When giving the path to a FIFO or other special file qutebrowser would
completely hang, and has to be killed.
Tested:
- Asks for overwrite: file, symlink to file
- Saves in dir: dir, symlink to dir
- Aborts: block dev, char dev, fifo, socket, and a symlink to all of these
I thought I put this in here before, but apparently I did not. So here it is,
together with a new test to verify it. Other tests needed to be updated with a
mock for os.path.expandvars.
In function File.validate the try-except block has been re-written to
differentiate raised errors.
In function File.transform there was a check for validity of the file path that
is alraedy performed by File.validate under the same conditions. This check has
been removed.
When downloading a bunch (7 or 8) of files I noticed qutebrowser was using a lot
of CPU (>60%).
I did some looking, and in the `downloadProgress` callback qutebrower emits the
updated signal which causes everything to be updated. We don't really need this,
since _update_speed() calls it every 500ms anyway.
I tested by downloading 3 copies of the 1GB file [on this
page]( http://www.thinkbroadband.com/download.html ) qutebrowser consistently
pulls about 25% CPU on my system.
When removing this call, the system pulls about 17% CPU. Not a great amount, but
still significant enough to warrant a pull request ;-)
Some other notes:
- wget uses about 1.5%-2% for each process when downloading.
- When not doing any UI updates & speed calculations qutebrowser uses about 15%.
- Doing some quick profiling and strategic commenting seems to indicate there
isn't any other low hanging fruit to be improved on here.
Error messages for validate() are more specific.
Return of standarddir.conf() is explicitly tested for None to avoid ambiguity
with other falsey values.
This fixes#716, which sufficiently annoyed me to make this quick fix. It's not
a great fix, but it's not worse than what we had already, and the current
behaviour is very surprising IMHO.
Before, the completion was shrinked every time any item was removed/added to
the completion (rowsRemoved/rowsInserted signals), which was >3000 times when
completing history.
Also, the signals got connected multiple times if setting the same model, which
made the situation worse.
Fixes#734.
... As I want to copy only the domain fairly frequently.
I also changed the message in the statusline to show the actual text being
copied, which I find helpful. But if you disagree, then just undo it (it's not
that important or anything).
This sets the third-party cookie policy.
- I created a new ThirdPartyCookiePolicy() class, since this setting seems to be
unique in the way it is set...
- I set the default to 'never', which is the most secure/private setting, but
*may* break *some* features of a (very) limited number of sites; these are
usually "non-critical" features.
For example, on Stack Exchange sites you're logged in all 200+ sites if you
sign in on one of them, this features required 3rd party cookies. You can
still sign in with out, but you have to do so 200+ times (this is actually the
only example I've ever noticed).
AFAIK all "major" browsers accept 3rd-party cookies by default, except for
Safari. Firefox also made this change, but reversed it (see:
https://brendaneich.com/2013/05/c-is-for-cookie/), but they don't offer any
good arguments to *not* have it IMHO, at least not that I could find.
In any case, in my humble opinion "secure and private by default" is the best
way to ship. But you're of course free to change it if you disagree ;-)
source is undefined when you type stuff in the console, I *think* this is the
only scenario? But maybe not?
<script>
setInterval(function() {
if (window.__qute_jseval__) {
throw new Error('jseval hack failed. Sorry :-( ' + window.__qute_jseval__);
}
}, 1);
</script>
