The challenge is this: 2 coordinated tables, 1 with an overview that is laid out on a grid, the other contains detailed information about the cells in the first table. (This overview is used for other information as well, which has been removed from the minimal example below.) Mousing over either one will highlight both. Clicking on the overview table will hide or show the rows of the data view.
The problem is that the data is being defined by a JSON data object and the order of that object is very important. It's important that the data is mapped to the overview without reordering either. There may be cells not represented in the data view that are in the overview and nothing bad should happen.
http://fiddle.jshell.net/g8z5h/14/
The problem is using select all lets d3 define the order things will be taken. I need to coordinate the mapping myself. I'm hoping there's an elegant solution to this that doesn't involve writing separate mouseover and mouse click functions.
You can use a key function to tie the selection to your data:
http://bost.ocks.org/mike/selection/#key
http://bost.ocks.org/mike/constancy/#key-functions
Related
first post to StackOverflow, though I've been finding great tips for a few years here. Hope to resolve a vexing problem with handsontable.
I have an application that presents a handsontable instance with several columns whose content/format is dependent upon the value in a cell in another column. I determine the content and format of the dependent cells by way of a database query and server-side processing, the results of which are sent back to handsontable for processing in the afterChange function.
My problem is that whenever the user updates one cell in the table, ALL cells blink while the code laboriously reconsiders the formatting for ALL cells in the table. The preferred behavior would be to ONLY update the cells for which the formatting has changed, i.e., that small set which is dependent on the value that changed in the first cell. It's only these dependent cells whose contents are being changed anyway. Why redraw the WHOLE table??
I have searched and search here and elsewhere, and the closest to an answer I could find is that this is the intended behavior for the table. I don't agree...when the table is anything larger than 10 rows, the updating process is painfully slow and distracting to the user.
Any suggestions? I'm open to directly editing the handsontable js code, but would prefer to flip some flag I'm unaware of if possible.
Thanks all!
afterChange fire only one time for the cell you just updated (documentation example).
Your problem might come from an afterChange "chain reaction". The first afterChange modifies one or more related cells, triggering others afterChange and so on. If that's the problem, you can put a filter on the source parameter.
Another solution could be to work on a copy of the data to do your updates (see data binding), and once everything is fine, inject it back on the table with loadData, and ignore the loadData source in the afterChange callback.
Finally, if you still have some performance issues, you can check the performance tips, and remove any option that might slow your table (example with columnSorting)
(sorry I can't use comments to identify precisely from where your issue come...)
So before I continue developing my accessible table I read this, on exposing data tables and the general MSAA documentation and I want to clear up a few things before I actually go ahead and write out this implementation, just so I can get everything just right.
For the record, this table is strictly owner-data and is row-based, where a given row can be selected, and a single cell within this row can be focused. Otherwise it behaves similarly to a listview control, but is not one, so I'm not going to try to take advantage of its accessibility features. In particular, any column can contain text, images, or checkboxes (and I might add more features); there might be background colors and text editability in the future as well. All cells in a given column have one of these types. The type will not change during runtime.
I am choosing to use MSAA because I still need Windows XP support, and I would rather not require developers and end users to have both SP3 and the .net framework together, which UI Automation seems to require (and even if the SP3 part is wrong, the "is not standalone/does not come out of the box with the OS" part still makes me uneasy about implementing that).
Is it process-safe/thread-safe to call my internal control functions within the accessible object method implementations, or should I use WM_USER messages to make sure everything happens all on one thread of one process? (In addition, would my error handling/debugging traps be run in the client or in the server in either case?)
From what I can gather, I need to build a three-level IAccessible hierarchy: the table itself on top, followed by each row, followed by each cell in the row. However, the various IAccessible methods that take a child object only take a VT_I4 VARIANT to specify the child. Does this mean I'm going to have to create an IAccessible for each row as well, or for each cell as well, or something else? Should I have a standard accessible object for each of these? Can they share? And what about NotifyWinEvent(); how would I indicate that a given cell has changed? Or should I say the whole row has changed...?
