https://kotlinlang.org logo
Join Slack
Powered by
The irrelevance of IDE auto-completion is staggeri...
# intellij
l
The irrelevance of IDE auto-completion is staggering… I'm shocked by how it ignores the parameter names and most obvious types, i.e. the most basic hints. (Android Studio 2025.1.3) It's actually annoying now. Such a far cry from how Java completion was working very well even 10 years ago. I guess it's because IDE's K2 mode was pushed before actual feature/behavior parity, and I don't like it 😞
💯 12
id
as first result for the parameter 
email
? Really?
Do you want to replace 
TODO()
with 
TODO()
?
stability and code analysis improvements.
Hum…
o
I can relate. Things could be better. I hope that this turns out to be a price worth paying for getting K2 out and stable, opening up Kotlin's evolution for its next steps. But yes, I really hope that auto-completion will catch up soon.
e
l
P3 💀
o
On the JB side it's Major.
👍🏼 1
w
It's not only K2, but also the ML completions I think. At least I remember declining quality around the time ML completion was introduces. I also feel that sometimes the initial completion is ok-ish but then more options show up, rearrange the initial ones and they're not helpful anymore. I also hate that parameter names are not prioritied
And this really comes up regularly — both reduced quality of Kotlin completions and K2 feature parity 😕 I got a huge amount of notifications for tickets closed around K2 inspections so hopefully it'll get better soon™
t
Hey everyone! Thanks for bringing attention to this problem. We are continuously working on improving completion, honestly, with every release. Completion is a big and complex system, so moving it from K1 mode to K2 isn’t something we can just “lift and shift.” The K2 completion is an almost complete rewrite (which was necessary due to the new compiler), and that’s why there is a drop in the quality currently, but it is improving steadily and will continue to improve. I completely understand how frustrating this feels in daily work. I hope you will feel the difference in the next release already. With Android Studio, it’s a bit trickier because of the version gap, which means completion improvements land there a bit later than in IDEA. By the way, do you also notice similar issues with full-line completion?
BTW, every screenshot and code snippet you post is very useful because we can easily reproduce them and check our completion improvements against them. We are grateful for them, please add more when you notice them.
Do you want to replace 
TODO()
with 
TODO()
?
And many of them are really funny and silly too
👍🏼 1
👍 1
👍🏾 1
l
full-line completion isn't supported in Android Studio, I guess for commercial reasons? Last time I used it in IDEA, it was working well maybe 2/3 of the time, but the last 1/3 it was distracting me with mistake-full suggestions, and it was breaking my flow, not fun, so I disabled it, preferring to slow down a little, but keep my focus. I didn't like the flipped boolean expressions and wrong pattern guesses.
For full-line completion, I wish it was available with a special shortcut to invoke it, for the times where I feel lazy and want a little help.
Noted for the screenshots, I will keep posting them in the issue (https://youtrack.jetbrains.com/issue/KTIJ-32954/)
thank you color 1
t
Hello! Have you tried completion in the new 2025.3 release? How do you find it? Is it better performance or quality-wise? We have introduced parallel completion, and our tests indicate improved performance. We are continuing to make it better, but would appreciate your feedback.
w
Have you tried completion in the new 2025.3 release?
Yes (since EAP)
How do you find it?
not much better, quality-wise 😞 named parameters being the biggest offender, but there's a lot more of unhelpful suggestions (compose functions in non-compose context, duplicated suggestions for data objects where one inserts just type name and the other also parentheses, suggestions not matching parameter names, type-matching completion still wonky with nested types) and other stuff around ergonomics (like overload resolution not being helpful when a single parameter type doesn't match)
Is it better performance or quality-wise?
quality-wise I noticed that type-matching completion started working every once in a while (for enums I think), otherwise not really. Performance-wise difficult to say as a user obviously. I feel it got a better with recent EAPs and then went down again slightly with the final release. But also I've been tweaking settings to disable more ML stuff and AI plugins to see if it helped. Anyway I don't feel performance is an issue with 2025.3 if that helps (I have ML-assisted completion off, full-line completion off, and disabled jetbrains ai assistant)
I want to add that I'm not discounting all the improvements being made and appreciate all the work being put into Kotlin ecosystem. It's just that as a user I still agree with the original statements that > The irrelevance of IDE auto-completion is staggering… and > It's actually annoying now. I'm a simple man, I mostly just want to write part of the code and press 
enter
to complete it, and that simply doesn't work. Btw I'm genuinely curious why this feedback doesn't come from JB engineers in the first place, since they must be working on tons of Kotlin code themselves.
❤️ 1
👍🏼 1
👍 1
t
I believe the problem is that we have been dogfooding K2 since 2023, and the completion back then was in a much worse state than it is now, and we had to accept it because it’s what we are working on. I’d say some of us have just grown accustomed to the current quality, which is the main reason. But when we notice problems, we fix them. And sometimes, different technologies could contribute to that, like for Spring, we have Spring completion contributor on top of the regular one, which is affecting the result, and we couldn’t dogfood as extensively, so this is where your feedback comes in. We plan to target quality completion in this release, both for regular completion and type-matching completion.
26 Views
Open in Slack
PreviousNext
Powered by