Alexis Métaireau - code

llm command-line tips

2023-09-27T00:00:00+02:00

I’m using llm more and more, and today I had to find back prompts I used in the past. Here is a command I’ve been using, which allows me to filter the results based on what I want. It leverages sql-utils, a cli tool which is able to talk to a SQLITE database and answer in json, and jq a command-line tool capable of doing requests for json.

All in all, it’s pretty satisfying to use. I finally got a simple way to query databases! I’m also using glow, which is capable of transforming markdown into a better version on the terminal.

“`bash sqlite-utils “$(llm logs path)” “SELECT * FROM responses WHERE prompt LIKE ‘%search%’” | jq ‘.[].response’ -r | glow ““

Which got me a colored response :-)

Setting up a IRC Bouncer with ZNC

2023-09-27T00:00:00+02:00

It’s been a while since I’ve used IRC, but I needed to connect to it today to discuss around Peewee.

The main issue with IRC is that you need to be connected to see the answer, and to get the context of the conversation. Unless… you set up a bouncer.

The bouncer is named ZNC, and the IRC client I use is Weechat.

So, that’s what I did:

Installation of ZNC

apt install znc
sudo -u _znc /usr/bin/znc --datadir=/var/lib/znc --makeconf
sudo systemctl enable znc

You can answer the questions asked by --makeconf, it will generate you a configuration file like this (stored in /var/lib/znc/configurations/znc.conf):

AnonIPLimit = 10
AuthOnlyViaModule = false
ConfigWriteDelay = 0
ConnectDelay = 5
HideVersion = false
LoadModule = webadmin
MaxBufferSize = 500
ProtectWebSessions = true
SSLCertFile = /var/lib/znc/znc.pem
SSLDHParamFile = /var/lib/znc/znc.pem
SSLKeyFile = /var/lib/znc/znc.pem
ServerThrottle = 30
Version = 1.8.2

<Listener listener0>
    AllowIRC = true
    AllowWeb = true
    IPv4 = true
    IPv6 = true
    Port = 6697
    SSL = true
    URIPrefix = /
</Listener>

<User alexis>
    Admin = true
    Allow = *
    AltNick = alexis_
    AppendTimestamp = false
    AuthOnlyViaModule = false
    AutoClearChanBuffer = true
    AutoClearQueryBuffer = true
    BindHost = skate.notmyidea.org
    ChanBufferSize = 50
    DenyLoadMod = false
    DenySetBindHost = false
    Ident = alexis
    JoinTries = 10
    LoadModule = chansaver
    LoadModule = controlpanel
    MaxJoins = 0
    MaxNetworks = 1
    MaxQueryBuffers = 50
    MultiClients = true
    Nick = alexis
    NoTrafficTimeout = 180
    PrependTimestamp = true
    QueryBufferSize = 50
    QuitMsg = See you :)
    RealName = N/A
    StatusPrefix = *
    TimestampFormat = [%H:%M:%S]

    <Network liberachat>
        FloodBurst = 9
        FloodRate = 2.00
        IRCConnectEnabled = true
        JoinDelay = 0
        LoadModule = simple_away
        RealName = N/A
        Server = irc.libera.chat +6697
        TrustAllCerts = false
        TrustPKI = true

        <Chan #peewee>
        </Chan>
    </Network>

    <Pass password>
        Hash = REDACTED
        Method = SHA256
        Salt = REDACTED
    </Pass>
</User>

You can access a web interface on the exposed port. I had to make a change in my Firefox configuration, in about:config, set network.security.ports.banned.override to 6697, otherwise, Firefox prevents you from connecting to these ports (which might actually be a good idea).

Weechat configuration

Now, to use this in weechat, here are some useful commands. First, get the fingerprint of the SSL certificate generated on your server:

cat /var/log/znc/znc.pem | openssl x509 -sha512 -fingerprint -noout | tr -d ':' | tr 'A-Z' 'a-z' | cut -d = -f 2
````

Then, in weechat :

```weechat
/server add znc host/6697 -tls -username=<username> -password=<yourpass> -autoconnect
/set irc.server.znc.tls_fingerprint <fingerprint-goes-here>
/connect znc

And you should be all set!

