🤖 User test baselines have changed

[typescript-bot]

Please review the diff and merge if no changes are unexpected.
You can view the build log here.

cc @weswigham @sandersn @RyanCavanaugh

[sandersn]

Looks like a regression. I'll have to look at what got merged yesterday.

[sandersn]

Seems like a mistake: updateBadgeFn is initialised with function() { }, then called twice as updateBadgeFn() and once as updateBadgeFn(url). The 1-parameter call used to be an error but no longer is.

    window.runLighthouseForConnection=function(
        connection,url,options,categoryIDs,
        updateBadgeFn=function(){}){
        const config=options&&options.fastMode?new Config(fastConfig):new Config({
            extends:'lighthouse:default',
            settings:{onlyCategories:categoryIDs}});



        const runOptions=Object.assign({},options,{url,config});
        updateBadgeFn(url); // Here!

        return Runner.run(connection,runOptions).
            then(result=>{
                updateBadgeFn();
                return result;
            }).
            catch(err=>{
                updateBadgeFn();
                throw err;
            });
    };

[sandersn]

Very likely a result of #29576. @ahejlsberg or @weswigham mind taking a look?

Note that I just cut-and-paste the snippet from lighthouse-background.js. I didn't confirm that the error repros standalone, but it seems likely.

[ahejlsberg]

Best I can tell this change is to be expected with #29576. Assuming there is no known type for window.runLighthouseForConnection, the contextual type for the function expression is any. This means we give type any to updateBadgeFn (and issue an implicit any error if that is enabled). Previously we'd use the type inferred from the initializer (and we'd then get confused during type inference, but that doesn't apply here).

[sandersn]

Ah, I see now. There should be no contextual type, because this isn't a normal assignment, it's a property-assignment declaration, just one that we don't recognise. With no contextual type, there would be no contextual type and we'd go back to the initialiser type.

We discussed whether we should recognise window.x = function() { ... } in JS just like the currently supported exports.x = function() { ... } and this.x = function() { ... }. I think the answer is yes; at least chrome-devtools-frontend uses it frequently.

So the fix is to keep #29576 and start treating window.x = function() { ... } as a declaration instead of an assignment to an undefined property.