Thanks for pushing this, John.

I think the things you mention already work, so no worries there:

• Right-clicking a variable in the workspace pane and selecting "open in variable Editor" will show the Variable Editor pane and open a tab inside it with the matrix or variable.
• Alternatively, the command "openvar <variable name>" will do the same.

The things that need fixing have been mentioned in comment #45 - comment #47.
In addition the "Autofit column width" setting is slow (see comment #39).
Copy/paste from external programs also has deficiencies (only works in Linux, and AFAICS pasting works transposed).

OK, nice, I'm very glad this has been included in core :-)

John, just to verify, your new patch (file #41199) builds & cross-builds and works fine in Linux and Windows.

I checked the pointer styles and AFAICS they're already compliant.

There's just one mention if "if error_state" in the octave-link.cc hunk. I'll leave it to someone else to change that, it's beyond me :-(  hopefully the fix will enlighten me.

Yes, I'd like to get this feature included in the development version soon.

Patch updated to Octave HG ID 293dec495256.

Philip, where are the changes for error catching and pointer style that you mention in comment #49?


(file #41193)

Updated patch that applies with a few fuzz lines to dev Octave hg id 4e770056f74a "don't attempt to access intepreter before it is ready (bug #48494)"


(file #37863)

As I said in the maintainers ML I've experimented with classdef and Java objects in Matlab's variable editor. The attached pics give some examples.

Java objects are just shown as such, Java arrays cannot be explored deeper (like double-clicking on individual cells like with cell arrays etc.).

Classdef stuff in the variable editor is treated similarly to structs (as expected; at low level they are -special- structs AFAICS).
For this example I used the first "mydate" example code from
https://yagtom.googlecode.com/svn/trunk/html/objectOriented.html

You're quick :-)
I'll try right away

BTW in the last weeks it occurred to me that it's not only structs that don't work (yet) - also Java objects and any Octave objects (incl. classdef) fail. Same for stuff like graphics handles.
Maybe all of those need to be catched (in openvar?) as long as they don't work before the VarEd attempts to display them.

char strings go fine; char arrays less so:

> b = {"thisisatext"; "anothertext"}
b =
{
  [1,1] = thisisatext
  [2,1] = anothertext
}
>> cell2mat (b)
ans =

thisisatext
anothertext

>> class (ans)
ans = char
>>

gives the attached pic.
Char arrays aren't used that much AFAIK so maybe no big deal.

My pleasure. Actually I was a bit afraid you'd be put off by all my comments :-)

I'm very happy to have a variable editor available => makes debugging my scripts and functions a LOT easier. So I'm happy to test (& comment) further as well.

As to the MXE build instructions, that has changed a little nowadays (the "octave" target should rather be "default-octave" and I've hit a few other glitches). I usually build mxe-octave once and then use it for a month or so building just new Octave versions with the prebuilt dependencies. Each new mxe-octave "cycle" seems to bring its own issues (now ~familiar to me but maybe harder to overcome for newbies).
Copying to the wiki could be easy but I have other priorities :-)

Would you agree to bring up the variable editor development in the maintainers ML? Other devs could "interfere" then (it wouldn't be your baby anymore) but development could be more rapid.

(Obviously in hindsight) it turns out that the "Autofit" setting causes the long delay when updating the VarEd pane. So that setting may not be a good idea after all.
Just a user-specified but otherwise fixed column width + row height setting would do. The row height setting works perfectly

I had time to compare a few things with Matlab (r2015b prerelease).

Matlab:
Clearing a cell in a cell array sets the cell to empty
Octave:
Sets cell contents to numeric zero

Matlab:
shows a text string as a string in one cell
Octave:
shows it as a an array of chars

Using:

a = {1 {'TL', 'TR'; 'BL', 'BR'}; 'text', [10, 20 25 35 45; 30 40 45 55 65]}

Matlab:
shows the nested cell array as "2x2 cell" (w/o double quotes);
shows numeric array as far as it fits (truncates contents)
Octave:
shows cell contents
shows numeric array as far as it fits (truncates contents) - Good!

a = struct ('b', 'field1')
a.c = [1 2; 3 4]
a.d = {1, 'text'; [10 20], {'TL', 'TR'; 'BL', 'BR'}}

gives the output of struct1.png (attached)
Then,

a(2).b = 'field2'

gives struct2.png
Double-clicking field a(1).d (~cell 1, 3) gives struct3.png

So maybe maybe (but obviously you know it better) displaying structs doesn't have to be that difficult and could be fairly straightforward.

First of all: thanks for your prolonged efforts.
I think it might be about time to bring it up on the maintainers ML for polling interest on inclusion in the next dev version. Would you agree?


As to the long variable name, I think I saw that on Windows.


As regards my remark in comment #31:
For uniform character arrays and cellstr arrays, I think quotes do not need to be shown around char variables in cells. That would only be required for mixed numeric/char cell arrays (e.g.,  {uint32(1), "1"}).
I. m aware that editing individual cells (individual characters) would require some attention then as quotes wouldn't be required in char/cellstr arrays but are needed in mixed-type arrays.
(My main motive here is that I'd like to limit use of screen space for redundant info, e.g., lots of double quotes in char strings.)


Below are some findings with the new patch on Linux (64b Mageia-4) where I could quickly build & test (Windows will take longer). I think most of it is small things:
--------------------------------------------

For variables already shown in Variable Editor, "apply"-ing the default row height setting doesn't work, nor does "Plus font height" setting, Closing the tab + renewed "open in variable editor" works fine though.

The "By Column | Uniform" Autofit preference neither works on row height (or I overlooked/missed the effect), nor is this setting persistent - the Variable Editor prefs pane always comes up with "By Column".

"Use terminal font" could be grayed out if it is unchecked (?)

The Toolbar size setting works perfectly.
(On Linux) 8 pt is really too small; 12 pt is like "Small" icon size in General tab, 24  matches "Normal", 32 matches "Large" icon size. Can this simply be linked to the General tab Icon size settings (also in the Preferences pane), please? If so this setting could be removed from the VarEd prefs pane.
I'll report later this weekend on how it looks on Windows.

Am I right to conclude that the row height setting applies to the space under the text? it looks like cell contents are consistently "top-adjusted" in cells.

Maybe nested cell contents had better be shown as "cell <dimensions>" within a Variable Editor cell, rather than attempting to display their contents. I suppose this will tackle the following issue (and perhaps could aid in making structs display in the V.E. easier as well):

Double clicking a nested cell array comprising just one cell doesn't work, steps to reproduce:

>> c = {1, 2; 3, {"text"}}

c =
{
  [1,1] =  1
  [2,1] =  3
  [1,2] =  2
  [2,2] =
  {
    [1,1] = text
  }
}

Now open c in variable editor and double-click cell (2, 2) =>
"

>> parse error:

  invalid constant left hand side of assignment

>>> {

  [1,1] = text
}
        ^
"

Compare with:

>> d = {1, 2; 3, {"text11", "text12"}}

d =
{
  [1,1] =  1
  [2,1] =  3
  [1,2] =  2
  [2,2] =
  {
    [1,1] = text11
    [1,2] = text12
  }
}

where double-clicking cell (2,2) just works fine.

Another issue:

a = {1 2; "text", [10, 20; 30 40]}

as first stmt after starting Octave with "Autofit" set to "Uniform", and then displaying variable a in the variable editor, Octave crashed consistently when double-clicking cell (2,2). However, later on I couldn't reproduce this anymore.

"Clear" in the context menu for cells in fact means "set to zero". I saw that when editing char strings: clearing one char sets it to 0. Same for numerical input and nested cell array contents.
This could be deceiving, at least merits some discussion IMO. Perhaps chars had better be reset to space rather than 0? or individual chars in a 1x<length> string, and individual cells in a cell array, should be cleared to "empty" (effectively be deleted). I'll check what Matlab does.

After entering 'a = {1 2; "text", [10, 20; 30 40]}' and opening a in the variable editor, then double-clicking cell a(2,2), a new tab opens with the 2x2 double array in a(2,2). Now, clearing a{2,2} in tab "a" doesn't update the other tab contents for a{2,2} - that only happens when displaying it in the command line terminal (by e.g., typing "a").


OK, 'nuff comments for now :-)

Later today I'll start an mxe Windows (cross)build + put it on Dropbox, when finished I'll post the link here.

Thanks for that.

I have made some changes just for you. You can change default row height and also set it to change relative to font size. It defaults to 2 pixels taller than QFont::height() for the text  settings. You can also shrink the toolbar. I had no idea how big it defaulted to so the largest selectable size is massive at present, although the default is set to the current size.

When you say:

- sometimes has trouble showing var.name and size on the pane.

Would this be long variable names, or something else?

(file #34679)

(Hi Michael, nice to hear back from you)

Tried your latest patch:

- Applies cleanly to very recent development tip
- Alternating row colors item in prefs cannot be selected (grayed out)
- Row height is large IMO, could take less real estate. Autosize as well?
- Same for icons. I have been searching around where icon size is set in the other panes but haven't pinpointed it down (there's an icon size pref in the first General prefs tab).

The previous patch:
- shows character strings as char arrays (one char per cell)
- sometimes has trouble showing var.name and size on the pane.

Anyway nice!

I have a Windows (cross-)build for 64bit Octave (not: 64-bit indexing) in my dropbox. I'm building a new one now, once its finished and uploaded I'll send you the link (hopefully Sat. morning).

I've implemented some options. Not got alternate row colours to actually show at this time but everything else seems to work. Default width is only set when the variable is first opened while autofit should update with every change.

Also replaced the multiply symbol with x since that's what disp() uses. I hadn't realised there was no standard ascii code for that.

Michael

(Would be good if I could get a windows executable with this in it since I'd like to have it with me at BMVC in september.

(file #34628)

Updated patch to build with Octave 4.1.0+ as of June 28


(file #34334)

I think I've fixed the editor position being saved between sessions. I do not understand why it opens in a column on its own, but once it's dragged somewhere else it stays there.

Not sure what to do next.

Michael

(file #34159)

The bug I reported in comments #21 & #22 appears to have been fixed in your latest cs, nice! (That bug was a little showstopper.)

I've made a small patch to clarify the "not Found" message when variables are cleared or get out of scope (e.g., when doing "dbup" or "dbdown" while debugging). I found that once scope returns to the function containing the variables, the Var.Editor pane again shows the variable, which is also very nice.


(file #34086)

(Sorry I was away for over a week, have to catch up but my daily job has priority (of course).)

A bug I just hit (a serious one I suppose) is that the variable editor doesn't cope with cleared variables opened in the variable editor.

>> a = rand (3); b = num2cell (a); b{2, 2} = {1, "a"; "b", true};
>> clear b
>> Variable b not foundVariable b not found

(ignoring the duplicate popups).

Same for variables opened in the variable editor in functions openened in debug mode. Once the function gets to the end and returns, the variable editor tab keeps hanging, and has to be closed manually as well.

Apparently the variable editor doesn't get notified that b is gone, or went out of scope.

The relevant variable editor tab has to be closed manually before other work can continue in the terminal.

As to structs, well, too bad :-) 
I find it already immensely useful to browse arrays and to see what happens inside while debugging, even if only numeric and cell arrays. Thanks again for your work on this!

I

Thanks for the advice.

I'll have a go with structs this week. No promises because they're quite a departure from other variables but I'll see how it works out. There will need to be conditions added to things like menus to prevent the user adding or removing columns, validation when changing variable names so they don't get duplicated and are required to be valid strings...

I'll aim to get a new patch up next weekend that is either just cleaned up or adding struct support.

You're right, I still see the ctrl-v related crash when no variables are open bug with a gcc-5.1.0 cross-build.

I didn't get it compiling under Windows, but I found a ctrl-c / ctrl-v related crash when no variables are open. Hopefully this might fix the crashes you were experiencing.

I've also got editing on double-click working again for cells containing matrices and strings.

With a little luck this should be all of the serious stuff.

(file #33954)

Okay, I'm going to try to get Octave compiling under Visual Studio to track this paste bug down.

Michael

Deleting rows works now.

sq-strings are accepted now.

Typing text w/o quotes in a numeric cell still popups a graphics window. The same in a cell containing text doesn't.

Thanks for these fixes!

On Windows I had some crashes when pasting text into the terminal. But when I did it slowly and carefully I experienced no crash. I'm still at loss what I did wrong - maybe a focus issue (I have a delayed "focus follows mouse" setting on Windows; but the RMB click should have set focus I suppose).

Would it be possible to add "openvar" in the right-click menu when clicking on a cell? (of course only really useful if that cell is a nested cell array but it gives a way to show nested arrays)

("our new patch" -> "your new patch" - my wireless keyboard needs new batteries :-( )

A quick check shows that the Copy / Ctrl-C crash issue has disappeared. Nice!

openvar("varname{2,2}") works fine AFAICS.

When entering text strings in a cell array, ' (single quotes, sq-strings) aren't accepted, only double quotes (dq-strings) work.

Inf and NaN are shown correctly now. I do not know that much about Octave internals nor C++ so I can't give any advice.

There's "variable editor" title on the docked pane -> fixed.

Typing "text" (w/o quotes) in a cell from e.g. tmp = ones(3) gives a graphics window pop-up and in the cell I get a number like -6.00174.

Sorry for my conspicuous absence, I think this is an advance.

I found a possibly related pasting error on Linux which I fixed by rejecting copy and paste requests if the variable editor does not have focus.

I have currently disabled opening subeditors on a double click. Commands such as openvar("varname{2,2}") still work but I haven't tested them.

My fix for lower case inf and nan is hacky but I've traced the issue back to octave_value::print returning the values in lowercase. Should I use something else?

GUI issues are still outstanding.

My biggest concern now is how to handle invalid entries. For example:

temp = ones(10);
openvar("temp");

select a cell and type in some text.

The editor will then recover perfectly when any variable, including ans, is changed. I haven't managed to automatically do this.

Suggestions welcomed.

Michael

(file #33823)

Right-clicking a row head and selecting "Delete" does not delete the row. For columns it works well.

Thanks, I've just built a tip with it and it works. I like it!

Some of the glitches I found:

Clicking on a cell containing another cell array in a nested cell array will crash octave (signal 6).

Double-clicking in an element of a struct array gives some incomprehensible error (("error: 'scalar' undefined near line 1 column 1") plus a popup that does show the proper values.
=> Looks like a simple fix.

NaN / Inf values are shown in small caps, not 'NaN' or 'Inf'

Clear <variable> when <variable> is shown in the variable editor is catched with <variable>: not found. Nice; maybe the subpane can be closed instead?

Empty strings are shown as empty cells instead of consecutive double quotes (which I would think to be more consistent).

The variable editor pane position and dock state isn't remembered across a restart of Octave.

All in all I find it very usable.

I'll change status to "in progress"

AFAICS your patch is on top of the one in patch #8316, which unfortunately doesn't apply anymore against current dev version (or 4.0.0-rc3).

Your patch itself applies cleanly but lacks e.g., variable-editor.h

Could you supply one changeset that includes the one from patch #8316 and yours that applies to the current dev sources?