Consider collapsing multiple repeated type-related error messages when in --pretty

[JoshuaKGoldberg]

Search Terms: error message pretty collapsing

The TS team started a few conversations in TSConf 2018 about making TypeScript error messages more friendly to newcomers. One good first one to approach could be repeated, near-identical errors around the same type.

Code

For example, if a type is missing multiple times in the same file:

let a: MissingType;
let b: MissingType;
// ..
let c: MissingType;

Expected behavior:

In a debatably perfect world, TypeScript could give a single error message for the repeated class of error:

src/index.ts Error TS2304: Cannot find name 'MissingType'.

1   let a: MissingType;
           ~~~~~~~~~~~
2   let b: MissingType;
           ~~~~~~~~~~~
4   let c: MissingType;
           ~~~~~~~~~~~

Actual behavior:

Three separate errors.

Perhaps a post-processing step in --pretty mode that receives all of a file's errors and smooshes them down when it can?

Some open questions:

• Which classes of errors would be in this bucket?
  • Example: TS2304 missing type if the type is the same
  • Example: TS2339 property does not exist if the type and property are the same
• How should this look?
  • As suggested above, which is similar to existing --pretty output but with line+column removed?
  • If there are a many of the same error, should the output truncate with a summary?

[DanielRosenwasser]

and smooshes them down when it can?

and smooshes them down when it can?

and smooshes them down when it can?

and smooshes them down

and smooshes

[DanielRosenwasser]

#10489 and #22789 both discuss the idea of related spans which could be leveraged here, but you really want to cut down on these. We could defer all "cannot find" error messages to the end of the checking process and potentially de-dupe in that fashion.

The problem is that you really want to still present the related spans outside of pretty, so we'd need to report a different error depending on pretty which feels bad-ish.

[RyanCavanaugh]

Revisit once we have general multi-location diagnostic support.

Some notes in #23444

[DanielRosenwasser]

I think this should also cover Cannot redeclare block-scoped variable '{0}'

[mhegazy]

We have had a discussion about this offline. An observation here is that usually the duplicate declaration errors are caused by including the same module/declaration file more than once. so the result is multiple duplicate declaration errors for the same 2 files. A proposal for improvement here is if we notice a single file has more than X (say 5) duplicate declaration errors with the same file, then only report the first one with an elaboration saying that there are multiple errors there. #25324 track this suggestion.

[DanielRosenwasser]

Also, related to this is #15550

[mhegazy]

with #25324, I think we can call the underlying issue addressed.