<Swing Dev> [13] RFR JDK-8214702:Wrong text position for whitespaced string in printing Swing text
Sergey Bylokhov
Sergey.Bylokhov at oracle.com
Thu May 16 15:43:03 UTC 2019
Hi, Prasanta.
Should not we use the real "screen" FontRenderContext instead of "DEFAULT_FRC"?
Did you check do we need to update the similat cases in the same class(we use getJustifiedLayout() 4 times mostly in the same code pattern)?
On 15/05/2019 22:23, Prasanta Sadhukhan wrote:
> Not sure if you are saying it's ok to "push"? I guess I need a 2nd reviewer if you are ok with this.
>
> Regards
> Prasanta
> On 16-May-19 12:01 AM, Phil Race wrote:
>> OK. Let's try that. I am sure it is still going to be non-ideal in a couple of ways
>> 1) where printed is shorter than screenwidth and we don't adjust, there may
>> be space after the text and before the "edge" of the component, but at least
>> there's no clipping and the text should look natural initself
>>
>> 2) Where printed is longer than screenwidth and we do adjust we may still
>> run into similar cases as this ..
>>
>> -phil.
>>
>> On 4/19/19 12:56 AM, Prasanta Sadhukhan wrote:
>>>
>>>
>>>
>>> On 18-Apr-19 12:31 AM, Phil Race wrote:
>>>>
>>>>
>>>> On 4/16/19 3:38 AM, Prasanta Sadhukhan wrote:
>>>>>
>>>>> Hi Phil,
>>>>>
>>>>> It seems screenWidth or "advance of the string at screen-resolution" is 533 whereas string advance calculated using printer fontmetrics is 503
>>>>>
>>>>> Now, TextLine#getJustifiedLine() [called from TextLayout.getJustifiedLayout] again calculates "justifyAdvance" for TextLineComponent which comes out to be 503,then it calculates
>>>>> the actual justification "delta" by subtracting justifyAdvance from screenWidth which is 533-503=30 and it then does TextJustifier.justify(delta)
>>>>> which calculates the amount by which each side of each glyph should grow or shrink.
>>>>>
>>>>> Then TextLine.getJustifiedLine() applies this value by calling TextLineComponent.applyJustificationDeltas() where it
>>>>> handle whitespace by modifying advance but handle everything else by modifying position before and after,
>>>>> so "spaces" seem to grow in size resulting in shifting of text with whitespaces.
>>>>
>>>> The spaces being used to add needed or absorb surplus space is what I understood the current code to be doing,
>>>>
>>>> However I am not sure what you mean by this :-
>>>> "but handle everything else by modifying position before and after,"
>>>>
>>>> Code run through text layout will always have glyph positions assigned, so I would suppose modifying the position
>>>> is how the spaces were handled too .. and of course this means running through all preceding glyphs to make
>>>> it consistent .. and I thought it was only adjusting the spaces, so what is "everything else"
>>> It seems when TextLineComponent.applyJustificationDeltas() is called, ExtendedTextSourceLabel.applyJustificationDeltas() handles that
>>> http://hg.openjdk.java.net/jdk/client/file/dc6c5c53669b/src/java.desktop/share/classes/sun/font/ExtendedTextSourceLabel.java#l1015
>>> where it handles "whitespace" a bit different (if block) from other glyph (else block) where advance is is not considered, which is also conveyed in the comment @1011
>>>>>
>>>>> Since it was mentioned in TextLayout.getJustifiedLayout() that " For best results, justificationWidth should not be too different from the current advance of the line"
>>>>
>>>> So for "best" results don't try to adjust a string that's 3 words and 80 pixels to a 500 pixel width.
>>>>
>>>>> I am proposing a fix to conditionally call TextLayout.getJustifiedLayout() only if the difference between justificationWidth and advance with for printer is more than 10.
>>>>>
>>>>
>>>> I had proposed investigating what happens if you simply don't use justification when the text will fit.
>>>> Maybe you are refining that but I am not sure about the details.
>>>> 10 is one arbitrary number and is not proportional to the length - which is what I would be
>>>> thinking about first in succh an approach
>>>> And I am not even sure you mean what you say.
>>>> You say "more" than 10, yet your code implements "less" than 10.
>>> It was a typo in mail.
>>> Modified webrev to not use justification if string width of text is within screenWidth
>>> http://cr.openjdk.java.net/~psadhukhan/8214702/webrev.4/
>>>
>>> Regards
>>> Prasanta
>>>> And moreover its an absolute value you are using, which re-introduces the clipping problem, doesn't it ?
>>>> ie you are purely looking at the difference and it isn't what I has proposed and I thought you were trying.
