Daniel Shahaf wrote: > Julian Foad wrote on Mon, Dec 10, 2018 at 18:23:00 +0000: > > Robustness: teach it to only accumulate new entries, and not to wipe the > > file when there's a network glitch. > > That's easy enough: just write 'set -e' [or use] double ampersands
I have done that but I think the script will still generate bad output if the 'curl' commands fail; see TODO below. > Another nice to have improvement would be making the 'svn ci' invocation > silent — by adding --quiet, or by running it under chronic(1) (from package > 'moreutils') — so the people on the crontab's MAILTO list don't get emails for > successful commits. Added '-q'etc. I also made it: * not commit if the only change is in the the "# Generated ... <date>" line. * search only in the "publish" subtree, so it doesn't pick up the example in its own script New version: # Update our Haxx-URL-to-Message-Id map (a manual cron entry, for now) 0 4 * * * fn=publish/.message-ids.tsv; cd ~/src/svn/site && svn up -q && tools/haxx-url-to-message-id.sh publish > $fn.tmp && if diff -q -I'^# Generated' $fn $fn.tmp > /dev/null; then rm $fn.tmp; else mv $fn.tmp $fn && svn ci -q -m "* $fn: Automatically regenerated" $fn; fi TODO: * only add new entries (actually it happens to work this way already, because it finds its own previous output file as part of the scan); * only query 'haxx.se' for new entries, not all the existing entries again each time; * the 'curl | perl' part of the script wants a way to fail if the 'curl' fails (Bash has 'set -o pipefail' for this). -- - Julian
