1228542 - Multiple tests cause ASSERTIONS when system add-ons import AUTHOR_SHEET style sheets (ASSERTION: Style set already has document sheets?: 'aStyleSet->SheetCount(SheetType::Doc) == 0')

Status: RESOLVED FIXED

(Core :: CSS Parsing and Computation, defect)

Description

[Mark Banner (:standard8)]

In bug 1223573 we're trying to move Hello to a system add-on. As part of our code we have:

  let sheets = ["chrome://loop/content/addon/css/loop.css",
                "chrome://loop/skin/platform.css"];
  for (let sheet of sheets) {
    let styleSheetURI = Services.io.newURI(sheet, null, null);
    if (styleSheetService.sheetRegistered(styleSheetURI,
                                          styleSheetService.AUTHOR_SHEET)) {
      styleSheetService.unregisterSheet(styleSheetURI,
                                        styleSheetService.AUTHOR_SHEET);
    }
  }

However, when we run it through try, there is a debug assertion raised on lots of different tests:

https://treeherder.mozilla.org/#/jobs?repo=try&revision=2be8d2c8c797&selectedJob=14172398
https://treeherder.mozilla.org/logviewer.html#?job_id=14172398&repo=try
 14:48:52     INFO -  [1191] ###!!! ASSERTION: Style set already has document sheets?: 'aStyleSet->SheetCount(SheetType::Doc) == 0', file /builds/slave/try-m64-d-00000000000000000000/build/src/dom/base/nsDocument.cpp, line 2365
 14:49:49     INFO -  #01: nsDocument::ResetStylesheetsToURI(nsIURI*) [xpcom/glue/nsCOMPtr.h:403]
 14:49:49     INFO -  #02: nsDocument::ResetToURI(nsIURI*, nsILoadGroup*, nsIPrincipal*) [mfbt/RefPtr.h:268]
 14:49:49     INFO -  #03: nsHTMLDocument::ResetToURI(nsIURI*, nsILoadGroup*, nsIPrincipal*) [mfbt/RefPtr.h:40]
 14:49:49     INFO -  #04: nsDocument::Reset(nsIChannel*, nsILoadGroup*) [mfbt/RefPtr.h:40]
 14:49:49     INFO -  #05: nsHTMLDocument::Reset(nsIChannel*, nsILoadGroup*) [dom/html/nsHTMLDocument.cpp:234]
 14:49:49     INFO -  #06: nsHTMLDocument::Open(JSContext*, nsAString_internal const&, nsAString_internal const&, mozilla::ErrorResult&) [xpcom/glue/nsCOMPtr.h:721]
 14:49:49     INFO -  #07: mozilla::dom::HTMLDocumentBinding::open [mfbt/AlreadyAddRefed.h:138]
 14:49:49     INFO -  #08: mozilla::dom::GenericBindingMethod(JSContext*, unsigned int, JS::Value*) [dom/bindings/BindingUtils.cpp:2671]
 14:49:49     INFO -  #09: js::CallJSNative(JSContext*, bool (*)(JSContext*, unsigned int, JS::Value*), JS::CallArgs const&) [js/src/jscntxtinlines.h:236]

If we change AUTHOR_SHEET to either USER_SHEET or AGENT_SHEET, then the tests no longer fail, however due to the CSS cascade doing this means we have to add a lot of !important annotations to the majority of our rules to get them to apply correctly.

Comment 1

[Mark Banner (:standard8)]

For now, I've worked around this in attachment 8692901 [details] [diff] [review], but as you can see, its really ugly, and I won't be surprised if it causes us issues at some stage.

Comment 2

[Boris Zbarsky [:bzbarsky]]

Looks like nsDocument::FillStyleSet adds style sheets from the stylesheet service to the style set, but nsDocument::ResetStylesheetsToURI never removes them, right?

Looks like a bug in the initial patch for bug 676054 which no one ever noticed, perhaps because no one ever tried to call document.open() in a debug build with any actual author sheets in the stlylesheet service...

Gabor, do you have bandwidth to look at this?

Comment 3

[Gabor Krizsanits (INACTIVE)]

Sorry hit the wrong "assigned to".

Comment 4

[Gabor Krizsanits (INACTIVE)]

Attachment: resetting AuthorStyleSheets. v1

Thanks for spotting out the problem. I totally overlooked it. The test is not pretty but I didn't have any better idea at the moment... I'm not even sure I want to land that test, just wanted to see if it fixes the problem.

Comment 5

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 8693554 [details] [diff] [review]
resetting AuthorStyleSheets. v1

Nice test!  r=me

Comment 6

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/d9a1efa7d1ab

Comment 7

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/d9a1efa7d1ab

Comment 8

[Mark Banner (:standard8)]

I've just tried removing our workaround, and try shows it as working fine now.

Thank you for fixing this quickly.