Age | Commit message (Collapse) | Author |
|
If we have an URL with e.g. %2F in the path, we shouldn't treat that as
separator, and it shouldn't be converted to a "/".
Fixes #5908
(cherry picked from commit 49d92d463424fd56d3ebd9ff6ed6b44e6d0eb6d9)
|
|
|
|
Closes #5905
(cherry picked from commit ba152cbe4ab3c97110a21df02ee858dcc8c8f14b)
|
|
This reverts commit b55675ad4afe12be7e4a4f39960bb58818c502f8.
(cherry-picked from commit 19586576c5ed5cfc82439dc5f5a4123fbd9e8d91)
|
|
Partially cherry picked from a16e8b2bcde372dd638a8fe6bece3f312e49a297
and b3feb9ece2ede28ee9a5e2af4226c5893e0b11ab
|
|
With macOS Big Sur (11.0), the default monospace font family we get from
QFontDatabase is ".AppleSystemUIFontMonospaced". The leading dot in
there leads to a stylesheet like:
font: 10pt .AppleSystemUIFontMonospaced;
Which fails to parse, thus messing up styling entirely on Big Sur.
Fixes #5663
(cherry picked from commit 56b1d21d66204fb1c2507c6819ae731aab463f00)
|
|
Follow-up to 120f8ce1d48cee39b785b9e74cb541652a2ca26d
(cherry picked from commit a35ad6920046a44d39ba9e88d70b2985145f58ac)
|
|
(partially cherry picked from commits
451d912bc76fffa5478ce051fa89e101438170e0 and
a14d0d433eec6edb5c9ecb872d458b5793c7f2bd)
|
|
See #5892
|
|
The --no-referrers flag was removed in Chromium 79:
https://chromium-review.googlesource.com/c/chromium/src/+/1808144
We now instead use the request interceptor to remove the referrer, which
is possible ever since Qt 5.12.4 and 5.13.1 and something we've wanted
to do anyways, see #4793.
The setting is still marked as "requires a restart" (and it doesn't
support URL patterns) because that's partially true: The `same-domain`
value is still handled via CLI arguments.
Fixes #5892 (together with the last commit for 'same-domain')
|
|
Fixes #5892 (together with the next commit for 'never')
|
|
|
|
(cherry picked from commit 907e0cc33c02bb5df348a12251106755fc569137)
|
|
This avoids MS Defender mistakenly thinking we're a downloading a
trojan...
Fixes #5902
|
|
(cherry picked from commit c58788935f82cf7d7f199b3f1928b61cb51ec09e)
|
|
To agree with what we do via Black for Python
(cherry picked from commit c02c98a0aa6b2d15a08631df6cd5205a04903594)
|
|
It looks like our views don't update correctly when fetchMore() is
called on models, see 32fa1ff and #5897.
Since Qt 5.11, Qt actually calls fetchMore in models internally when
their view is scrolled to the maximum, so that this condition doesn't
actually work properly anymore (thus regressing #2841), because
`rowCount()` on the model will return the updated row count after
already fetching more data.
As a workaround, we now ask the view whether the index below the current
one is being displayed, and if not, use the existing expandAll() hack to
force it loading new data.
This also fixes a crash introduced by the new PgUp/PgDown
bindings (`:completion-item-focus {prev,next}-page`): If we're moving
towards the end of visible data, PgDown will at some point focus the
last visible item. If we then press PgDown again, we move the selected
item to an invisible (unloaded) item, thus causing a ZeroDivisionError
when trying to get the element height of those items.
Fixes #5848
(cherry picked from commit 98e4457b8d06a6fedea35836b8633a3bc316ed96)
|
|
On some setups, it looks like setting WA_TranslucentBackground causes
worse performance. Additionally, it also breaks dmenu window embedding
with its `-w` flag. This adds a setting to make it optional.
I originally tried to set this in a config change handler, but even with
a subsequent .hide()/.show(), for some odd reason that causes
rendering/repainting artifacts, possibly due to a Qt bug.
