| Age | Commit message (Collapse) | Author |
|---|
|
|
|
We only need showLinkText in the hint's local frame, so do not transmit
it to remote frames.
|
|
The "reason" a hint is available is only ever needed in the hint's own
frame, so there's no need to pass it to other frames.
|
|
Previously, we passed each hints rect to every other frame. However,
the rect is only actually needed in its own frame.
Here, we instead keep (and use) each hint's rect locally only.
|
|
Previously, we were rendering hints in all frames, but displaying only
in their own frame
With n frames and m hints, we rendered n*m hints.
With this, hints are only rendered in their own frame, so we only render
m hints.
|
|
We take a copy of the frame Ids and ports at the point which link-hints
mode is launched, and use that subsequently. Therefore, a newly created
frame cannot cause the hint coordinator to become confused.
Also: add debugging code.
|
|
This fixes an error in 39adee9090fc5aadfd5dd681f91b80025084858a.
Specifically, if there are no hints to select from, or no
documentElement, then link-hints mode exits without initiating an actual
mode (i.e. without calling its super() constructor).
With 39adee9090fc5aadfd5dd681f91b80025084858a, that leaves a mode in
place which blocks all keyboard events, thereby rendering Vimium
entirely hung.
See this line:
https://github.com/philc/vimium/commit/39adee9090fc5aadfd5dd681f91b80025084858a#diff-e9abcb9ebcdb5af8e9f33651364673a1R59.
Here:
- we explicitly remove the keyboard-blocking mode
- we add exitOnEscape (so that the situation is at least
recoverable), and
- we add an indicator (so that we can see what's going on).
It is proposed that the indicator is a temporary feature, while the
global link hints feature is in shake down.
|
|
There's no need for the previous complicated approach to UI component
initialialisation, in particular for the Vomnibar.
We only initialise the Vomnibar in the top frame. However, if for some
reason it hasn't been initialised by the time it's needed, then we can
just initialise it then. We are only initialising it early to avoid
flicker, so it's not a correctness issue. And the only reason it
wouldn't be initialised is if Vimium is disabled in the top frame, but
enabled in some other frame -- which is not a common case.
|
|
We need to wait for documentReady() here to ensure that the document is ready in the top frame.
|
|
I've not observed this, but it could possily happen...
If the stableSortCount gets out of sync in different frames, then
different frames might make different decisions about the ordering of
the hints.
Ti avoid this potentially happening, we initialise the stableSortCount
every time link-hints mode is activated.
|
|
Previously, we could select divs, uls, and ols for scrolling, but we
couldn't get back to scrolling the document body.
This makes it possible to select document.body for scrolling.
Unfortunately, sometimes the hint appears in a rather odd place (because
it's "on top of" something else which is clickable.
|
|
There seems to be an issue on this page:
- http://i.imgur.com/PdmUjij.jpg
whereby "DOMContentLoaded" isn't firing. The page structure is unusual
(involving a shadow-DOM element), which might be the source of the
problem.
What happens is that the "DOMContentLoaded" event fires as required, but
the document state is still "loading".
Here, we just say that if the "DOMContentLoaded" handler has fired
once, then we're good to go.
|
|
This affects focusFrame and link hints.
We do not register tiny frames. The reason for doing this is that link
hints messages all (registered) frames to collect their hint
descriptors. On some sites (e.g. Google Inbox and other Google
properties), there are many tiny iframes (on the order of 12 or 15 or
so). Those frames cannot provide useful hints, so -- by not registering
them -- we don't ask them for hints.
The intention is to speed up the link-hints activation sequence.
|
|
The reason for doing this is that we will be using the same test to
decide whether or not to register a frame.
|
|
This is a no-op.
It separates the process of registering a frame with the background page
from the port initialization. The idea is that -- soon -- we will only
register some frames; in particular, we should not register the many
tiny iframes on pages like GMail.
The eventual goal is to speed up the global link-hints initialisation
sequence.
|
|
This seems to be considerably faster than using sendMessage().
|
|
Without considering the size of the data passed, ports seem to be about
5 times fast than sendMessage(), so we use ports of link-hint messages
wherever possible.
|
|
If document.documentElement isn't ready, then we can'r generate hints.