Resources : The ZNC Wiki on Weechat and the Debian page on ZNC

How to run the vigogne model locally

2023-09-22T00:00:00+02:00

Vigogne is a LLM model based on LLAMA2, but trained with french data. As I’m working mostly in french, it might be useful. The current models that I can get locally are in english.

The information I’ve found online are scarse and not so easy to follow, so here is a step by step tutorial you can follow. I’m using pipenv almost everywhere now, it’s so easy :-)

llm install -U llm-llama-cpp
wget https://huggingface.co/TheBloke/Vigogne-2-7B-Chat-GGUF/resolve/main/vigogne-2-7b-chat.Q4_K_M.gguf
llm llama-cpp add-model vigogne-2-7b-chat.Q4_K_M.gguf -a vigogne
llm models default vigogne

Creating a simple command line to post snippets on Gitlab

2023-09-18T00:00:00+02:00

I’m trying to get away from Github, and one thing that I find useful is the gist utility they’re providing. Seems that gitlab provides a similar tool.

You can use it using python-gitlab:

pipx install python-gitlab

And then :

gitlab snippet create --title="youpi" --file-name="snip.py" --content snip.py --visibility="public"

I now wanted a small bash script which will just get the name of the file and infer all the parameters. I asked GPT-4, and iterated on its answer.

Here’s the resulting bash script:

#!/bin/bash

if [ -z "$1" ]
then
    echo "Please provide a filename"
    exit 1
fi

file="$1"
base=$(basename "$file")
title="$base"
visibility="public"

# Use `cat` to fetch the content of the file
content=$(cat "$file")

result=$(gitlab snippet create --title="$title" --file-name="$title" --content="$content" --visibility="$visibility")

id=$(echo "$result" | awk '/id: / { print $2 }')
echo "https://gitlab.com/-/snippets/$id"

I can now do snip README.md and that will create the snippet for me :-)

Creating an online space to share markdown files

2023-09-17T00:00:00+02:00

I wanted to create a space on my server where I can upload markdown files and have them rendered directly, for them to be shared with other people.

I stumbled on the markdown module for nginx which does exactly what I want, but seemed to ask for compilation of nginx, which wasn’t exactly what I wanted in terms of maintainability (it would make it complicated to update it)

I then thought that the Caddy server does that by default, and so I’ve tested it out. Turns out it’s not, but it offers ways to do this thanks to its template mecanism.

It also, setups automatically and transparently SSL certificates for you (using Let’s Encrypt!), so I wanted to have a look.

Here is the Caddy configuration file I’m now using :

md.notmyidea.org {
        root * /home/caddy/md.notmyidea.org
        rewrite * /index.html
        file_server
        templates
        encode zstd gzip

}

And the template:

{{$pathParts := splitList "/" .OriginalReq.URL.Path}}
{{$markdownFilename := default "index" (slice $pathParts 1 | join "/")}}

{{if not (fileExists $markdownFilename)}}
    {{httpError 404}}
{{end}}

{{$markdownFile := (include $markdownFilename | splitFrontMatter)}}
<!DOCTYPE html>
<html>
    <head>
        <title>{{ $markdownFilename }}</title>
    </head>
    <body>
        {{ markdown $markdownFile.Body }}
    </body>
</html>

This is a minimalistic version, but it works :-)

Conversion d’un fichier svg en favicon.ico

2023-09-13T00:00:00+02:00

Il y a plusieurs sites qui permettent de faire ça automatiquement, mais j’aime bien faire les choses depuis mon terminal, voici donc une commande qui permet de faire ça simplement, en utilisant ImageMagick. Merci à ce gist

convert -density 256x256 -background transparent favicon.svg -define icon:auto-resize -colors 256 favicon.ico

Découverte de nouveaux outils pour le développement: LLM, Helix et plus

2023-09-12T00:00:00+02:00

LLM

LocalAI permet de faire tourner des modèles en local avec la même API HTTP que celle d’OpenAI
Le modèle Vigogne est un modèle entrainé (fine-tuned) avec des données en Français. Notamment ce modèlequi prends LLAMA2 en entrée.
LangChain semble être un framework pour travailler avec les différents concepts utiles. A voir.

