Fix start position of intersection/union node

[dhcmrlchtdj]

Fixes #30995

• I am not sure how to write a baseline test for startPos, do we have docs or examples?
• parseOptionalToken use scanner.getStartPos() as start position. What about use scanner.getTokenPos() instead?

[rbuckton]

Can you add a test for this output. It seems like this change would result in missing comments:

type A = 
  /*leading0*/
  | /*leading1*/ 1 /*trailing1*/ 
  | /*leading2*/ 2 /*trailing2*/;

When we emit this declaration today, we include the leading and trailing comments for each type in the union, but do not emit /*leading0*/. I would like to see the test verify that we still preserve the comments around each type and whether this will now also include the leading comment before the first |.

[rbuckton]

Also, I would move this line inside of the if statement that follows it? There is no need to determine startPos if we end up not using it.

[ajafff]

should the NodeArray also include the leading operator? currently only the Union/IntersectionType contains the operator

[ajafff]

• I am not sure how to write a baseline test for startPos, do we have docs or examples?

Have a look at my attempt: #31265

For the test @rbuckton requested:

• add a file to tests/cases/compiler with a comment // @declaration: true at the top and the desired content after that.
• run gulp runtests-parallel (will fail)
• run gulp baseline-accept (and verify the diff) now all tests should pass

• parseOptionalToken use scanner.getStartPos() as start position. What about use scanner.getTokenPos() instead?

In my PR (linked above) I used scanner.getStartPos() to avoid creating a Token object that is never really used.

[mysticatea]

Hi. There is a discussion about how this should be fixed on the original issue #30995. Especially, this PR leaves the leading operator as floating if the second operator didn't exist.

[dhcmrlchtdj]

#31265 by @ajafff is a better solution.
So I will close this PR.

Thanks for your review. @rbuckton @ajafff