Moreover, this would crash -- thereby hanging global link hints.
|
|
|
|
Following on from f0911e52f0e71c6d2539bdc74a09ff2dbd5ab125, I omitted to
verify that the height of the dialog was correct on taller screens.
|
|
With global link hints, hints might be launched in one frame when the
settings are not yet loaded in another. This could lead to one frame
using one settings value, and another another value. (Because we use
the default value if the settings are not yet loaded.) And this in turn
could cause link hints to crash if filetered hints are used (because, in
that case, we would not calculate hint.linkText in one frame, but then
try to use that value later in another frame).
|
|
|
|
|
|
Problems:
- The meanings of some of the Mode/handlerStack constant names is far
from obvious.
- The same thing is named different things in different places.
This changes various constant names such that:
- the names used in the handler stack and in the modes are the same.
- ditto vis-a-vis DomUtils.
Also, break out the core of the handler stacks' `bubbleEvent` method
into a switch statements. This makes it more obvious that the cases are
mutually exclusive.
|
|
Which:
- uncovered a typo in 39adee9090fc5aadfd5dd681f91b80025084858a.
Also:
- make Mode.debug a class variable, which is more helpful while trying
to debug. Specifically, you can turn debugging on or off from
within the tests, for example.
|
|
Because there is a small amount of time between link-hints mode being
requested and it being activated, it was possible to launch other Vimium
commands (e.g. the Vomnibar) after requesting link-hints mode, and
before link-hints mode starts.
This prevents that.
|
|
The exit sequence is clearer like this.
|
|
|
|
See the newly-added comment.
|
|
|
|
Context: filtered link hints with wait-for-enter enabled. Once a link
is selected, it is highlighted. We then consume all keyboard events
until the user hits enter, at which point the link is activated.
Problem: When we're waiting, the link is highlighted. It looks to the
user like `Escape` should cancel.
This implements escape to cancel at that point in the exit sequence.
|
|
|
|
hintMarkers was previously passed around (within the link-hints mode
class) from function to function. With #2048 (global link hints), it is
better if this becomes @hintMarkers (that is, is available directly to
all methods within the class.
With #2048, we used bind to avoid having to do this - to keep the diff
clearer. Here, we do it directly.
|
|
These variable names are misleading. The things being manipulated are
actually hint descriptors.
So this renames the variables accordingly.
|
|
This code (LocalHints) has been embedded into the middle of the
link-hints mode class. And it shouldn't be.
This moves it out, and allows us to unwind some of the gymnastics #2048
(global link hints) introduced to avoid having an incomprehensible diff.
|
|
Testing the `scrollHeight` is cheaper than testing scrollability. There
are a lot of non-scrollabile divs, so it makes sense to use this cheaper
test as a filter.
|
|
We should start by checking the *parent* of the candidate descendant.
|
|
We recognise elements with a class names containing the text "button" as
clickable. However, often they're not, they're just wrapping a
clickable thing, like a real button.
Here, we filter out such false positives.
This has two effects:
- It eliminates quite a number of real false pasitives in practice.
- With fewer hints close together, fewer hint markers are obscured by
the hints from (non-clickable) wrappers. This reduces the need for
rotating the hint stacking order, e.g #1252.
|
|
Fixes #425.
Conflicts:
content_scripts/scroller.coffee
|
|
|
|
... Also, do not set an active hint marker initially (because it's not
predicatble which hint will be selected).
|
|
Here's why:
- b7535a604954b5873d825eb66bfecd08f1f2c99b is complicated and complex (O(n^2)).
- With experience, it is not obviously better than what was there
before, and in some cases it's worse.
- b7535a604954b5873d825eb66bfecd08f1f2c99b selects way too much text in
some cases; e.g. on Google Inbox, it selects text lengths in the tens
of thousands of characters. That cannot be useful.
- With global hints, this extra cost (resulting from passing large
objects between frames) is significant and noticable.
- The simpler approach of accounting for text length when scoring
filtered hints (tweaked here: 5cbc5ad702a01a81b98f8c82edb3b6d227c2c7b5)
works better in practice.
|
|
Just tweaks.
|
|
- Better comments in places.
- Better variable and message names in some places.
|
|
|
|
Because we're running hint modes in multiple frames, we need to ensure
the right frame has the focus for selectable elements (inputs).
|
|
|
|
Here's the issue (and we may have to address this ouside of this PR).
If we put the HUD in the top frame, then the top frame grabs the focus
when the HUD is displayed.
If we open link hints with the help dialog open, then the help dialog
loses the focus, and we can't `Esc` out of it.
|
|
We cannot use "request" and "name" to describe a link-hints message.
The message is then accepted (and fails) on the options page where there
is no handler.
So, here, use "messageType" instead of "name".
|
|
|