>>>>
>>>> -phil
>>>>
>>>>> webrev: http://cr.openjdk.java.net/~psadhukhan/8214702/webrev.3/
>>>>>
>>>>> Regards
>>>>> Prasanta
>>>>> On 26-Feb-19 12:30 PM, Philip Race wrote:
>>>>>>
>>>>>>
>>>>>> On 2/25/19, 10:21 PM, Prasanta Sadhukhan wrote:
>>>>>>>
>>>>>>> Thanks Phil for review. So, you are doubting it will regress swing printing tests. As you told earlier, I have ran the following regression test with this fix
>>>>>>>
>>>>>>
>>>>>> It may not regress a test if the test is not being tested under the
>>>>>> same conditions for which it is created but I am telling you for a
>>>>>> fact that the fix is wrong. screenWidth should be "width on the screen"
>>>>>>>
>>>>>>> (https://bugs.openjdk.java.net/browse/JDK-6488219 The bug above is covered by java/awt/print/PrinterJob/SwingUIText.java)
>>>>>>> and I did not notice any regression. Any other test we have for swing printing that we can run?
>>>>>>>
>>>>>>
>>>>>> No idea which tests will show this today but I know it is an issue.
>>>>>> No way we can push this and then just wait for the complaints.
>>>>>>
>>>>>>> >>Previously we were using DEFAULT_FRC to make it a screen width which except for maybe needing to be updated for hi-dpi screens is what we want.
>>>>>>> This issue was there from jdk1.6. If it is for hidpi screen, we would have seen it from jdk9 onwards where we supported hidpi, no?
>>>>>>
>>>>>> What I am saying here is that DEFAULT_FRC means "screen frc" and
>>>>>> I think that should have been updated in 1.9 but was missed because
>>>>>> it (hidpi for windows) was not a small or contained task.
>>>>>> This is an ancilliary observation of something that should be looked
>>>>>> at entirely independent of this bug.
>>>>>>
>>>>>> -phil.
>>>>>>
>>>>>>>
>>>>>>> Regards
>>>>>>> Prasanta
>>>>>>> On 26-Feb-19 3:25 AM, Phil Race wrote:
>>>>>>>> The current fix confused me for a while as I could not see how it was
>>>>>>>> at all different than the existing code, since I can't imagine when we'd
>>>>>>>> ever take the "else" branch here :
>>>>>>>> 533 TextLayout layout;
>>>>>>>> 534 if (!isFontRenderContextPrintCompatible(frc, deviceFRC)) {
>>>>>>>> 535 layout = createTextLayout(c, text, g2d.getFont(), deviceFRC);
>>>>>>>> 536 } else {
>>>>>>>> 537 layout = createTextLayout(c, text, g2d.getFont(), frc);
>>>>>>>> 538 } Eventually when walking through it I noticed this 531 FontMetrics fm = g2d.getFontMetrics();
>>>>>>>> 532 float screenWidth = SwingUtilities2.stringWidth(c, fm ,trimmedText);
>>>>>>>>
>>>>>>>> "fm" from line 532 is getting a FontMetrics from the PRINTER - ie the scaled FontRenderContext.
>>>>>>>> It then uses this to calculate the advance width for such a case - ie the printer
>>>>>>>> but then *assigns it to a variable called screenWidth*.
>>>>>>>>
>>>>>>>> Previously we were using DEFAULT_FRC to make it a screen width which except
>>>>>>>> for maybe needing to be updated for hi-dpi screens is what we want.
>>>>>>>>
>>>>>>>> So in the updated proposed fix the wrong width is passed to getJustifiedLayout().
>>>>>>>>
>>>>>>>> This may not matter here because there is plenty of space, but in other cases
>>>>>>>> Swing printing will be clipped as a result. And there were many, many, bug reports about
>>>>>>>> that. Which is why the code is laying out to the screenwidth because that is where the
>>>>>>>> UI component size available came from. Buttons & Label text are the typical cases where
>>>>>>>> this showed up.
>>>>>>>>
>>>>>>>> There maybe other things to change, as well but the incorrect screenWidth is the
>>>>>>>> main problem I see here.
>>>>>>>>
>>>>>>>> -phil.
>>>>>>>>
>>>>>>>> On 2/25/19 12:05 AM, Prasanta Sadhukhan wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On 21-Feb-19 4:50 AM, Sergey Bylokhov wrote:
>>>>>>>>>> On 13/02/2019 22:53, Prasanta Sadhukhan wrote:
>>>>>>>>>>> Hi Sergey,
>>>>>>>>>>>
>>>>>>>>>>> I believe drawChars() also has same printing issue [and should be changed like modified drawString()] but I am not able to test it as reproducer testcase uses JLabel whose constructor can only accept "String" and not char[] so I can only test drawString(). Using drawChars() implementation in drawString() still reproduces the issue.