If the answer to question 2 is "yes", I'm going to wind up with a lot of IAccessible objects that need to be notified when the table control is destroyed. If the answer to question 1 is "it is not thread-safe", then is it safe to hold a lock in my IAccessible methods whenever I access my table control and have that hold the lock when it's time to invalidate all those IAccessibles? Or should I investigate some other approach? I see talk on MSDN of "proxy objects" but I'm not really following how that would help... especially if I need to make lots of IAccessibles.
From what I can gather I need to return (number of rows) + 1 rows in my table's get_accChildCount() method, with the first row being full of column header cells. But I'm not using a custom header control; I'm using the standard comctl32.dll header control. Should I hide that control and construct my column headers according to the Exposing Data Tables document or should I relegate the first row to the header control? I don't see a way to do both... (I do not have row headers, so will not be implementing that.) Columns can be user-resized, but may be automatically sized to fit until the user first does so (maybe).
Is it safe to relegate IDispatch methods to the window handle's standard accessible object? Even in the case of the row and cell IAccessibles? Or will the standard accessible object's IDispatch not call my IAccessible methods?
What should accDoDefaultAction() do for a checkbox cell, toggle the checkbox state? And for an editable text cell, should it enter editing mode?
Thanks. (Hoping none of this sounds dumb...)
UPDATE 6 January 2015
I changed the wording of some of the paragraphs above and have one more question:
How do I correctly check for errors from LresultFromObject(), by casting to an HRESULT and checking that, or by comparing a signed version of the LRESULT value and seeing if it's less than zero, or something else?
This is sort of a best practice question, since I can think of a few ways that would work.
I want to implement an outline view. Let's suppose I want to implement the one in OmniFocus (my aim is very similar):
(I refer to the outline view in the main pane of that screenshot, not to the sidebar.)
So my data type is a task. Any task can have subtasks. Each task has one or zero parent tasks. Classic data model for displaying in an outline view, right?
BUT! I would like to offer the user the ability to group these tasks visually by a property of their choice. They could group them by milestone, or by assigned user, or by component… there'll be a bunch of these. And this grouping should appear in the outline view, at the top level of the hierarchy, just like the "Inbox", "Home" and "Work" items in this OmniFocus screenshot.
So it's fairly obvious that to display the grouping, I should use the built in NSOutlineView methods for drawing a group cell: the outlineView:isGroupItem: delegate method. The problem is, this setup is assuming that each item in the outline view is represented by an item in whatever data model I've connected it to: so both my tasks and my group headers have to be represented in the data model.
Therefore, just binding to the Core Data table that represents my tasks is a no-no. I could forego bindings all together and just go back to the old style data source methods, but then I lose a bunch of useful stuff. So is there a middle ground?
I could, for example, create another class TaskTemporaryGroup. I give that class an ivar NSSet that is subtasks, the same key path as could be used on a task to get its children. Then I can bind my set of TaskTemporaryGroup objects to a tree controller, and it'll arrange the whole tree regardless of the fact that the top level items are a different class to those lower down. I'd have to take care when implementing drag and drop, to make sure that the grouping rows can't be arbitrarily dragged about, but it might work.
The other point of contention is that I would like clicking on the table headers to sort the rows within each group, but not to sort the top level groups. I assume NSTreeController applies the sort descriptors to each level of the hierarchy? I could have my TaskTemporaryGroup class return the same thing (i.e. its desired sorting order, that I've calculated separately) for any of the key paths that I want to be able to sort the rest of the cells on, so that to the framework it would look like they stayed in the same order no matter what property they were sorted by. Is this a good way to go or is it a hack?
So, to summarize:
How do people suggest I implement this kind of hierarchy, where the top level is a special case and all subsequent levels are the same kind of object?
Is the way I described, with a class to represent the top level objects that happens to respond to the same key paths as the child objects, a good way to go?
