SVN hook errors get lost

[strk]

I've found errors in the PostGIS SVN post-commit hook and found out they get lost which explain why they survived for so long.

One of the errors is a mail to an inexistent address. The mail is sent as svn_postgis@…, but that address does not exist, so the reply from MAILER_DAEMON never get to be read by anyone.

The other error is asyncronous, coming from a backgrounded wget process spowned by the script. The wget'ed url is non-existent but the hostname resolves and the firewall does not respond so that the client hangs indefintetly (till timeout).it takes a long time to timeout. I still don't know what will be output on timeout, but whatever it is, it doesn't hit anyone.

I think the sane thing to do here would be to ensure SVN errors for a projects are sent to a responsible person from the project.

[strk]

I found that "svn_postgis@…" address is really specified in the commit hook script itself, so not something generally inherited. I've seen that trac mail (for "-tickets" mailing lists) are sent with from set to "trac@…". Does _that_ mailbox end up anywhere ?

[neteler]

FWIW: You may want to look at

/var/www/svn/repos/grass/hooks/post-commit

for a working solution. It posts to

​https://lists.osgeo.org/pipermail/grass-commit/

In the mailman setting of grass-commit we have specified the grass-dev list as reply-to (in order to avoid that comments pollute the grass-commit list and in order to reach the developers right away).

[strk]

Thanks Markus, that script is not much different from the one we had in PostGIS, so it suffers of the same problem, that is shall the target email (grass-commit@…) have a delivery error, that error will be sent to svn_grass@… which is an unexistent address.

This ticket is to define a policy about who should receive errors on such occurrences (if anyone).

BTW, good idea about using -dev for reply-to!

[robe]

this ticket is moot