Created attachment 129090 [details]
freetype bug fix example

Hm... i found this path in ubuntu freetype 2.5.2 and bug fixed!

Sample: https://bugs.freedesktop.org/attachment.cgi?id=129090


# Description: revert commit which creates regressions in gtk
# Ubuntu: https://bugs.launchpad.net/freetype/+bug/972223
# Debian: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=636776
--- a/src/truetype/ttdriver.c
+++ b/src/truetype/ttdriver.c
@@ -246,10 +246,7 @@
     FT_Request_Metrics( size->face, req );
 
     if ( FT_IS_SCALABLE( size->face ) )
-    {
       error = tt_size_reset( ttsize );
-      ttsize->root.metrics = ttsize->metrics;
-    }
 
     return error;
   }

Please see

  http://lists.nongnu.org/archive/html/freetype-devel/2012-04/msg00001.html

and the following e-mails in the thread for an explanation why your report is invalid.

Okay, commit is invalid. But look at the screenshot: https://bugs.freedesktop.org/attachment.cgi?id=129090
It bug. This bug exists several years, but not fixed. Why? 
Maybe there is another bug-report? Else why this bug closed?

Kirill, I can only repeat: There is *no* bug. You are simply *accustomed* to
the old value.

(0) You should always mention the used DPI value.  I guess you are using 96dpi,
    right?  Then 10pt and 11pt correspond to the ppem values 13ppem and 15ppem
    (ppem values are always rounded to integer values), respectively.

(1) The code snippet you are referencing to doesn't apply to the issue at all!
    It comes from the TrueType driver, but if I test with `ftgrid' I get a `p'
    height of 10px for and 11px for 13 and 15 ppem, respectively.  (I tested
    DejaVuSansMono version 2.34.)

(2) Consequently, you are using the auto-hinter.  Here it is completely up to
    FreeType to do the hinting.  I no longer remember the reason, but there
    was obviously a change in the auto-hinter that resulted in modified heights
    of lowercase glyphs.  You might consider this as a bug, *I* consider it as
    a global improvement for hinting.  There is more than a single font that
    the auto-hinter has to take care of.

(3) As mentioned earlier, ppem values must be integer, but pt values can be
    fractional!  If I select 10.5pt, I get 10.5*96/72=14ppem, and – tadaa –
    FreeType's auto-hinter uses a height of 11px for glyph `p'.

Corollary: Either set up your OS to use the TrueType bytecode engine, or
select a size of 10.5pt.  All these issues should have been asked and
discussed on the FreeType mailing list instead of opening a bug report!

By the way, with 10.5 DeJavu Mono underscore symbol is invisible. Is it freetype's bug?

Using `ftview' I see character `_' for all resolutions.  However, underlining is normally not a character but created as a graphics.  So I'm rather sure this is not a FreeType bug.