Does anyone have any tips about how to make sorting work well in this scenario?
Or will trying to use bindings for this task cause me a world of hurt?
Thanks,
Amy
I’m working on an inventory display app (master/details) for Windows Phone 7 where the user has the ability to change the sorting. The source of the data is XML from the web, and depending on the sort chosen, the results are either a flat list (sorted by name, model number, etc.) or grouped by the selected attributes (color, size, etc.). I'll pre-determine whether the sort returns flat vs. group and it won't be user selectable.
I’m displaying the results in a WP7 pivot control and want/need the different sorts to appear in the same PivotItem (putting the sorts in different PivotItems isn’t really an option). The source for the list is an Observable Collection (actually at this point it’s a CollectionViewSource, but I’m not wedded to either). I’m using the MVVM Light framework.
I’ve played with LongListSelector from the Silverlight Toolkit and it looks like it does what I want. I've read a number of posts (starting with WP7 ListBox Grouping) and actually have it working for the flat list (no sort applied, just default as it is from the source). Now I’m trying to figure out the right way to use it.
My questions:
Is it possible to re-structure the way the data is displayed (flat/sorted vs. grouped) without having to build/maintain multiple observable collections of the same info (one for each sort variant)?
If not, is the best method to save a local copy of the data locally on the phone and work from there? I expect I'd take a hit on complexity but pick up savings on perf/responsiveness.
To display the various results in the same PivotItem, should I create user controls with the appropriate formatting and swap them in to the pivot based on the chosen sort, or can I do it via dynamic ItemSource & ItemTemplate changes?
Is Observable Collection/CollectionViewSource the way to go here?
It’s entirely possible (even likely) the answers are staring me in the face, but I’m new to this stuff and I’ve hit data overload. I could really use some guidance.
Thanks,
Rich
Is it possible to re-structure the way the data is displayed (flat/sorted vs. grouped) without having to build/maintain multiple observable collections of the same info (one for each sort variant)?
I think what you could try is using LINQ to query your given collection, and use that query as the source for your LLS. I'm not sure on how the performance would be, but you can give it a try. Alternatively, IIRC the various *ViewSource collections have ways of changing the view of your data, so you can check those out.
I have a datagrid with many columns. This makes it pretty wide. Now we want to add more information to the table. Aside from removing or shortening existing columns what are some ways we might be able to add additional information without adding new columnes.
The data we want to add would be one of several values. For example:
Projected
Actual
Other
For other cases when the value was an off/on or true/false we would change the color of the row. In this case that doesn't seem to be a good option.
Another thing we considered is using an icon to indicate the information.
Any other ways this could be done?
A solution i've seen implemented with grid components is to have a column chooser - some sort of popup dialog that lists the columns and you can select which ones you would like to see in the grid. You should be able to invoke this popup by triggering it from the grid, e.g. it might appear as an option when the user right clicks and causes the context menu to appear.
Can you group related information into tabs?
an overflow area? ie a number of fields underneath the table that populate based on the selected row.
or just only show the minimum needed info and the have full details in a popup when doble clicked or something..
1) Popup on row hover
2) Drop open inline in the grid with extra info on row click
One technique I've used in the past was to create a "container" type of class that has its own labels and textboxes, and you can arrange them however you want, then insert this class into a single grid column. You still have to do some tricks on binding multiple controls that are not native "grid column" controls, but should help you along. Then, you can actually have each row a single container control in a single grid column...
You can't add completely new data to a grid without reserving a column to display it. The best solution I've seen is to provide only the essential information in the grid displaying all records, and then create a drilldown view that shows all of the data for one row. The drilldown can either be a new view in the same form, a popup for an additional window, or perhaps a mouseover popup.
I've worked on systems that use all sorts of shortcuts to display every last bit of information on a single page, and I found that it just made everything more confusing and harder to use. "Oh, that little icon there means that <insert something totally unrelated to the icon picture>."