Re: PSC Motion: require SQLite 3.11 as minimum version
Even Rouault <[email protected]> Thu, 07 Nov 2019 19:51:08 +0100
| Newsgroups | gmane.comp.gis.proj-4.devel |
|---|---|
| Message-ID | <2021685.NZW87uO35g@even-i700> |
On jeudi 7 novembre 2019 18:27:59 CET Kristian Evers wrote: > I am also +1 on this. I am just wondering if 3.11 is the lowest possible > version that is feasible to use with PROJ. From the linked issue I > understand that SQLite 3.8.x is too slow and that it works with 3.11. I > take it that 3.11 was chosen because it was readily available for Even. > Would 3.9 or 3.10 work as well? Or was there a specific change in 3.11 that > we know is the one that makes the difference? 3.11 is likely not the minimum version. I tried to look at the SQLite release notes to find the right version, but nothing obvious struck me, so this would require doing bisection to find out the minimum version. I don't think it really matters to spot the exact version. From the feedback we got up to now, the practical concern is with RHEL 7 that ships something too old. Other environments have at least 3.11. And anyway, even if it did work for now with something older than 3.11, I wouldn't test routinely against it, and our CI doens't test that anymore (it is on Xenial on master, so using 3.11), so it might break in the future. On thinking, what has intriguied me is that on Travis, the 6.2 branch runs with Ubuntu Trusty / SQLite 3.8.2. Looking at logs, I can indeed see that the runtime of test_cpp_api reported there is abnormaly long (45s) but we didn't notice. Locally, with a same -O2 build of the 6.2 branch, I get 5.2s with sqlite 3.11 vs 32s with 3.8.2. So there might be some amortizing of the penalty for runs in the same process / reusing the same database connection. -- Spatialys - Geospatial professional services http://www.spatialys.com _______________________________________________ PROJ mailing list [email protected] https://lists.osgeo.org/mailman/listinfo/proj