Information regarding cvs [server aborted]: received broken pipe signal

[Mark Lancisi]

 

I am submitting this info as suggested at :

 

http://www.cvshome.org/docs/manual/cvs_21.html#SEC183

 

in the following entry:

 

cvs [server aborted]: received broken pipe signal

This message seems to be caused by a hard-to-track-down bug in CVS or the systems it runs on (we don't know--we haven't tracked it down yet!). It seems to happen only after a CVS command has completed, and you should be able to just ignore the message. However, if you have discovered information concerning its cause, please let us know as described in H. Dealing with bugs in CVS or this manual.

 

 

If this issue has not been tracked down as yet ( and I could not find any evidence on the net that it has been ) here is some info regarding at least one of the causes. Note that I worked around the problem and do not intend to fix it (or get it fixed) in cvs (since there is a work around)..

 

Details:

 

I ran into this error message

cvs [server aborted]: received broken pipe signal

after implementing a tool (in perl), via the loginfo file, to remind engineers (via email) to double update when checking into subordinate branches as listed in a self maintained control file.  The error seemed benign when checking files into a single directory, (though it turned out even this was not benign because some automation I had written that was attempting to do a commit was catching the error and bailing out, which is how I found the problem in the first place), however when  you did a top level  commit expecting files from multiple directories to be checked in, the server would bail after processing and checking in files from the first directory.. ( so you cannot really "ignore the message" as is suggested in the manual, if you expect to be able to do multi-level commits).

After much frustration and testing in a cvs project off to the side (but in the same repository), I found that the problem occurred only when the script I wrote didn't send any mail to anyone because there was no reason too (ie the branch or trunk they were checking into was not listed as a subordinate of any other branch or the trunk..)

So - the problem, I believe, is that the server does not like any script that is called via loginfo to NOT send mail. This could be a slight simplification of the problem, as I didn't prove that this was the only cause of the broken pipe message, but it certainly was behaving this way with regard to my script, so I know it is at least one cause.

I'd been using loginfo for a year prior to send mail on commit but it turns out that the script I am using there is set up to always send mail on every commit, so I never saw this problem.

To get around the problem, I changed my new script to send mail to a nobody user set up with an inbox of (essentially) /dev/null whenever it doesn't need to send any mail at all...  This seems to sufficiently fake out the server and keeps from flooding a dummy inbox while avoiding the "broken pipe signal" thereby fixing multi-level commits (and my automation ;-)  ) .

Feel free to post this anywhere on the net if it can be of help.

 

Thanks,

Mark Lancisi

Release Engineer

Netezza Corp.