>>>>>>>>>>
>>>>>>>>>> Is it possible temporary replace the call to drawString() by the drawChars(), to check how drawChars() will work?
>>>>>>>>> As I told, it behaves similarly to unmodified drawString and the issue can still be seen. I think we should commit this drawString() change in this fix and I can open another bug to investigate drawChars() impl and reproducer. Will that be fine?
>>>>>>>>>
>>>>>>>>> Regards
>>>>>>>>> Prasanta
>>>>>>>>>> or probably we can implement drawChars() on top of drawString()?
>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Regards
>>>>>>>>>>> Prasanta
>>>>>>>>>>> On 14-Feb-19 4:12 AM, Sergey Bylokhov wrote:
>>>>>>>>>>>> Hi, Prasanta.
>>>>>>>>>>>>
>>>>>>>>>>>>> I modified the fix to use deviceFRC if not compatible and in sync with the comment which says "obtain a TextLayout with advances for the printer graphics FRC"
>>>>>>>>>>>>> I used SwingUtilies2.getStringWidth() which calculates the advances of the string if text layouting is used.
>>>>>>>>>>>>>
>>>>>>>>>>>>> http://cr.openjdk.java.net/~psadhukhan/8214702/webrev.2/
>>>>>>>>>>>>
>>>>>>>>>>>> Can you please take a look to the existed drawChars() method, which is implemented in the similar way as drawStringImpl() and new version of drawString(), but they have some small difference. Why we cannot use the same logic?
>>>>>>>>>>>>
>>>>>>>>>>>> For example in the drawChars:
>>>>>>>>>>>> =========
>>>>>>>>>>>> FontRenderContext deviceFontRenderContext = g2d.
>>>>>>>>>>>> getFontRenderContext();
>>>>>>>>>>>> FontRenderContext frc = getFontRenderContext(c);
>>>>>>>>>>>> if (frc != null &&
>>>>>>>>>>>> !isFontRenderContextPrintCompatible
>>>>>>>>>>>> (deviceFontRenderContext, frc)) {
>>>>>>>>>>>> String text = new String(data, offset, length);
>>>>>>>>>>>> TextLayout layout = new TextLayout(text, g2d.getFont(),
>>>>>>>>>>>> deviceFontRenderContext);
>>>>>>>>>>>> String trimmedText = trimTrailingSpaces(text);
>>>>>>>>>>>> if (!trimmedText.isEmpty()) {
>>>>>>>>>>>> float screenWidth = (float)g2d.getFont().
>>>>>>>>>>>> getStringBounds(trimmedText, frc).getWidth();
>>>>>>>>>>>> layout = layout.getJustifiedLayout(screenWidth);
>>>>>>>>>>>>
>>>>>>>>>>>> ==========
>>>>>>>>>>>> Similar but not the same logic in the fix:
>>>>>>>>>>>>
>>>>>>>>>>>> 524 FontRenderContext frc = getFontRenderContext(c);
>>>>>>>>>>>> 525 if (frc.isAntiAliased() || frc.usesFractionalMetrics()) {
>>>>>>>>>>>> 526 frc = new FontRenderContext(frc.getTransform(), false, false);
>>>>>>>>>>>> 527 }
>>>>>>>>>>>> 528 FontRenderContext deviceFRC = g2d.getFontRenderContext();
>>>>>>>>>>>> 529 String trimmedText = trimTrailingSpaces(text);
>>>>>>>>>>>> 530 if (!trimmedText.isEmpty()) {
>>>>>>>>>>>> 531 FontMetrics fm = g2d.getFontMetrics();
>>>>>>>>>>>> 532 float screenWidth = SwingUtilities2.stringWidth(c, fm ,trimmedText);
>>>>>>>>>>>> 533 TextLayout layout;
>>>>>>>>>>>> 534 if (!isFontRenderContextPrintCompatible(frc, deviceFRC)) {
>>>>>>>>>>>> 535 layout = createTextLayout(c, text, g2d.getFont(), deviceFRC);
>>>>>>>>>>>> 536 } else {
>>>>>>>>>>>> 537 layout = createTextLayout(c, text, g2d.getFont(), frc);
>>>>>>>>>>>> 538 }
>>>>>>>>>>>> 540 layout = layout.getJustifiedLayout(screenWidth);
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>
>>>>
>>>
>>
>
--
Best regards, Sergey.
More information about the swing-dev
mailing list