Julian Foad wrote on Mon, Nov 19, 2012 at 15:16:44 +0000: > Stefan Fuhrmann wrote: > > relatively rare and constructing an error object can be > > expensive (NLS). svn_stringbuf_from_file2 is also a very > > convenient function to use. > I wonder if we could change our generic error handling scheme to defer > NLS look-up until the time when we print the error. That would > eliminate that overhead in cases where we handle and clear an error > without printing it. Something to think about for the future.
I thought we considered error object creation to be expensive for other reasons, too, not just for NLS reasons? (eg, using a global pool) Daniel
