Re: [ft] With Freetype 2.8 my desktop looks bad

[Werner LEMBERG]

> It wouldn’t have occurred to me that I could set fractional values.

Yeah, this misconception is popular.  People mix up ppem and pt...

> It did seem, though, like some programs might not be handling the
> fractional correctly as some fonts in their windows looked
> noticeably smaller than before. (One example is my graphical package
> manager, Octopi.)

It is definitely a bug if progams can't accept fractional pt values.
You should report it to the maintainers.

> I got more consistent results by keeping the font sizes at 10 and
> setting the text scaling factor, which is a separate setting offered
> in Gnome, to 0.975.

OK, good to know that such a feature exists.

> So do I understand correctly that the key for an even display is to
> arrive at some integer (any integer) for the measure you call ppem?

For *your* distribution, this might be true if you want to get similar
results as before.  In general, however, this isn't a valid assumption
since the auto-hinter uses a completely different algorithm for
hinting than native bytecode hinting, and in most cases people
consider something `good' if it looks exactly as before...  So it's
rather a matter of habit.

> I was able to understand the issue better by using quantity calculus
> on your formula
> 
>   13ppem * 72dpi / 96dpi = 9.75pt
> 
> after I slightly change the units (I think your choice of dpi after
> 72 has to be changed for correctness, the other units I am only
> making more explicit), which gives me:
> 
>   13 pixels per em * 72 points per inch / 96 pixels per inch
>   = 9.75 points per em

Ah, yes, thanks.  It was rather quick writing :-)

>> So it seems that `Arch' is another distribution that used FreeType
>> < 2.8 unmodified...
> 
> Out of curiosity: would it have been advisable for distributions or
> users to apply some patch?

Yes.  From FreeType's `CHANGES' file:

    - `Light' auto-hinting  mode no  longer uses TrueType  metrics for
      TrueType  fonts.   This bug  was  introduced  in version  2.4.6,
      causing   horizontal  scaling   also.    Almost  all   GNU/Linux
      distributions (with Fedora as  a notable exception) disabled the
      corresponding patch for good reasons; chances are thus high that
      you won't notice a difference.

I got reports about the issue – however, due to some stupidity on my
side I didn't recognize the reason and rejected a reversal (the patch
fixed a real problem, BTW); only a recent, new bug report enlightened
me so that I could properly fix the issue.


    Werner