Calc with it's huge data-sets is the app that this would most likely benefit most from; I attach a small sample document and a callgrind profile showing the unzip + SAX parse + namespace resolution at >10% of document load time.
10% is not that much, but we other ongoing work to improve other calc core code performance :-)
Created attachment 77071 [details]
sample sheet with 100k random doubles in a column
callgrind profile at: http://users.freedesktop.org/~michael/callgrind-load-doubles.txt.gz - download, gzip -d and run kachegrind to see the result (if you're interested).
Michael, were these patches https://lists.freedesktop.org/archives/libreoffice/2017-August/078392.html solution of this enhancement?
If yes, then please close this as Resolved Fixed.
Yes - thanks, lets close this - the proportion nowadays is ~50% of document load time - and again we need to thread more - but ... ;-) that needs a different bug really.