[RFC] Introduce infiniteWidth document

[vjeux]

For some reason I don't understand, breakParent doesn't have the intended effect. But having a document that always puts the remaining with to < 0 does.

I'm pretty sure that we should actually fix the issue with breakParent but wanted to show a solution that would work.

Fixes #625
Fixes #616

[jlongster]

That's strange, I would like to look into this more because it feels like it's working around a bug. Does it involve conditionalGroup? Remember that breaks are not propagated across them, so you need to use willBreak and manually manage breaking.

[vjeux]

Just rebased it

[jlongster]

Thanks. I've been very busy recently so sorry I haven't had a chance to look at this yet. Tomorrow is pretty busy as well but I'll try to find time for this. (Usually if I have time I skim through all the other issues/prs :p)

[jlongster]

I took a look at this. I don't know the solution yet, but here's what I found:

• The problem is that we forcibly break objects if they were expanded in the original source. That causes these kinds of problems, because if a source has a collapsed object and then we naturally expand it, running prettier on it again will see that it's expanded and forcibly expand it. In case Add testing #1, the code does not see any hard breaks, so any checks like willBreak will be false, but in case Nuclide run-through #2, it does see hard breaks, so willBreak and friends will be true. This changes how various code interacts and deals with breaks.

I still wish we didn't keep objects expanded. :)

• Your PR happens to fix it, but not for the reason you think. This simple patch makes the problem go away:

diff --git a/src/printer.js b/src/printer.js
index 3a9e6e8..171670c 100644
--- a/src/printer.js
+++ b/src/printer.js
@@ -711,18 +711,17 @@ function genericPrintNoParens(path, options, print) {
             ifBreak(
               canHaveTrailingComma && shouldPrintComma(options) ? "," : ""
             ),
             indent(
               parentIsUnionTypeAnnotation ? 2 : 0,
               concat([options.bracketSpacing ? line : softline, rightBrace])
             ),
             path.call(print, "typeAnnotation")
-          ]),
-          { shouldBreak }
+          ])
         );
       }
 
     case "PropertyPattern":
       return concat([
         path.call(print, "key"),
         ": ",
         path.call(print, "pattern")

But that removes the feature of breaking an object if anything inside it is broken, so we obviously can't do that. Your code happens to remove this so it "works", but you don't need infiniteWidth at all.

This points to why this is happening: the shouldBreak is false in the first pass, and true in the second pass.

I don't know the right solution yet. I'll come back to this tomorrow!

[jlongster]

Overtaken by #1032