local/home/toolsadm/CVS_REPO/OIL/README.txt,v" 
> symtag="anothert\
>     ag", rev="1.14", date="2008.12.22.20.39.16"
> 
>     cvsnt [tag aborted]: end of file from server (consult 
> above messages if any)

Yes - Solaris unfortunately does that quite a lot, we are trying to make
all the debugging Solaris safe but it takes time... I'll try and find
what TRACE statement we've missed based on your results...

Regards,


Arthur