Created attachment 6646 [details] [review]
Patch

This patch combines scaled fonts with the same face into the one subset.

Thanks Adrian!

I'm not quite sure how this works for Type3 fonts, specially with respect to
hinting and scale, but I'll test further.

From a quick look, this doesn't look right:

-    sub_font->sub_font_glyphs = _cairo_hash_table_create
(_cairo_sub_font_glyphs_equal);
+    if (parent->merge_scaled)
+        sub_font->sub_font_glyphs = _cairo_hash_table_create
(_cairo_sub_font_glyphs_equal);
+        sub_font->sub_font_glyphs = _cairo_hash_table_create
(_cairo_sub_font_glyphs_equal);

I also think you don't need a separate equal/hash function for the merged case.
 Just keeping a pointer version of base.hash should be enough.  I mean, no
font_face_t and scaled_font_t have the same address, right?

Created attachment 6657 [details] [review]
Patch

I have rearranged the patch a bit. It has been tested with truetype, type1, and
type 3 fonts. 

There is no problem with hinting and scale of type 3 fonts as the PS/PDF
backends turn off hinting and the FontMatrix in the type 3 font dictionary is
the inverse of the scaled font it was created from.

(In reply to comment #3)
> Created an attachment (id=6657) [edit]
> Patch
> 
> I have rearranged the patch a bit. It has been tested with truetype, type1, and
> type 3 fonts. 

Thanks.

> There is no problem with hinting and scale of type 3 fonts as the PS/PDF
> backends turn off hinting

A user still can be setting hinting on.  I agree that fonts hinted to 1pt
boundaries are not that useful, but they are possible currently, and we use
enable metrics hinting in the test suite for example.  Not sure how that will be
affected.

> and the FontMatrix in the type 3 font dictionary is
> the inverse of the scaled font it was created from.

> A user still can be setting hinting on.  I agree that fonts hinted to 1pt
> boundaries are not that useful, but they are possible currently, and we use
> enable metrics hinting in the test suite for example.  Not sure how that
> will be affected.

After some further testing I can see I misunderstood how the font options work.
Another problem is this will not work for bitmap fonts.

I'm open to suggestions.