Pour la première fois, j’ai commencé à utiliser un peu plus l’outil llm pour m’aider dans les tâches de programmation.

!! warning J’utilise actuellement par défaut le modèle en ligne d’OpenAI “GTP4”, à travers leur API. Cela me pose des problèmes éthiques, mais mon approche est pour le moment de voir le type de résultats que j’obtiens pour ensuite comparer avec des modèles locaux type LLAMA2.

Deux choses que j’ai trouvées utiles :

git diff | llm "write me a commit message"
git diff | llm "find a branch name"

Qui peuvent faciliter la vie plutôt que de chercher à décrire des choses manuellement (le “branch-name”, je le trouve particulièrement utile)

J’ai aussi trouvé l’idée intéressante d’automatiser l’écrite de tests, avec pytest. Bien sur ça ne fait pas tout, mais ça donne un template de départ qui est intéressant :

cat code/test_file.py | llm "write tests for this, using pytest"

J’ai par exemple demandé à llm de me lister 20 titres pour cet article :

cat ~/dev/til.notmyidea.org/content/$(date +%F).md | llm "Liste moi 20 titres pour cet article. Quelque chose de simple"
1. "Découverte de nouveaux outils pour le développement: LLM, Helix et plus"
2. "Explorer les modèles LLM pour la génération de code"
3. "L'amélioration du développement grâce à LLM et Helix"
4. "Introduction à l'éditeur Helix et à LLM"
5. "Boostez votre productivité de développement avec LLM et Helix"
6. "Comment j'intègre LLM et Helix dans mon flux de travail de développement"
7. "Les avantages de l'utilisation de LLM pour la programmation"
8. "Approches innovantes pour le développement de code avec LLM"
9. "L'avenir du développement de code avec LLM et Helix"
10. "Optimisation des tâches de programmation avec LLM"
11. "Ouvrir de nouvelles possibilités de codage avec LLM et Helix"
12. "Utiliser des modèles LLM pour améliorer la qualité du code"

Helix

Je me mets à tester un peu plus l’éditeur Helix. Je suis un habitué de Vim pour toutes les petites modifications que je fais depuis mon terminal, et Helix semble intéressant.

xpour sélectionner une ligne (xd pour sélectionner et supprimer)
ma pour sélectionner “around” quelque chose. maw pour sélectionner le mot
c pour effacer la sélection et passer en mode insertion.

Divers

J’ai fait confiance, j’ai appris. — Thomas

J’aime beaucoup ce que ça dit. Faire confiance est peut-être nécessaire, même si on est déçu au final, on aura au moins appris. Ça me touche.

Running the Gitlab CI locally

2023-08-19T00:00:00+02:00

Sometimes, I need to change how the continuous integration is setup, and I find myself pushing to a branch to test if my changes are working. Oftentimes, it takes me multiple commits to find the correct configuration, which is… suboptimal.

I discovered today Gitlab CI local which makes it possible to run the CI actions locally, without having to push to the remote CI. The same thing exists for Microsoft Github.

Under the hood, it’s using Docker, so you need to have it running on your system, but once it’s done, you just have to issue a simple command to see the results. Very helpful :-)

Here is an example :

