Re: Font sizing (was Re: [dia] Windows created diagrams are no longer readable)

[Lars Clausen <lars raeder dk>]

On Mon, 2004-05-17 at 15:19, Hans Breuer wrote:
> At 14:14 14.05.04, Lars Clausen wrote:
> > On Thu, 2004-05-13 at 22:15, Federico Maggi wrote:
> > >       hi,
> > > I am sorry if some similar post already exists.
>
> There were many posts, but none of them had such a misleading subject yet ;)
> [For me "are no longer readable" sounds like a much more serious problem 
> than just some size mismatch.]

Most of those were the Pango 1.2.4+ problem, where fonts indeed did get
unreadable.

> > > I am using the same dia version (0.92.2) both on Windows (XP) and Linux
> > > (GNU/Linux, debian/testing).
> > Once the Windows version of it comes out, try 0.93, it uses font-config,
> > so may agree more on fonts.
>
> even if there would be a font-config based build it wont solve the problem 
> which is (part of, but also read below) :
>
> _different fonts_ provide by any system will generate different sized 
> bounding boxes. There is no magic in Pango or font-config to resolve this 
> issue long time known in Dia land.
> Note: this is not a matter of the platform you are running Dia on but a 
> matter of concrete font files. Ensure they are the same, and appropriately 
> mapped/named in you font configuation and the problem is gone.

That is good to hear.  I shall have to try that at work, where I've had
somewhat of the same problem.

> Sorry Lars - but I've a different understanding of the problem. I've tried 
> Dia on win32 with both backends Pango/win32 and default 
> font-config/Pango/FT2 and beside marginal different aliasing the sizes were 
> exactly the same.

Good.  I was afraid the different hinting implementations would add up.

> Also aren't you comparing apples to elephants (Pango and Win32)? Shouldn't 
> that read FT2 and win32 font rendering ? Or did you mean font-configs font 
> list compared to the one provided by win32::EnumFontFamilies() ?

Just too lazy to write win32 font rendering.

> UPDATE: Repeating my test with 0.93 (they were originally done some month 
> ago with some 0.92 or even 0.91 ?) did show a IMO serious regression 
> though. Font sizes between the backends are no longer the same.
> To test yourself try the attached UML-Test (original from Dias 
> distribution, I've put in an additional layer showing the UML Class box 
> sizes with Dia-0.90-win32.
> Loading the file with 0.93(Pango/win32) shows some small deviation, some 
> few percent as expected. But loading the same file into .93(Pango/FT2) 
> shows that the boxes - and thus the reported text length - are about 30% 
> smaller. To me this looks like an unacceptable regression - I simply have 
> too much diagrams done with 0.90 ...

Eeek!  First thing I think about there is the magic 70% size reduction
that we introduced with Pango in the first place.  Always hated that
thing, 'cause I totally didn't understand it.  Ugh.  Maybe we will have
to have a 0.93-1 after all.

-Lars