| Ticket UUID: | 94547cbef7a87a11323edd2e8dd56bc406877da3 | |||
| Title: | Revised [text]: borders not always redrawn | |||
| Type: | Bug | Version: | revised_text | |
| Submitter: | fvogel | Created on: | 2021-07-14 15:38:58 | |
| Subsystem: | 18. [text] | Assigned To: | fvogel | |
| Priority: | 5 Medium | Severity: | Minor | |
| Status: | Closed | Last Modified: | 2022-06-07 21:21:17 | |
| Resolution: | Fixed | Closed By: | fvogel | |
| Closed on: | 2022-06-07 21:21:17 | |||
| Description: |
package require Tk
pack [text .t]
.t configure -bd 15 -relief raised
.t insert 1.0 "Line 1"
frame .f2 -bg #ff0000
place .f2 -in .t -relx 0.4 -relwidth 0.2 -rely 1.0 -relheight 0.2 \
-anchor s -bordermode ignore
update
destroy .f2
When executing this, a relic of the red frame is still visible in the text widget bottom border. This means that borders were erroneously not redrawn after destruction of frame .f2, while they should have been.
Further notes:
| |||
| User Comments: |
fvogel added on 2022-06-07 21:21:17:
All this now done. I have found bug [aad0231f07] in the process. Closing this one now. fvogel added on 2022-05-23 06:44:26: Oh, right! I didn't notice these tests because they were marked as 'nonPortable', meaning they normally don't run. I'm now dealing with them in a specific test branch nonPortable_tests off core-8-6-branch, with the objective of removing this constraint when it is no longer needed. If no surprise shows up, I'll merge it up to trunk, then revised_text, and finally to bug-94547cbef7. griffin added on 2022-05-23 00:19:35: Yes, my mistake in the text region changes. I am seeing textDisp-7.2 and -7.3 failing. Both of these tests expect the border to be drawn. From what I can tell, the test does not damage the border. It is what I see observationally and what is defined by the place command. Place defines -relx and -rely to be inside the border when -border is not present, i.e. when the default value is "inside". I also note that a focus change will also cause the border to be redrawn. These tests make no assurance about the current state of focus for the text widget. Consequently, the place operation can potentially change focus. This can result in unstable results from these tests. Regardless of what the original text widget produces for these tests, I still think the current expected results are incorrect. The test, as designed, should not redraw the border. Can you see anything that I may have missed in this analysis? fvogel added on 2022-05-20 21:24:29: Brian, sorry but I disagree. I had a deeper look at the problem and I have found that the width and height of the textRect were jsut wrongly calculated. I have fixed this in [1a49fd12]. Now the test script from my original report in the present ticket behaves correctly, and, very important, no test result needs to be changed compared to the legacy text widget. For easy testing here is an interactive test:
package require Tk
proc doit {x y} {
frame .f2 -bg #ff0000
set inset [expr [.t cget -borderwidth] + [.t cget -highlightthickness]]
set x [expr $x - $inset - [.t cget -padx]]
set y [expr $y - $inset - [.t cget -pady]]
place .f2 -in .t -x $x -width 50 -y $y -height 100
update
after 300
destroy .f2
}
pack [text .t -insertofftime 0]
.t configure -bd 15 -relief raised
.t insert 1.0 "Line 1"
bind .t <Button-1> {doit %x %y}
griffin added on 2022-05-18 20:30:55: After some deliberation, I've convinced myself that the RectContainsRect() is the wrong test completely. The text region and border are two independent (re)draw regions, and any test to determine damaged areas needs to be independent as well. For the border, there is actually 4 regions to examine, so any test limited to comparing two regions is insufficient. Beyond that, I believe the cursor extent values potentially force an overlap between the text region and border/padding, which complicates any damage calculation. With respect to the border, anything dealing with the text region is immaterial. Looking closely at test textDisp-7.2, I've determined that the border should not be drawn. The damaged area is completely contained within the text region by definition. So the current test results are wrong and indicate an unnecessary redraw of the border. I have not investigated other test discrepancies yet. I've updated the changes to TextInvalidateRegion() and textDisp-7.2 as per my observations above. These changes are on the [bug-94547cbef7] branch. griffin added on 2022-05-18 06:32:54: Very impressive! Thank you! fvogel added on 2022-05-18 03:47:48: Yes I'm aware of this issue with test constraints. There is a third way to deal with this, which is to make the tests font agnostic. I have started the implementation work, see branches less_tests_constraints for 8.6 (legacy version) and less_test_constraints_revised_text for the revised_text branch. griffin added on 2022-05-18 00:40:33: Thanks for the explanation! I determined that the reason the tests passed on Linux, at least for me locally, is because all of tests that should have failed were skipped due to the textfont constraint. This is a serious flaw in the regression system, making it pretty much useless for doing any development work. There are 2 fixes for the general font problem that come to mind: 1) Assert the presence of a particular font so that no Tk test passes without that font present, or 2) build and release Tk with a specific set of default fonts. fvogel added on 2022-05-17 22:28:57: The rectangle provided in the expose event is clipped by the text area rectangle inside the text widget. The coordinates associated to this expose event are not what we would expect. The bug is not where you have changed the code, function RectContainsRect is totally correct and TextInvalidateRegion as well. It's the clipRect (obtained from the invalidated region) that is too small. And it's too small because the invalidated region vehiculated by the expose event is a rectangle unexpectedly clipped by the limits of the text widget area (inside the border). griffin added on 2022-05-17 16:28:42: I changed the computation done on expose events to determine if the border should be redrawn. I replaced the test with original test from text widget, which appeared correct to me since it actually included the border width info whereas the revised version did not. On further inspection, it appears that the revised text combines the text area and border together before determining redraw regions. That could explain the extra border redraw with my change. fvogel added on 2022-05-17 11:46:39: Without having yet really looked at the proposed patch I'm observing many new failures at Github actions. These were not present before the patch. They are probably all the same and reveal that borders are now redrawn in more than 30 newly failing tests while borders were not redrawn in those tests before the patch. Note that tese tests do pass with the legacy widget. What I also find interesting is that whereas the patch is changing generic code, the tests fail on Windows and macOS but pass on Linux. I have no explanation about this fact so far, but digging in that direction could reveal something. fvogel added on 2022-05-17 05:55:32: I will have a look, yes thanks. griffin added on 2022-05-17 03:01:21: This problem should be fixed now with change [632732f2]. Francois: can you please confirm and close? Thanks. fvogel added on 2021-07-24 12:34:04: textDisp-7.5 and textDisp-7.7 show the issue as well. | |||
Home
Timeline
Branches
Tags
Forum
Tickets
Download
Wiki