This post documents how I’m changing the search methods for pkg-message. It uses full-text search, but instead of using a special column, I’m going to use an index. I will elaborate. When searching pkg-message was added to the website, a different approach was taken than in the past. In Converting a column-based expression to a …
Moving from storing the tsvector values to a tsvector index Read More »
One of the primary uses for FreshPorts is searching various characteristics of the ports. One highly used search feature is pkg-messages (background at pkg-message). A problem which has been vexing me for about 2 months was the lack of results on prod, but results which worked on dev. The issue was raised on GitHub. Initially, …
When tsvector was working as expected on most hosts, but not one Read More »
Tonight I was setting up a new FreshPorts node for development. It took me a while to remember the steps. sudo su -l postgres psql template1 < globals.sql createdb -T template0 -E SQL_ASCII freshports.org psql freshports.org create language plperl; exit; time pg_restore -j 32 -d freshports.org ~dan/dump/freshports.org.dump
Based on the Add links to GitHub and GitLab mirrors for ports and commits request, the website has some new links arriving soon. Link to GitHub from each commit A similar link to GitLab Use the one you like. Each git web interface has it’s own followers. Also, while there, the commit hash at the …
New links to GitLab and GitHub, in addition to the exist cgit.freebsd.org Read More »
The Libreoffice homepages seem wrong. dev and test do not show a homepage for arabic/libreoffice, but stage and prod do. Could this be a commit which corrected something after dev and test were corrected? I think so. I think this commit did that. Let’s check from the command line. I’m doing this in my empty …
By “post correction”, I meant “post-correction script”, i.e. after the correction script completed. A recent commit started the process of moving the WWW from pkg-descr to Makefile. These were huge commits, about 27930 ports each. FreshPorts is dumb. It sees a commit and it refreshes all ports touched by that commit. Refreshing every port takes …
Today I learned, or at best, relearned, a new make target: root@freshports:/usr/ports/sysutils/anvil # make create-manifest root@freshports:/usr/ports/sysutils/anvil # ls -lt | head total 14 drwxr-xr-x 3 root wheel 3 Sep 10 17:26 work -rw-r–r– 1 root wheel 337 Sep 10 16:06 pkg-descr -rw-r–r– 1 root wheel 182 Sep 10 16:06 distinfo -rw-r–r– 1 root wheel 1050 …
A long-standing bug, perhaps dating back to April 2003, has been fixed. It was an optimization in a trigger: if the values in the trigger have not changed, let’s not run the trigger. Turns out, that was a bad idea. The table The ports_categories table documents the connections between a port and the categories in …
The July 4th tweet announced “dev, test, and stage .freshports.org are now running Python 3.9”. Three days later, “Is it coincidental that dev, test, and stage are all stuck on commits from July 4?” I took notes as I explored the problem. The first clue was: 2022.07.07 00:12:03 git-delta.sh /usr/local/libexec/freshports/git-to-freshports-xml.py –repo ports –path /var/db/ingress/repos/ports –branch …
I need some help with the git processing shell script. Look at line 82 in this script please. See those hardcoded branch names? That’s my problem. Let’s reduce the shell script to a simple proof of concept. % cat ~/tmp/poc.sh #!/bin/sh git for-each-ref –format ‘%(objecttype) %(refname)’ \ | sed -n ‘s/^commit refs\/remotes\///p’ \ | while …