applying fixes can create syntactically invalid code.

[bradzacher]

I've been trying to use ts-fix to add annotations for isolatedDeclarations and I've run into a few cases where the fixers are not correctly merged together.

The most obvious example I've seen is when multiple imports need to be added from a module

// before fixing:
import type {
  A as B,
  C,
  D as E,
  F as H,
  H,
  I,
  J,
} from './mod1';

// after fixing -- syntax errors and duplicate imports
import type {
  A as B,
  C,
  D as E,
  F as H,
  H,
  I,
  J,
  H // missing commas
  I // missing commas
  J // missing commas
  K // missing commas
  L // missing commas
  M,
} from './mod1';
import { N } from './mod2';
import { JSX } from 'react/jsx-runtime';
import { O, P } from './mod3';
import { N } from './mod2'; // duplicate import

This of course makes at-scale codemodding difficult because many files will need to be fixed up by hand.

---

It's worth noting that if I manually open the file in VSCode and select the Add all missing type annotations quick fix:

Then the resulting file is syntactically correct.
Which suggests that TS has a built-in mechanism for merging the fixes to produce valid code.

It would be great if we could switch to use this mechanism.

[nikelborm]

I have the same issue as I've described in #43 PR comment

Hi! I'm trying to use ts-fix in my project. I want to enable isolatedDeclarations and automatically add all required inference type information in all places where it was missing.
And I just faced the same problem with ts-fix built from the latest commit.

It outputs redundant imports (19-22, 23-27) that are almost identical (the highlighted line 25 is the only difference between those 2 blocks):

I know in VS Code, there's the "Organize imports" command in the command palette. Do you know by any chance if it calls some simple method inside the TS language service, that can be reused in the codebase of ts-fix?

I don't know how ts-fix works, I just assume and hypothesize (correct me if I'm wrong) how it works by its outputs.

Maybe, if there are many fixes in a row to a certain file, you have to maybe not think about how all fixes can be applied at once, but do it incrementally?

like this:

Is the initial first version of the file bad?
yes.
Apply any available fix for this file.
Is the second version of the file still bad even after the fix?
yes.
Apply the new relative diff fix to the second file version to get the third version.
etc, until "no."...

That way, if one automatic fix added new lines with imports, the next automatic fix will continue from the point where the previous fix left off, adding only necessary import type { OctokitResponse } from '@octokit/types';. Is it easy to implement in this codebase?