Hi, Andi discussed this with me and we agreed to simply revert the change and re-do regression testing and release-packaging then, I hope nobody objects.
Anybody who would like to make use of the performance gains in single-threaded applications can call the setter outside of POI easily, see the comment in the bug at https://bz.apache.org/bugzilla/show_bug.cgi?id=61350 Thanks... Dominik. On Wed, Aug 23, 2017 at 11:15 PM, Greg Woolsey <greg.wool...@gmail.com> wrote: > Once again legacy xmlbeans disappoints. I would prefer making the old > behavior the default, with a setting/flag/property that can change to the > faster behavior. Consumers can then test it in their environments. > > On Wed, Aug 23, 2017, 13:49 Dominik Stadler <dominik.stad...@gmx.at> > wrote: > > > Hi, > > > > regression results are in, unfortunately they don't look too good from a > > quick look. > > > > There are a number of NullPointerException and > > ArrayIndexOutOfBoundsException somewhere deep in XmlBeans code which > cannot > > be reproduced when running a single file. I suspect "#61350 - Use > > unsynchronized xmlbeans" to be related. The test runs in two threads at > the > > same time, but always handles separate files. This is something that we > > documented as allowed and is used by many downstream projects, only > > handling the same item in multiple threads is not supported. > > > > Do we need to make this synchronization setting configurable instead so > you > > can get the additional performance if you do not use more than one thread > > at all? > > > > Additionally a few "IllegalArgumentException: typeface can't be null nor > > empty", probably related to some HSLF changes and a few others with only > > very few occurrences that I cannot match to recent changes easily. > > > > See http://people.apache.org/~centic/poi_regression/reports/ and > > http://people.apache.org/~centic/poi_regression/reportsAll/ for details. > > > > Thanks... Dominik. > > > > > > On Tue, Aug 22, 2017 at 10:49 PM, Dominik Stadler < > dominik.stad...@gmx.at> > > wrote: > > > > > Hi, > > > > > > My regression tests are started and will hopefully finish in the next 2 > > > days. > > > > > > Dominik. > > > > > > On Tue, Aug 22, 2017 at 1:06 AM, Andreas Beeker <kiwiwi...@apache.org> > > > wrote: > > > > > >> You are right ... I simply wanted to push out that message, to not > leave > > >> the trunk changes > > >> unaccompanied for too long. > > >> > > >> As I have to clarify with the Tika devs anyway, what their stance on > > JDK8 > > >> is, this will probably > > >> take anyway a bit longer to be announced. > > >> > > >> Andi. > > >> > > >> On 8/22/17 1:00 AM, Javen O'Neal wrote: > > >> > > >> Let's leave this vote open until Dominik and/or Tim can run tests > > against > > >> our corpus and Tika. > > >> > > >> If this is the last Java 6-compatible release, let's make it a good > one. > > >> > > >> On Aug 21, 2017 3:43 PM, "Andreas Beeker" <kiwiwi...@apache.org> < > > kiwiwi...@apache.org> wrote: > > >> > > >> > > >> Hi, > > >> > > >> I've prepared artifacts for the release of Apache POI 3.17 (RC1). > > >> > > >> The most notable changes in this release are: > > >> > > >> - Various modules: add sanity checks and fix infinite loops / OOMs > > caused > > >> by fuzzed data > > >> - OPC: fix linebreak handling on XML signature calculation (#61182) > > >> - SS Common: fix number formatting (github-43/52, #60422) > > >> - SXSSF: fix XML processing - unicode surrogates and line breaks > > (#61048, > > >> #61246) > > >> https://dist.apache.org/repos/dist/dev/poi/3.17-RC1/ > > >> > > >> All tests pass. ASC files verify and MD5/SHA1 are correct. Docs look > > fine. > > >> I'll add the summary to the change log on releasing the artifacts. > > >> > > >> Please vote to release the artifacts. > > >> The vote keeps open for 72hrs, 2017-08-25, 23:59 UTC, > > >> planned release announcement date is Saturday, 2017-08-27. > > >> > > >> Here is my +1 > > >> > > >> Andi > > >> > > >> > > >> > > >> > > >> > > >> > > > > > >
