19601 - rounding problems in dashed and dotted borders

Status: VERIFIED FIXED

(Core :: Layout, defect, P3)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

This bug may have resulted from the fix to bug 1054.  It started around the same
time.

DESCRIPTION:  There are rounding problems on dashed and dotted borders.  One
edge ends up a little inside the ends of the sides that meet it.  That is, the
borders look like this (exaggerating greatly, since the errors are 1px):


-------|
-----  |
   |  |
   |  |
   |  |

STEPS TO REPRODUCE:
 * load http://www.w3.org/Style/CSS/Test/current/sec5517.htm

ACTUAL RESULTS:
 * rounding problems.  screenshot to be attached

EXPECTED RESULTS:
 * no rounding problems

DOES NOT WORK CORRECTLY ON:
 * Linux, apprunner, 1999-11-22-08-M12


WORKS
CORRECTLY ON:
 * older builds, *I think*

Comment 1

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: screenshot showing problems

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

The screenshot shows problems on
 * the right edge of the second-to-last (eleventh) and last (twelfth) paragraphs
 * the bottom edge of the first and second-to-last (eleventh) paragraphs

Comment 3

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

See also bug 19963.

Comment 4

[Patrick C. Beard]

This is part of layout, or CSS rendering, I'd guess.

Comment 5

[dcone (gone)]

It looks as if the bug occurs just inside of tables, my regression test work
just fine on all platforms.. I will have to create some tests that duplicate
this problem.. the test case included with this bug has way to many things going
on to be helpful.  It would be nice to have an isolated example for a specific
bug.  Also, is this a Linux bug only, which makes it a GFX bug.. or does it
happen on all platforms.  Knowing if it does happen on older builds does help,
the -- "I Think" part does not help..  I will research these questions.

Comment 6

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: testcase clearly showing rounding problems

Comment 7

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

In the test case I just attached, the DIVs along the top all have the same top,
and are successively taller by 0.05pt (1 twip).  The DIVs along the left all
have the same left, and are successively wider by 0.05pt (1 twip).  This shows
that the problems are with rounding, and don't have anything to do with tables.
It's just the test cases inside the table end up rounded at different points
because they're offset by the table's border width and cell padding.

Comment 8

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

And I forgot to mention that I see the problem along the bottom edge for some of
the ones along the top (for me, the first 8 and the 8 separated by four from the
end) and along the right edge for the ones along the left (for me, the 8
separated by 8 from the end).  Which ones you see it for probably depends on
your logical resolution.  (I realize I may not have included enough along the
left.)

Comment 9

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

The problems in attachment 3440 [details] first appeared in Linux viewer between the
builds of 1999-11-02-12-M11 and 1999-11-04-08-M11.  There was a significant
change in how dashed borders were drawn between those builds.

This means I think the problems were introduced in version 3.80 of
nsCSSRendering.cpp, which was overall an improvement to these borders, except it
caused this bug.

Comment 10

[dcone (gone)]

Thank you for all the input.. this will help out a Ton is solving this one!!!

Comment 11

[dcone (gone)]

Fixed the translation matrix, I had to add an error amount to the calculated
width and height of the transform.  This is because x0,y0 are transformed first
and rounded.  The width and height were then calculated and rounded.  This
rounding would take off some fractional part that was needed to calculate the
x1,y1.  So I calculated an error and added this amount to the width and height
before rounding.  I also made a speadsheet to prove and understand these errors.
I will look at all the routines in the nsTransform class for this kind of error.

Comment 12

[dcone (gone)]

Fixed transformation matrix.  The height and width did not include the a
rounding error that can be introduced.  An error is now built into the Transform
to prevent such errors.

Comment 13

[John Unruh]

Verified. Worksforme.