<p>In early 2019 the university computing service upgraded its firewall rules which took the
server offline completely.
<p>
Wookey eventually managed to find us free space (a virtual machine)
on a debian mirror server somewhere in Leicestershire (we think).
This move to a different secure server means that all ssh to the server now needs to use cryptographic keys tied to individual machines. There is an expo-nerds email list (all mailing lists are now hosted on wookware.org as the university list system restricted what non-Raven-users could do) to coordinate server fettling.
<p>The troggle software has been migrated to git, and the old erebus and cvs branches (pre 2010) removed. Some decrufting was done to get rid of log files, old copies of embedded javascript (codemirror, jquery etc) and some fat images no longer used.
The loser repo and expoweb repo need more care in migration (expoweb is the website content - which is published by troggle). Loser should have the old 1999-2004 CVS history restored, and maybe Tom's annual snapshots from before that, so ancient history can usefully be researched (sometimes useful). It's also a good idea to add the 2015, 2016 and 2017 ARGE data we got (in 2017) added in the correct years so that it's possible to go back to an 'end of this year' checkout and get an accurate view of what was found (for making plots and length stats). All of that requires some history rewriting, which is best done at the time of conversion.
Similarly expoweb is full of bloat from fat images and surveys and one 82MB thesis that got checked in and then removed. Clearing that out is a good idea. I have a set of 'unused fat blob' lists which can be stripped out with git-gilter. It's not hard to make a 'do the conversion' script, ready for sometime after expo 2019 has calmed down.
Wookey has now moved 'expoweb' from mercurial to git largely "as-is" and will to use the git tools to patch up the history and to remove redundancies, rather than the original plan to tidy them up "at the time of conversion". Mark Shinwell is working on loser with him.
<p>Sam continues to work on upgrading django from v1.7 on python 2.7.17 . We would like to upgrade django as quickly as possible because old versions of django have unpatched security issues.
Upgrading to later django versions is a real pig - not helped by the fact that all the tools to help do it are now out of date for these very old django releases.
<ul>
<li>"Django 1.11 is the last version to support Python 2.7. Support for Django 1.11 ends in 2020." see: <ahref="https://docs.djangoproject.com/en/3.0/faq/install/">django versions</a>. You will notice that we are really outstaying our welcome here, especially as python2.7 was <ahref="https://python-release-cycle.glitch.me/">declared dead in January</a> this year.
<li>For a table displaying the various versions of django and support expiry dates
<li>Ubuntu 20.04 came out on 23rd April but it does not support python2 at all. So we cannot use it for software maintenance (well be can, but only using non-recommended software, which is what we are trying to get away from).
</ul>
<p>We planned to upgrade from django 1.7 to django 1.11, then port from python2 to python3 on
the same version of django, then upgrade to as recent a version of django as we could. But we have
discovered that django1.7 works just fine with <ahref="https://docs.djangoproject.com/en/1.10/topics/python3/">python3</a>, so we will probably move to python3 during June and
then upgrade the server operating system from Debian <var>stretch</var> to <var>buster</var> before
tackling the next step: thinking deeply about when we migrate from django
<p>Sam was a bit overworked in trying to get an entire university to work remotely and Philip got troggle on django 1.7 to work on python 3.5 and then 3.8. He then did the slog of migrating it through the django versions up to 1.11.29 - the last version before django 2.0 when a new URL routing architecture becomes available. 1.11.29 is an LTS (long term support) version of django. In doing this we had to retreat to python3.7 due to a django incompatibility; which is fine because the server will run python3.7 when it is updated to debian 10 (buster) [Wookey to do].
<p>
In the course of these migrations several unused or partly-used django plugins were dropped as they caused migration problems (notably staticfiles) and the plug-ins pillow, django-registration, six and sqlparse were brought up to recent versions. This was all done with pip in a python venv (virtual environment) on a Windows 10 machine running ubuntu 20.04 under WSL (Windows Systems for Linux) v1. Also missing troggle functions were repaired and partly-implemented pages, such as the list of all cavers and their surveyed passages, were finished and made to work. The survex file parsing was completely rebuilt to reduce the excessive memory footprint while parsing.
<p>We plan to stick with debian 10 (buster), django 1.11.29 and python 3.7.7 (the standard on buster) until spring 2021 when we will upgrade debian to the forthcoming stable release 11. At that point debian will have python 3.8 as standard and we will migrate to django 2.x, hopefully getting as far as django 2.2 which is an LTS and actually in support until April 2022.
<p>With any luck that will be the last of our involvement with django migrations as we will keep using django 2.2 until we stop using django altogether, see <ahref="trogspeculate.html">troggle architecture speculations</a>.
