As I already mentioned, a very long day. Three talks and random coffee.
The first talk of the day from Zhenya Matsyuk is his classic base on UI tests, such a meta-talk about his and others' talks from other conferences. I've already watched all this, so I didn't take away much new, unfortunately. Perhaps it was interesting to hear about all sorts of cloud services on which these tests can be run without going crazy from building your own infrastructure. Emulator.wtf, Firebase Test Lab, and Marathon Cloud were noted. Well, and I remembered two interesting thoughts: the East, unlike the West, loves to write their own bicycles, and mostly UI tests are written by QA for everyone. Probably my shame for never having written proper UI tests can be reduced a bit, QA is to blame.
Random coffee brought us together with Alexey Khaiminov, who at the round table was responsible for medium projects, and the coincidences turned out quite funny. He's originally from Almaty, I'm now from Almaty by fate's will, so we found common topics. We discussed both local features and Cypriot ones, where he settled, and Android and multiplatform. You can guess what conversations immigrant Android developers have. It was cool to exchange thoughts.
The first evening session was about generating autotests using LLM. Plus minus the most useless talk for me at this Podlodka (seems to be the fate of all talks that mention AI). The talk in one sentence: there are all sorts of ChatGPT and Copilots, they are unavailable to us, unsafe, so don't use them, use local models. Some features, let's call them that, of the Russian market affect the talks too. And I heard almost no test specifics, for the whole talk they showed a couple of calculator-level tests.
Having heard all this, you start to appreciate even more places where they distribute enterprise Copilot and ChatGPT. Even had a bit of a flashback to a world where you can't access the normal internet from your work laptop, brr.
The last session of the day - a workshop on Geminio. Zero questions about the tool, although when I had the idea of generating feature and core modules in the project I just went and wrote my Gradle task, I was missing something from Geminio. More questions about generating 10 classes for a feature on Decompose + MVIKotlin. Some internal beliefs don't allow me to hide this behind a generator yet. It seems that this is too implicit a way to hide architecture complexity instead of directing the developer with all sorts of checks and code not to write crap, which is more important. Creating modules is ok because in the studio it's quite troublesome to do according to some pattern without extra garbage, well, and there are not so many different interpretations in this.
Today is the last day, I'm already tired, honestly.