Last modified: 2014-08-28 20:16:22 UTC
If a user uses something like > window.windowDotSomething = true; > justSomething = true; to declare a few variables they will be global variables, and if the user uses > var varSomething = true; in a ResourceLoader-enabled gadget or in their new global.js provided by the GlobalCssJs extension, this will create a local variable (they are inside of a mw.loader.implement wrapper). This is the expected behavior. However > var varSomething = true; results in a *global* variable if inserted in user subpages like [[Special:MyPage/common.js]] or [[MediaWiki:Common.js]]: > window.varSomething true Using "var" should not create a global variable in any case, and users should not rely in the current behavior when writing their scripts, because this usually causes bugs when porting user scripts to gadgets, due to the need to replace lots of "var example"s with explicitly declared globals like "window.example". TLDR: drop support for executing page.js scripts in the global scope.
This will break badly written user scripts and the users will blame us. :) Doesn't mean that it shouldn't be done, but it probably should be part of a larger cleanup action. Also related: bug 33836, bug 33837.
Indeed. If this is ever implemented, there should be more than one announcement in as many channels as possible or something like that, and giving users a good amount of time to (at least ask how to) fix any scripts they care about. And MediaWiki could also get this implemented under a $wgWrapScripts flag which could be set to false on WMF initially but could be true for MediaWiki software (for which there are release notes).
Removing blocker of tracking bug 9968 as this has nothing to do with the phasing out of MediaWiki global variables (specifically PHP, but those of JS as well). This bug doesn't relate to the exposure or support of global interfaces. It just relates to a script executing in the global scope and thus producing local variables as globals. Re-filing as low priority enhancement because them executing in the global scope right now is that way for legacy reasons and very much intentional, not a bug.
