GitHub Pages + Jekyll wins for me on one axis the others can’t match: full ownership of the site structure, source files, and publishing workflow. The repo is the blog. Version control, diffs, PRs, and deploys are the same workflow I already use every day.
Jekyll is a simple, blog-aware, static site generator perfect for personal, project, or organization sites. Think of it like a file-based CMS, without all the complexity. — jekyll/jekyll
It’s Ruby-based and tightly integrated with GitHub Pages — push to main, GitHub builds and serves the site. No CI to wire up, no server to maintain.
I’ll keep writing here, directly in this repo. I also want to migrate some old notes over, which will take time. And I’m tempted to try a GitHub Pages + Astro setup at some point, just to see how it compares. We’ll see.
Stay tuned.
]]>Our MVP setup looked clean on paper: a Presenter per screen, a View interface, a Model behind a use case. In practice, three things kept biting us.
view.showLoading(), view.showError(),
view.hideError(), view.showContent() — the Presenter ended up driving the UI step
by step instead of describing it.By the third or fourth time we wrote the same “restore state after rotation” code, we agreed it was time for something else.
MVI flips the relationship. Instead of the Presenter calling methods on the View, the View renders a single state object, and that state can only be changed by emitting an intent. Three rules:
State data class.(State, Intent) -> State function — same idea as a Redux
reducer.The result is a screen that’s basically a function: state -> UI. Bugs become
reproducible because the state is one value, not a constellation.
Roughly:
data class SearchState(
val query: String = "",
val items: List<Item> = emptyList(),
val loading: Boolean = false,
val error: Throwable? = null,
)
sealed class SearchIntent {
data class QueryChanged(val q: String) : SearchIntent()
object Submit : SearchIntent()
data class ResultsLoaded(val items: List<Item>) : SearchIntent()
data class Failed(val cause: Throwable) : SearchIntent()
}
fun reduce(state: SearchState, intent: SearchIntent): SearchState = when (intent) {
is QueryChanged -> state.copy(query = intent.q)
is Submit -> state.copy(loading = true, error = null)
is ResultsLoaded -> state.copy(loading = false, items = intent.items)
is Failed -> state.copy(loading = false, error = intent.cause)
}
The screen subscribes to a stream of SearchState (we use RxJava today; we’re
watching Kotlin Flow, which stabilized this fall) and re-renders on every emission.
Side effects — the actual network call triggered by Submit — live in a separate
“effect handler” that maps an intent to an observable of follow-up intents.
That’s the whole pattern. There’s no per-method dance with the View anymore.
]]>I want to mark a special thank you to M. Christian Gendreault, my very first boss. He gave me a chance, trusted a young engineer fresh out of school, and showed me what a good director and a good adviser looks like. Three years under his wing taught me more than I realized at the time.
I also want to thank my project lead Jean-Christophe, whose work and guidance I particularly admired, and every colleague who worked alongside me and put up with me (in alphabetical order, lul): Bénédicte, Charlotte, Charline, Dimitri, Flora, Jules, Mikaël, Paul, Sébastien, and Vincent.
You all made those three years in Castres genuinely good ones.
J’ai passé trois années avec vous, trois années bien agréables. Vous avez été très gentils avec moi et vous m’avez beaucoup aidé.
Merci beaucoup à M. Christian Gendreault qui a été pour moi un bon directeur, un bon conseiller.
J’ai particulièrement apprécié le travail de mon chef de projet Jean-Christophe et je le remercie aussi.
Je remercie aussi les collaborateurs qui ont travaillé avec moi et qui m’ont supporté : Bénédicte, Charlotte, Charline, Flora, Dimitri, Mikael, Jules, Paul, Vincent, Sébastien (l’ordre alphabétique, lul) !!!
Maintenant, je me plais bien à Paris, même si c’est différent de Castres.
Je vous souhaite à tous une très bonne année 2019.
Paris was a big change from Castres — bigger pace, bigger noise, bigger distances. But the foundations I built at Exyzt traveled with me. To everyone there: thank you, and a belated bonne année — every year since.
]]>Business rules in the center, frameworks on the outside, dependencies always pointing inward. The database, the UI, the network, the web framework — all of them are details, plugged in from the edges. The core of the app shouldn’t know they exist.
That one rule does a surprising amount of work:
Read like that, half of the bugs we keep fighting stop looking like bugs and start looking like the consequence of letting frameworks reach into the core. We’re paying a tax we never agreed to.
Because it sounds, at first, like a rewrite. And rewrites are scary, expensive, and usually a bad idea on a project that’s already shipping.
Here’s the part I want to promote: Clean Architecture is not a rewrite. It’s a direction. You don’t need a green field. You don’t need a six-month freeze. You need to know which way is “in” and start pointing your new arrows that way.
I’m writing this because I want the team to read the original article. Not my summary — the source. Half an hour of reading, and the way you look at our code afterward is different.
A big rewrite is not always the answer. A direction is. Uncle Bob gave us one. Let’s start walking.
]]>Our field devices run plain AOSP — no Google services, no Play Store, just the stock open-source build. That means the only keyboard out of the box is the basic AOSP LatinIME. It’s fine for casual texting; it’s not fine for our work.
A small keyboard app that fits the work. Agents type so much faster now.
]]>