$ gitlab-ci-local test
parsing and downloads finished in 41 ms
test  starting python:3.8-alpine (test)
test  copied to docker volumes in 4.05 s
test  $ apk update && apk add make libsass gcc musl-dev g++
test  > fetch https://dl-cdn.alpinelinux.org/alpine/v3.18/main/aarch64/APKINDEX.tar.gz
test  > fetch https://dl-cdn.alpinelinux.org/alpine/v3.18/community/aarch64/APKINDEX.tar.gz
test  > v3.18.3-55-g2ee93b9273a [https://dl-cdn.alpinelinux.org/alpine/v3.18/main]
test  > v3.18.3-56-g4a3b0382caa [https://dl-cdn.alpinelinux.org/alpine/v3.18/community]
test  > OK: 19939 distinct packages available
test  > (1/17) Installing libgcc (12.2.1_git20220924-r10)
test  > (2/17) Installing libstdc++ (12.2.1_git20220924-r10)
test  > (3/17) Installing libstdc++-dev (12.2.1_git20220924-r10)
test  > (4/17) Installing zstd-libs (1.5.5-r4)
test  > (5/17) Installing binutils (2.40-r7)
test  > (6/17) Installing libgomp (12.2.1_git20220924-r10)
test  > (7/17) Installing libatomic (12.2.1_git20220924-r10)
test  > (8/17) Installing gmp (6.2.1-r3)
test  > (9/17) Installing isl26 (0.26-r1)
test  > (10/17) Installing mpfr4 (4.2.0_p12-r0)
test  > (11/17) Installing mpc1 (1.3.1-r1)
test  > (12/17) Installing gcc (12.2.1_git20220924-r10)
test  > (13/17) Installing musl-dev (1.2.4-r1)
test  > (14/17) Installing libc-dev (0.7.2-r5)
test  > (15/17) Installing g++ (12.2.1_git20220924-r10)
test  > (16/17) Installing libsass (3.6.5-r0)
test  > (17/17) Installing make (4.4.1-r1)
test  > Executing busybox-1.36.1-r2.trigger
test  > OK: 246 MiB in 55 packages
test  $ pip install -r requirements.txt
test  > Collecting pelican
test  >   Downloading pelican-4.8.0-py3-none-any.whl (1.4 MB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.4/1.4 MB 539.9 kB/s eta 0:00:00
test  > Collecting markdown
test  >   Downloading Markdown-3.4.4-py3-none-any.whl (94 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 94.2/94.2 kB 540.1 kB/s eta 0:00:00
test  > Collecting typogrify
test  >   Downloading typogrify-2.0.7.tar.gz (12 kB)
test  >   Preparing metadata (setup.py): started
test  >   Preparing metadata (setup.py): finished with status 'done'
test  > Collecting pelican-search
test  >   Downloading pelican_search-1.1.0-py3-none-any.whl (6.6 kB)
test  > Collecting pelican-neighbors
test  >   Downloading pelican_neighbors-1.2.0-py3-none-any.whl (16 kB)
test  > Collecting pelican-webassets
test  >   Downloading pelican_webassets-2.0.0-py3-none-any.whl (5.8 kB)
test  > Collecting libsass
test  >   Downloading libsass-0.22.0.tar.gz (316 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 316.3/316.3 kB 552.1 kB/s eta 0:00:00
test  >   Preparing metadata (setup.py): started
test  >   Preparing metadata (setup.py): finished with status 'done'
test  > Collecting docutils>=0.16
test  >   Downloading docutils-0.20.1-py3-none-any.whl (572 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 572.7/572.7 kB 549.2 kB/s eta 0:00:00
test  > Collecting rich>=10.1
test  >   Downloading rich-13.5.2-py3-none-any.whl (239 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 239.7/239.7 kB 485.3 kB/s eta 0:00:00
test  > Collecting jinja2>=2.7
test  >   Downloading Jinja2-3.1.2-py3-none-any.whl (133 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 133.1/133.1 kB 342.6 kB/s eta 0:00:00
test  > Collecting pytz>=2020.1
test  >   Downloading pytz-2023.3-py2.py3-none-any.whl (502 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 502.3/502.3 kB 547.3 kB/s eta 0:00:00
test  > Collecting pygments>=2.6
test  >   Downloading Pygments-2.16.1-py3-none-any.whl (1.2 MB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.2/1.2 MB 551.4 kB/s eta 0:00:00
test  > Collecting unidecode>=1.1
test  >   Downloading Unidecode-1.3.6-py3-none-any.whl (235 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 235.9/235.9 kB 554.2 kB/s eta 0:00:00
test  > Collecting blinker>=1.4
test  >   Downloading blinker-1.6.2-py3-none-any.whl (13 kB)
