Code completion suggestions are irrelevant / context unaware

[barthbv]

This tracker is for issues related to:

• Dart analyzer and linter

Summary:

Code completion suggestions are flooded with irrelevant entries and / or are not context aware since (roughly) Dart 3.
I've been noticing a decrease in accuracy of the code suggestions for a while that had the relevant suggestion buried below various references to unused packages or even keywords that would result in wrong syntax. (examples below)

Tested with Dart versions:

• 3.0.5
  to
• 3.3.4 (currently used)
• 3.5.0

Tested in IDEs:

• Android Studio
• VSCode
• IntelliJ Idea

System:
Mac OS 14.4.1 (but the issue was probably noticed before that)
Flutter 3.19.6 (current)
Dart 3.3.4 (bundled)

Examples:

Typing the keyword required for a named parameter up to the first 3 letters suggests the dynamic variables "r"/"re"/"req".
Expected: required first and possible others below (even though i wouldn't think declaring a variable this way would be recommended)

Typing the keyword this inside a class to reference a member points to an object in the dart:html package (this is not a web project, so the linter knows importing it is wrong).
Expected: this and throw first, and the rest, but imports from the dart:html package probably shouldn't be suggested in a non web project

Note: typing only "t" yields better results but still contains far fetched suggestions

Here, the constructor of a class contains a named bool parameter. Trying to type false as a value suggests a load of unrelated non bool objects, and even the factory keyword as the first pick in VSCode (happens on a base flutter project, so the suggestion order is influenced by the "import "package:flutter/material.dart")

Tests i've conducted:

The different sdk versions i've tried all suffer from this to various degrees, 3.0.5 was the one with the most relevant suggestions.
I've not tested with all the individual versions since but i've continuously kept my environment updated and the problem has persisted since.

In Android Studio, flipped every "code completion" parameter available.

In VSCode i tried various combinations of the Dart settings "useLegacyAnalyzerProtocol", "autoImportCompletions", "editor.suggestSelection", "editor.wordBasedSuggestions" and the likes, yielding various results, improving some situations here and there but i could see the issue was not resolved, and the suggestions were nowhere near the accuracy i was used to before.

The different IDEs all suffer from this, with some minor variations, although as expected Android Studio and IntelliJ Idea behave the same.

The issue persisted over 2 computers (albeit both apple) and several clean installations of several systems (probably Mac OS 12 - 14)

The dart sdk used is bundled in flutter since i'm using flutter, but to be sure i've installed it separately and made sure the IDEs used that one instead, same results.

In VSCode, tried setting both the Dart and the Flutter extensions to "pre-release", each separately and both at the same time, same results.

I know some issues are already opened on the subject but i felt they did not fully address my problem so i opened this one, thanks for reading !

[bwilkerson]

Thanks for the feedback!

And we're sorry about the regressions. I'm not sure exactly why they occurred, but they're almost certainly a result of some changes we're making to streamline the implementation of code completion. We take the quality of code completion seriously and will look into the known cases, like the ones above, as soon as possible.

[barthbv]

You're welcome, and for sure don't be sorry ! I didn't report to complain, i reported to inform 👍
Thank YOU for your work and dedication, if i can help with anything just let me know.