Multiline JSDoc types don't skip asterisks

In a JS file, this typedef doesn't work

/**
 * @typedef {{ 
 *     krazy: 'chad', 
 *     and: 'his',
 *     tag: 'team'
 * }} 
 * Show */

Expected behavior:
No parse errors, and the type is the same as the single-line version

Actual behavior:
Lots of parse errors.

The problem is that JSDoc type parsing uses the Typescript parser, which does not expect asterisks.

Brian Terlson (@bterlson)

Brian Terlson (@bterlson), who found this bug, suggests just skipping asterisks everywhere in the type parser which might work if we really never think that types will have asterisks in them.

looks like a duplicate of #23517

The symptom may be the same, but the cause is different #23517 is failure to skip asterisks inside (unrecognised) tag text, whereas this bug is a failure to skip asterisks inside a type. The fix will be different since (1) whitespace is still important in the text case (2) the jsdoc lexer and parser are used for the text case. Neither is true for the type parser.

See also #26067. As mentioned there, it seems like it would be easier to group the asterisk together with the newline/whitespace trivia while scanning (and set a scanner flag about the linebreak).

Nathan Shively-Sanders (@sandersn)

It feels like this could be implemented with a relatively minor change - if it were possible to differentiate between scanning a comment and scanning code:

diff --git a/src/compiler/scanner.ts b/src/compiler/scanner.ts
index 61ce8330bf..71541f356c 100644
--- a/src/compiler/scanner.ts
+++ b/src/compiler/scanner.ts
@@ -1447,6 +1447,10 @@ namespace ts {
                             return pos += 2, token = SyntaxKind.AsteriskAsteriskToken;
                         }
                         pos++;
+                        if (skipTrivia && (tokenFlags & TokenFlags.PrecedingLineBreak)) {
+                            // decoration at the start of a JSDoc comment line
+                            continue;
+                        }
                         return token = SyntaxKind.AsteriskToken;
                     case CharacterCodes.plus:
                         if (text.charCodeAt(pos + 1) === CharacterCodes.plus) {

This change fixes comment parsing but breaks code parsing (generator methods and any other multiline expression where the line starts with a *). It may also break the indent handling when parsing comments.

So far I don't see an easy way to skip * after newline only while parsing JSDoc types - it feels like there are too many branches of the type parsing to cover.

Nathan Shively-Sanders (@sandersn) - any thoughts on how to provide the scanner with more context (similar to skipTrivia)? Or any other ideas?

In case the "provide the scanner with more context" idea sounds like a good one, I've pushed up a branch that makes things work with multi-line JSDoc types in #26528.

All tests are now passing on #26528. I've also added cases to demonstrate that #26846 is fixed by the same.

Navigation Menu

Repository navigation

Multiline JSDoc types don't skip asterisks #23667

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Multiline JSDoc types don't skip asterisks #23667

Description

Activity

sandersn commented on Apr 24, 2018

mhegazy commented on Apr 25, 2018

sandersn commented on Apr 26, 2018

tschaub commented on Aug 6, 2018

tschaub commented on Aug 18, 2018

tschaub commented on Aug 18, 2018

tschaub commented on Sep 2, 2018

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions