Fri 11 Aug 2017 04:44:59 AM UTC, comment #12:
Looks good with fltk/qt (do we need all those drawnow() calls in
demo copyobj code?)
Demo still looks slightly off for gnuplot backend, but ptobably it is a different bug.
Dmitri.
|
Fri 11 Aug 2017 04:17:58 AM UTC, comment #11:
Ugh. orderfields was completely busted and the BIST tests were too simple to detect it. I changed it to use true permutation matrices and it now seems fine. See cset http://hg.savannah.gnu.org/hgweb/octave/rev/92a3b165689e.
|
Fri 11 Aug 2017 02:44:01 AM UTC, comment #10:
I did some more testing and used this version of copyobj demo
1) orderfiels.m is called 14 times!!
2) with a breakpoint in copyobj it stops after the first window is created.
3) The fist window looks good at this point.
4) the legend has 11errorbar and title is "original"
5) and box is good
6) continue from break-point
7)with a break-point in orderfields.m you can watch it construct the next window.
8) and at the end the title "original shows up on the 2nd window and the box gets small.
9) the first window now has a title of "new" and a legend of "errorbar22" and a good size box.
10) It is the newer window that in now titled "original" that has the small box.
|
Fri 11 Aug 2017 01:26:52 AM UTC, comment #9:
I have been thinking about this and my impression is that orderfields must be used deeper than what is visible in the demo copyobj. Is there an easy way to see if it is called more than once? maybe a breakpoint in orderfields?
|
Fri 11 Aug 2017 12:45:56 AM UTC, comment #8:
see comment #4
|
Fri 11 Aug 2017 12:26:38 AM UTC, comment #7:
What is your last good/bad bracket?
Dmitri.
|
Thu 10 Aug 2017 11:57:09 PM UTC, comment #6:
@Rik
I did some checking and the window with the bad legend box is the original, not the copy.
So this is even more confusing
( I made a different legend in the copy and then I realised that the original was small)
But why it would show up at this change set is ?????
|
Thu 10 Aug 2017 09:37:47 PM UTC, comment #5:
@Doug: You're definitely right that it is a change in orderfields, but I can't figure out what is going on. I made a copy of the old function and called it orderfields2. I then call both orderfields and orderfields2 and compare their outputs using isequaln during copyobj and everything is identical.
|
Thu 10 Aug 2017 07:41:59 PM UTC, comment #4:
this is the first bad one.
Changeset:
23349 (4f07b4770eec) orderfields.m: recode algorithm to halve running time (bug #50688). …
|
Thu 10 Aug 2017 07:03:19 PM UTC, comment #3:
One way to restore updating is to get the handle for the "legend" axes object and then change the property "cameraupvectormode" from "manual" to "auto". After that, slightly changing the plot size will force a redraw and the box will be calculated correctly.
The issue is why the axes size is not correctly. The "position" and "outerposition" properties match. There is something strange with the "tightinset" property which has negative numbers in it. But I can get those to go away after resizing the plot so that they are updated.
|
Thu 10 Aug 2017 06:26:54 PM UTC, comment #2:
Confirmed. This may have something to do with the fact that legend objects are not fully copied. It also could be a consequence of the change to listeners that is affecting colorbars.
|
Thu 10 Aug 2017 06:16:23 PM UTC, comment #1:
I will bisect
|
Thu 10 Aug 2017 05:30:43 PM UTC, original submission:
The textbox around plot keys is wrong in "demo copyobj 1"
(hg id d56c18dc1373+ tip @)
See attached.
Worked fine in 4.2.1.
Dmitri.
|