зеркало из https://github.com/mozilla/gecko-dev.git
7e9c9d9319
`NS_FRAME_IS_DIRTY` being set on a frame has always meant that that frame and all of its descendants need to be reflowed. What the main patch on this bug is changing is when it gets propagated to descendants; prior to that patch it gets propagated during reflow, whereas after the patch it gets propagated when the bit is set. This means that `NS_FRAME_IS_DIRTY` is now (after the patch) a somewhat more reliable indicator of whether a frame requires reflow than it was before. However, this has a strange interaction with the particular sequence of operations that test_list.html performs. In particular, test_list.html: (1) makes a style change that requires reflow, (2) flushes style (but not layout), and (3) gets the name of an accessible whose layout was dirtied by the style change. Getting this name ends up here: ``` nsTextFrame::GetRenderedText at layout/generic/nsTextFrame.cpp:9600 nsTextEquivUtils::AppendTextEquivFromTextContent at accessible/base/nsTextEquivUtils.cpp:127 nsTextEquivUtils::AppendFromAccessible at accessible/base/nsTextEquivUtils.cpp:174 nsTextEquivUtils::AppendFromAccessibleChildren at accessible/base/nsTextEquivUtils.cpp:162 nsTextEquivUtils::GetNameFromSubtree at accessible/base/nsTextEquivUtils.cpp:39 mozilla::a11y::Accessible::NativeName at accessible/generic/Accessible.cpp:1991 mozilla::a11y::HyperTextAccessible::NativeName at accessible/generic/HyperTextAccessible.cpp:1760 mozilla::a11y::Accessible::Name at accessible/generic/Accessible.cpp:149 mozilla::a11y::xpcAccessible::GetName at accessible/xpcom/xpcAccessible.cpp:245 NS_InvokeByIndex at xpcom/reflect/xptcall/md/unix/xptcinvoke_asm_x86_64_unix.S:106 ``` What the main patch in the bug has changed is whether this `nsTextFrame` has the `NS_FRAME_IS_DIRTY` bit set at this point. This means that `GetRenderedText` will bail out without reporting the text, because the frame is marked dirty. On the assumption that this action -- getting accessible names while frames are dirty -- is an artificial situation set up in the test and not a real-world situation that we need to care about, I'm proposing that we make the straightforward adjustment to the test: flush layout in addition to flushing style at step (2) above, so that the test works again. As long as that assumption is correct, I think this should be fine. (I'd note that this is the *only* test that breaks as a result of the patch.) If, on the other hand, we actually do care about what accessible names return while layout state is dirty and reflow is needed, we should probably improve `nsTextFrame::GetRenderedText` so that it can somehow report useful state even when `NS_FRAME_IS_DIRTY` is set, or something else more complicated. Differential Revision: https://phabricator.services.mozilla.com/D36423 --HG-- extra : moz-landing-system : lando |
||
|---|---|---|
| .. | ||
| browser | ||
| crashtests | ||
| mochitest | ||