test  > Collecting python-dateutil>=2.8
test  >   Downloading python_dateutil-2.8.2-py2.py3-none-any.whl (247 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 247.7/247.7 kB 235.7 kB/s eta 0:00:00
test  > Collecting feedgenerator>=1.9
test  >   Downloading feedgenerator-2.1.0-py3-none-any.whl (21 kB)
test  > Collecting importlib-metadata>=4.4
test  >   Downloading importlib_metadata-6.8.0-py3-none-any.whl (22 kB)
test  > Collecting smartypants>=1.8.3
test  >   Downloading smartypants-2.0.1-py2.py3-none-any.whl (9.9 kB)
test  > Collecting rtoml<0.10.0,>=0.9.0
test  >   Downloading rtoml-0.9.0-cp38-cp38-musllinux_1_1_aarch64.whl (846 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 846.2/846.2 kB 503.7 kB/s eta 0:00:00
test  > Collecting webassets<3.0,>=2.0
test  >   Downloading webassets-2.0-py3-none-any.whl (142 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 142.9/142.9 kB 551.8 kB/s eta 0:00:00
test  > Collecting zipp>=0.5
test  >   Downloading zipp-3.16.2-py3-none-any.whl (7.2 kB)
test  > Collecting MarkupSafe>=2.0
test  >   Downloading MarkupSafe-2.1.3-cp38-cp38-musllinux_1_1_aarch64.whl (30 kB)
test  > Collecting six>=1.5
test  >   Downloading six-1.16.0-py2.py3-none-any.whl (11 kB)
test  > Collecting markdown-it-py>=2.2.0
test  >   Downloading markdown_it_py-3.0.0-py3-none-any.whl (87 kB)
test  >      ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 87.5/87.5 kB 561.7 kB/s eta 0:00:00
test  > Collecting typing-extensions<5.0,>=4.0.0
test  >   Downloading typing_extensions-4.7.1-py3-none-any.whl (33 kB)
test  > Collecting mdurl~=0.1
test  >   Downloading mdurl-0.1.2-py3-none-any.whl (10.0 kB)
test  > Building wheels for collected packages: typogrify, libsass
test  >   Building wheel for typogrify (setup.py): started
test  >   Building wheel for typogrify (setup.py): finished with status 'done'
test  >   Created wheel for typogrify: filename=typogrify-2.0.7-py2.py3-none-any.whl size=13452 sha256=4ce329903e807671102eab7fd2bc49765b6efc3a4ae68c82053318b62789083c
test  >   Stored in directory: /root/.cache/pip/wheels/0b/e9/98/c888501e8dd2166da059e4f8418694de9b50b48a7192712be9
test  >   Building wheel for libsass (setup.py): started
test  >   Building wheel for libsass (setup.py): still running...
test  >   Building wheel for libsass (setup.py): finished with status 'done'
test  >   Created wheel for libsass: filename=libsass-0.22.0-cp38-abi3-linux_aarch64.whl size=13710320 sha256=3dcb4ce97c1aafc179a6343e0f312c17df88e56c4eb647ab54b09ead5ee00b92
test  >   Stored in directory: /root/.cache/pip/wheels/95/64/fa/47638d5037df216387cdc168e9871d5d9851fc995d636bd108
test  > Successfully built typogrify libsass
test  > Installing collected packages: webassets, smartypants, pytz, zipp, unidecode, typogrify, typing-extensions, six, rtoml, pygments, mdurl, MarkupSafe, libsass, feedgenerator, docutils, blinker, python-dateutil, markdown-it-py, jinja2, importlib-metadata, rich, markdown, pelican, pelican-webassets, pelican-search, pelican-neighbors
test  > Successfully installed MarkupSafe-2.1.3 blinker-1.6.2 docutils-0.20.1 feedgenerator-2.1.0 importlib-metadata-6.8.0 jinja2-3.1.2 libsass-0.22.0 markdown-3.4.4 markdown-it-py-3.0.0 mdurl-0.1.2 pelican-4.8.0 pelican-neighbors-1.2.0 pelican-search-1.1.0 pelican-webassets-2.0.0 pygments-2.16.1 python-dateutil-2.8.2 pytz-2023.3 rich-13.5.2 rtoml-0.9.0 six-1.16.0 smartypants-2.0.1 typing-extensions-4.7.1 typogrify-2.0.7 unidecode-1.3.6 webassets-2.0 zipp-3.16.2
test  > WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv
test  >
test  > [notice] A new release of pip is available: 23.0.1 -> 23.2.1
test  > [notice] To update, run: pip install --upgrade pip
test  $ make publish
test  > "pelican" "/gcl-builds/content" -o "/gcl-builds/public" -s "/gcl-builds/publishconf.py"
test  > Done: Processed 5 articles, 0 drafts, 0 hidden articles, 2 pages, 0 hidden pages
test  > and 0 draft pages in 0.50 seconds.
test  finished in 6 min
PASS  test

