[OpenJDK 2D-Dev] RFR: 8203485: [freetype] text rotated on 180 degrees is too narrow

[Philip Race]

+1 (approved) from me.

-phil.

On 5/22/18, 9:40 AM, Andrew Brygin wrote:
> corrected:
>
>   47     static final Font fnt = new Font(Font.SERIF, Font.PLAIN, 12);
>
> Updated webrev: http://cr.openjdk.java.net/~bae/8203485/webrev.01/
>
> Thanks,
> Andrew
>
>> On May 22, 2018, at 7:26 PM, Philip Race<philip.race at oracle.com>  wrote:
>>
>> Fix sounds good.
>>
>> The test has a bug
>>    47     static final Font fnt = new Font("Time New Roamn", Font.PLAIN, 1).deriveFont(12.0f);
>>
>> "Roamn" should be "Roman”.
>>
>> But since Linux is very unlikely to have that font, I think you should use Font.SERIF instead.
>>
>> Actually I am unclear why you need to use deriveFont and can't
>> test this with the constructor taking int of 12.
>>
>> -phil
>>
>> On 5/22/18, 8:50 AM, Andrew Brygin wrote:
>>> Webrev: http://cr.openjdk.java.net/~bae/8203485/webrev.00/
>>>
>>> Bug:
>>> https://bugs.openjdk.java.net/browse/JDK-8203485
>>>
>>>
>>> With freetype, text rotated on 180 or 270 degrees is too narrow.
>>> Attached test case and screenshots demonstrate the problem.
>>>
>>> The problem is caused by the rounding applied to the cases of vertical
>>> and horizontal text, see freetypeScaler.c, lines 768 and 773:
>>>
>>> http://hg.openjdk.java.net/jdk/client/file/80a5ff734fcd/src/java.desktop/share/native/libfontmanager/freetypeScaler.c#l768
>>>
>>>
>>> http://hg.openjdk.java.net/jdk/client/file/80a5ff734fcd/src/java.desktop/share/native/libfontmanager/freetypeScaler.c#l773
>>>
>>>
>>> The rounding routine is defined as ROUND(x) ((int) (x+0.5)) (see line 48),
>>> and it gives incorrect results for negative arguments.
>>>
>>> For example, say glyph advance is 8, and we render it without rotation,
>>> and with rotation on 180 degrees. In these cases, we get ROUND(8) = 8,
>>> whereas for 180 degrees we get ROUND(-8) = -7. This rounding error
>>> leads to decrease of the width of the rendered text.
>>>
>>> For the case of integer metrics, we can expect that FT26.6 advances
>>> produced by freetype are integer, i.e. fractional part (lower 6 bits)
>>> is zero. So, we can convert them to glyph info values without floating
>>> point division, just by using integer shift. In this case, there is no
>>> need to round the floating point value to integer, and there is no need
>>> to care about sign of the argument.
>>>
>>> Thanks,
>>> Andrew
>>>