|author||Patrick Spek <firstname.lastname@example.org>||2023-08-29 13:18:08 +0200|
|committer||Patrick Spek <email@example.com>||2023-08-29 13:18:08 +0200|
1 files changed, 74 insertions, 0 deletions
diff --git a/content/posts/2023/2023-08-29-releasing-raku-modules-with-fez.md b/content/posts/2023/2023-08-29-releasing-raku-modules-with-fez.md
new file mode 100644
@@ -0,0 +1,74 @@
+title: "Releasing Raku modules with fez"
+Last week I got a message on Matrix, asking me to update one of my
+[`Config::Parser::TOML`](https://git.tyil.nl/raku/config-parser-toml/). One of
+the dependencies had been updated, and the old one is no longer available
+through the module installer `zef`. Its not that big a change, and there are
+tests available, so its a reasonably small fix on itself.
+Recently I've set up [Argo Workflows](https://argoproj.github.io/workflows/) for
+my CI/CD desires, and I found this a good and simple Raku project to try and
+incorporate into a workflow. Since I had some additional quality checks ready to
+use in my workflow, this has resulted in [REUSE](https://reuse.software/)
+compliance for this Raku module, in addition to the regular `prove` tests
+already available in the project. Additionally, the de facto default module
+authoring tool `fez` also brings a few new checks that have been incorporated.
+While all that is good, there were some annoyances I encountered while
+configuring this. Notably, I've found `fez` to be a chore to work with when it
+comes to non-interactive use. All CI/CD jobs run in their own Kubernetes pods,
+and _should_ not require any interaction from myself during these runs. I am
+writing this blog post mainly to write down the annoyances I encountered, hoping
+that `fez` can be improved in the future.
+Lets start with the first issue I encountered while setting up the workflow:
+`zef install fez` fails by default. `zef` gives the advice to `--exclude` one of
+the dependencies, and going by the issues reported on their Github repository,
+this seems to be accepted workaround. However, I'd argue that this workaround
+should not be needed to begin with. Especially seeing as `fez` works fine and I
+have absolutely no clue what this `z` is or how I can supply it. Either drop
+this dependency, or document its use and upstream so people can package it.
+The second issue I encountered was with the `login` functionality of `fez`.
+There seems to be no way to handle this non-interactively. The way around this
+for me has become to use `expect` scripts, but this is obviously not very pretty
+and will break whenever the interactive interface of `fez` changes. A good means
+of non-interactive authentication would be great to have. I've considered to
+just mount `fez`'s config/cache into the containers, but the documentation warns
+that tokens aren't permanent to begin with.
+Next up there's the actual `upload` command. I'm running it twice in my
+workflow, once with `--dry-run` and once with `--force`. The first one is done
+as a preliminary quality check to see if there's any obvious issues that ought
+to be fixed beforehand. I noticed on a subsequent run (the one with `--force`)
+that the _dry_ run isn't all that dry. It leaves an `sdist` directory, which in
+turn will get included in the next step. There's a flag to create this `sdist`
+directory, but no flag to do the inverse. My solution is to end this step with
+`rm -fr -- sdist` to clean it up again.
+And lastly, when all quality assurance checks have passed, the `fez upload
+--force` command is ran on the working directory. I'd rather not force anything
+here, but the alternative is that another interactive question pops up and the
+job hangs forever. I don't know all the possible prompts `fez` can generate, and
+for this one I didn't even bother to try and look that up. Rather than a
+`--force` to practically say "yes" to everything, I'd prefer an option to say
+"no" to everything, failing the pipeline immediately.
+Another pet-peeve of mine is that `fez` seemingly doesn't use exit codes. No
+matter what happens, even something quite important such as `login` with
+incorrect credentials, it _always_ returns `0` as exit code. This should
+obviously be fixed sooner rather than later, as it is quite simple and it is the
+basis for _many_ systems to check the exit code to deduce something is wrong.
+Uploads of module updates are currently working, which is good, but I feel like
+a lot of workaround code I had to write should not be necessary. If `fez` can
+fix these issues, it will be much more of a breeze to use, which in turn
+hopefully encourages more automated testing and distributing of Raku modules.
+This can be a great boon for the module ecosystem and overall community.