ArchLinux et mise à jour du keyring

2023-08-18T00:00:00+02:00

Pour les mises à jour Arch, j’utilise yay. Je ne fais les mises à jour que de manière semi-régulière, et parfois après une longue période je me retrouve avec des soucis de clé qui ne sont plus à jour ou manquantes.

Avec une utilisation fréquente du système, aucun problème ne se pose car un service s’occupe de faire la mise à jour des clés de manière automatique.

Pour résoudre le souci, il suffit de mettre à jour le paquet archlinux-keyring, comme décrit dans la page Wiki qui va bien.

sudo pacman -S archlinux-keyring

Python packaging with Hatch, pipx and Zsh environment variables

2023-08-17T00:00:00+02:00

It’s been a while I didn’t packaged something new. I recently remembered an old package of mine that needed some attention : debts. It’s now time to package it, so I discovered hatch

hatch new —init

This does the heavy-lifting for you, actually porting the setup.py files to the new way of packaging with python (with a pyproject.toml file)

Then hatch shell will create a development environment, install dependencies, check the pyproject.toml file in one command, and give you a shell to test whatever you need to test.

Isolating system packages

I discovered that pipx is a convenient way to install user-facing applications on my system. I use multiple virtual environments for my different projects, but not for the install that are used system-wide.

pipx seems to solve this, and avoid using sudo pip install x.

Manipulating env variables with Zsh

I use Zsh as my main shell for years, and I just discovered that it’s possible to manipulate environment variables in an easy way.

If you’re like me, you never remember how to add something to your path. You can actually use +=, like this:

path+=('/Users/alexis/.local/bin')
export PATH

Profiling and speeding up Django and Pytest

2023-08-16T00:00:00+02:00

Éloi made a pull request on IHateMoney to speedup the tests, with some great tooling for pytest that I wasn’t aware of:

pytest-xdist allows to run tests in parallel, using -n auto
pytest-profiling makes it easy to get the call stack and time the function calls that take most of the time.
You can them analyse the .prof files with Snakeviz

So, I spent some time using these on the tests for La Chariotte, because they were slow.

I found two things :

Login calls are costly in the test, and it’s possible to speed things up ;
On my machine, calls to resolve my hostname were slow, using 5s during the tests for a lookup that wasn’t even useful.

Changing the hashing algorithm to speedup tests

By default, Django uses a slow (but secure !) hashing mechanism for checking the user credentials. In the tests, we don’t need this security, but we need the speed.

Changing them to use MD5 turns out to be a way to greatly speed them up! Here is how to do it with a pytest fixture :

@pytest.fixture(autouse=True)
def password_hasher_setup(settings):
    # Use a weaker password hasher during tests, for speed
    settings.PASSWORD_HASHERS = [
        "django.contrib.auth.hashers.MD5PasswordHasher",
    ]

Speeding DNS lookups

I’m currently using a MacOSX machine, and for for whatever reason, the local lookup was not configured properly on my machine. I don’t think I did anything specific to get this wrong, so it might be your case too. Calls to resolve the local domain were tooking 5s.

If the answer to scutil --get LocalHostName, hostname and scutil --get HostName differ, then you might be in this case. Here is the fix :

sudo scutil --set HostName <YourHostName>