Follow-up to #5546
Closes #5805
(cherry picked from commit 50cd6aa6388817ae1cc3e31c420ea7746a5bb182)
|
|
qutebrowser/mainwindow/tabwidget.py:645: DeprecationWarning: an
integer is required (got type float). Implicit conversion to integers
using __int__ is deprecated, and may be removed in a future version of
Python.
size = QSize(width, height)
This was a left-over from 66ab7f263472d95e8c19bb8d67f377a6da580b44.
See #4928
|
|
Possibly also address: https://github.com/qutebrowser/qutebrowser/issues/1752
(cherry picked from commit 07a6ad0bb7ee77f95bb5beb6172ac56ec044affe)
|
|
(cherry picked from commits c5d5197, b6bc90c and aacc68a)
|
|
|
|
With Qt 5.12, https://www.vr.fi/en fails to load connections because it
uses Object.fromEntries() which was added in Chromium 73:
https://caniuse.com/mdn-javascript_builtins_object_fromentries
however, Qt 5.12 is based on Chromium 69.
This adds a polyfill from here:
https://gitlab.com/moongoal/js-polyfill-object.fromentries/-/tree/master
Possible alternatives:
https://github.com/es-shims/Object.fromEntries
(too complex)
https://github.com/feross/fromentries/blob/master/index.js
(not actually a polyfill but a "ponyfill", uses other unsupported
JS features)
|
|
Before 4cd255c7372625b4c3a630844ffd2b42d6ddde58, we correctly raised
`NoWindow()` via `_window_by_index(-1)`.
After this change, `first_opened_window` and `last_opened_window()`
didn't do anything at all, because their for-loop iterated over an empty
range, thus causing the `Unreachable` to be hit.
This caused #1246 to manifest in a different way than usual. Now, we're
back to the old (still buggy) behavior in that case.
Closes #5849
(cherry picked from commit 18750a332a5474fe42bcdefcf14cea10eec41243)
|
|
With b997220c3c2976a71a45c42509ea55db2762749d, we introduced code which
calls .cancel() on all downloads on shutdown. However, this includes
downloads which were already finished, causing their finished signal to
be emitted a second time.
Combined with 4cd255c7372625b4c3a630844ffd2b42d6ddde58, this
caused #5849 (though there might be other ways to reproduce?):
- Window is shutting down, thus not available via objreg anymore
- Downloads get cancelled
- Completed PDF downloads emit finished again
- PDFjs tries to open again, but no window is available anymore
(partial cherry pick from 5eabf460b9709c62a6535bc8eda52c2141177cb2, but
with potentially regressing assertions removed)
|
|
|
|
Looks like trying to stabilize it didn't help yet...
|
|
- It only seems to catch some dependencies, not all
- Its PRs (without batching) are kind of noisy
- Despite the GitHub docs saying otherwise, it seems to open PRs for
forks...
- GitHub Actions seem to use semantic versioning pretty much
consistently, so this should hopefully work out without explicit
pinning.
|
|
|
|
|
|
|
|
|
|
|
|
See #5359
|
|
|
|
|
|
The "klass" is a functools.partial instance, not the "typ" (which is the
instance)...
Also, the second check is not needed as we're filtering out all
ListOrValue instances out already anyways...
|
|
|
|
|
|
The test file was taken from acid3, which does actually serve documents
for those locations.
For us, those documents never were available (as we're using `setHtml`
which uses a `data` URL internally), but the test does work just fine
without them. It was intended to test `appendChild` with different tags,
see the original commit (6c241f96ed626f600911cac77f357ba8eb131e0c) as
well as #2723 for details.
So far, them being there wasn't a problem, but with Qt 5.15.2 (or rather
the underlying Chromium 83), we get errors logged like this one:
"Invalid relative frame source URL (xhtml.3) within data URL.", source: (47)
So let's just not set them at all.
See #5752.
|
|
See #5787
|
|
Fix bad formatting in tabs.title.format docs
|
|
|
|
|
|
|
|
|
|
Speculative fix for #5787
|
|
Speculative fix - depending on the font, I suspect if we scroll
horizontally too much, the first hint will not be visible anymore.
See #5787
|
|
See #5787
|
|
See #5787
|