On Mon, Jul 30, 2012 at 09:14:11AM +0100, Michael Goffioul wrote:
> On Mon, Jul 30, 2012 at 1:33 AM, Mike Miller <
address@hidden> wrote:
> >
> > Yes, I'm seeing that the result before rounding is not near an even
> > integer. The _fixedFont loop is saying that all characters in REPCHAR
> > have the same width, but the width of the entire string is apparently
> > not the sum of its parts...
> >
> > This is my debug after setting the font to Inconsolata 16.
> >
> > _fontWidth = 11 (before round = 10.651515)
> > fm.width(REPCHAR) = 703
> > strlen(REPCHAR) = 66
> > fw = 11
> > fm.width("m") = 11
> > fm.width("l") = 11
> > _fixedFont = true
> >
>
> There's something fishy here. In a monospace font, the width of a string
> should be the sum of the width of its characters. For your information, in
> general the width of a string is a complex stuff: obviously it depends on
> the width of individual chars, but also on additional information (iirc
> called "kerning"). Basically one character (e.g. 'b'), when preceded by
> another specific char (e.g. 'i'), can be slightly offset to the left to
> make it look better. In a monospace font, all characters should have the
> same width and there should be no kerning.
Yep, I agree with that. And you've seen on my screenshot that the
characters do have equal width. I can produces lines of arbitrary
characters and they are fixed width.
> In attachment, I provided a patch I apply on qterminal/main.cpp to allow
> easy change of the font (on the command line) and give some debug
> information at startup. You might find it useful.
>
> I've tested the average font width (_fontWidth) with the mono fonts I have
> and I invariably get (for REPCHAR) a total width of 858, which is exactly
> 66*13. The only exception is FreeeMono. Would you mind experimenting with
> repChar (see my patch) to determine whether the non-constant font metrics
> is coming form a specific character or is general to the font? I'm really
> puzzled about your issue; I've tested it on my F16 and F17 boxes and I
> can't reproduce it.
Yes, I've applied your patch and added a little more.
With all the fonts I've tested I get a non-integer average font width.
With a point size of 16, the total width of repChar and average width
are:
Courier (actually Nimbus Mono L on my system): len=845, avg=12.803
Droid Sans Mono: len=831, avg=12.5909
FreeMono: len=831, avg=12.5909
Inconsolata: len=703, avg=10.6515
Liberation Mono: len=831, avg=12.5909
Ubuntu Mono: len=692, avg=10.4848
I added the following extra debug in main right after yours:
bool fixed = true;
for (int i = 0; i < repChar.length(); i++) {
if (fm.width(repChar[i]) != fm.width("m")) {
fixed = false;
qDebug("fm.width(\"%c\") != fm.width(\"m\")", repChar[i].toAscii());
}
}
if (fixed)
qDebug("All chars in repChar have equal width");
qDebug("fm.width(\"m\") = %d", fm.width("m"));
qDebug("fm.width(\"mm\") = %d", fm.width("mm"));
qDebug("fm.width(\"mmmmmmmm\") = %d", fm.width("mmmmmmmm"));
All of the fonts report equal width when testing a single character at a
time. However, the width of two or more m's is not N times the width of
one. The width of a single m is 13 for Nimbus Mono, Droid Sans Mono,
FreeMono, and Liberation Mono. 8 m's is 102 for Nimbus, 101 for the
rest. Inconsolata reports 11 for a single m, 85 for 8 m's. And Ubuntu
Mono reports 10 for a single m, 84 for 8 m's. This is the only one that
has a larger width for multiple characters, which is why I see the
characters being truncated by the cursor in that font.
Which version of Qt do you have in Fedora? I have 4.8.2.
That's about all I have time for until later this evening. I just did
some quick searching and found a few links indicating this might be a
feature of 4.8:
https://git.reviewboard.kde.org/r/102899/
https://bugreports.qt-project.org/browse/QTBUG-22040
I also see in the QFont API doc in 4.8 that it mentions supporting
"font engines that support fractional font metrics".
https://qt-project.org/doc/qt-4.8/qfont.html#StyleStrategy-enum
Thanks, these are very interesting findings. I'm also using 4.8.2 under Fedora, but it's possible that the packager has tweaked the package to change some default behavior, or it is part of the Qt style, I don't know. But if it is some system settings, it is possible those are different under Ubuntu (afaik both of you, Doug and Mike, are using Ubuntu). The link:
contains a konsole patch ("View Diff" tab in the top-right corner of the yellow area). Could you apply the change to TerminalView.cpp and give it a try. I tried it, but it didn't solve the problem with the FreeMono font (the only one showing the problem for me), but maybe it'll help for you for the other fonts. Note also that I tried FreeMono in konsole and it's full of misalignment, so I guess here the font is the problem.
Michael.