On Tue, Dec 18, 2018 at 10:15:56AM +0100, Stefan Sperling wrote: > On Tue, Dec 18, 2018 at 03:55:42AM +0100, Stefan Fuhrmann wrote: > > Error handling during tear-down is always a bit messy. > > > > I guess not using the request struct will mean we cannot > > tell what operation cased a problem - ever (not sure, > > though). If true, that's a high cost. > > > > One way to improve your solution would be to make > > the error handling degrade as the objects are being > > destroyed. But it adds / duplicates a bit of code: > > > > * keep the logging based on request context > > * apr_pool_pre_cleanup_register on the request pool, > > a switch to the new connection-based logging > > > > That's a fair suggestion. I'll try to implement this.
Implemented in https://svn.apache.org/r1850651
