Overview of misalignments with Babel

[armano2]

There are still few nodes which are misaligned and we have to decide if we want to follow babel on it.

Biggest one are:

• TSIndexSignature
  • typescript is able to produce ast for more than one parameter and provides array of arguments even if only one is valid
  • babel ast respects that and provides array of arguments with only one node allowed
  • we are converting it to "index" and we are reading only first one even if typescript provides us more
• Heritage in classes and interfaces (TSExpressionWithTypeArguments)
  • babel has one node type for Heritage for implements in classes and extends in interfaces
    • implements in classes and extends in interfaces is optional in babel and its named implements and extends
  • we are converting it to 2 separated nodes TSInterfaceHeritage and ClassImplements and our node is always present
    • in interfaces node name is implements and in classes its heritage

i added test cases for them in #110

Alignment test override for heritage

if we want to keep our nodes we should add this to as processing in babel. ```ts ClassDeclaration(node: any) { if (node.abstract) { node.type = 'TSAbstractClassDeclaration'; delete node.abstract; } }, TSInterfaceDeclaration(node: any) { if (node.extends) { node.heritage = node.extends; delete node.extends; } else { node.heritage = []; } }, TSExpressionWithTypeArguments(node: any, parent: any) { if (parent.type === 'TSInterfaceDeclaration') { node.type = 'TSInterfaceHeritage'; } else if ( parent.type === 'ClassExpression' || parent.type === 'ClassDeclaration' ) { node.type = 'ClassImplements'; } if (node.expression) { node.id = node.expression; delete node.expression; } } ```

---

what we should do with this... those are breaking changes, and i'd like to hear you opinion. @JamesHenry @j-f1 @ikatyang @uniqueiniquity

[armano2]

do we want to align those nodes?

[JamesHenry]

For TSIndexSignature do you know if there is a diagnostic which provides functionality similar to the babel check?

[armano2]

Yes there is and we are reporting it already, TS1096 "An index signature must have exactly one parameter."

[JamesHenry]

Ok cool, sounds like the alignment there should be pretty straightforward then 👍

I don't have any strong opinions on the heritage side of things, so probably makes sense to continue the trend of matching babel unless there are obvious flaws in its AST (which obviously sometimes there is)

[armano2]

ok, than i'm going to prepare PRs for that at evening, unless you want to take one of them 🐱

[JamesHenry]

I won't be able to until at least 7pm EST - but very happy to do it then

[armano2]

ok, than i will take TSIndexSignature :)

[JamesHenry]

Cool, I’ll do heritage later, thanks! On Fri, 11 Jan 2019 at 11:09, Armano notifications@github.com wrote:

ok, than i will take TSIndexSignature :)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/JamesHenry/typescript-estree/issues/112#issuecomment-453567853, or mute the thread https://github.com/notifications/unsubscribe-auth/AA29q7EwAn0wLZyVHYG8cRDc421OwWRrks5vCLckgaJpZM4Z5r0E .

[armano2]

there is one more broken part of conversion, Pattern vs Statement.

in my test project, ~600 files fails due to that :>

i was thinking about changing way how we determine it and follow on that estree spec... right now we are checking parents, and parents of parents.... and so on...

rather than that we should pass down information that we are processing pattern or statement while calling convertChild..

[armano2]

but we are on good path, between versions 13.5.2 and 15.0.0 we went down with failed tests from:

to

btw. i tried changing mocha to jest and looks like its way way slower xd

[JamesHenry]

Nice, thanks for maintaining that extra test suite!

[JamesHenry]

We can do an updated summary of the state of the two ASTs soon 👍

[armano2]

that's new results

[JamesHenry]

Awesome!