Here's a frustrating WPF scenario — you use the ApplicationCommands class to add Cut, Copy and Paste commands to toolbar and then put a TextBox on another toolbar. Click in the TextBox and the commands remain disabled. WTF, WPF?

The problem is with focus scopes. Your window is a focus scope and so are any menus or toolbars. This has the desirable property of allowing commands to target the control you were in immediately before invoking the command. You want paste to target the text box you're editing, not the menu item or button you clicked to request the paste.

So far so good. The problem is that the commanding system isn't smart enough to target the control with keyboard focus if it's in a nested focus scope. Remember that the window itself is a focus scope so our TextBox in a ToolBar (also a focus scope) is nested and immune to commands from our menu or toolbar.

Here's a simple window that demonstrates the problem:

Ignore the PreviewCanExecute handler for now. If you run this window and click in the main TextBox the paste button and menu item are enabled. Click in the toolbar TextBox and pasting isn't an option. Well, Ctrl-V still works and there's a context menu but you know what I mean.

The problem can be fixed by adding a command binding for ApplicationCommands.Paste and handling the PreviewCanExecute event:

When the window loads we're making note of the focus scopes for the toolbar and menu. Then when PreviewCanExecute fires we check to see if the element with the keyboard focus is in a different focus scope (and also that the window doesn't have keyboard focus). We then set the CommandTarget for the menu item and button to the element that has keyboard focus.

A handler isn't required for CanExecute as the command will take care of this with respect to the new CommandTarget.

Run the window again and you'll see that the paste button is enabled for both of the TextBox controls. When you click the button (or menu item) our PreviewCanExecute handler ignores the new keyboard focus and the command is sent to the desired control. 

One drawback of this approach is that keyboard focus isn't returned to the TextBox after the command executes. The CommandTarget remains in place so you can keep pasting and the command remains enabled but you lose the visual cue that lets you know where the target is. I haven't figured out a clean approach to this yet. When I do, I'll update this post. Better yet, if you've figured it out leave a comment.

I'm slowly converting a number of blogs from Blogger to BlogEngine.NET. The least fun part is dealing with the Blogger export file. For this blog I used a Powershell script but had problems with comments not exporting correctly and it was quite painful to fix everything up. Blogger allows you to export a copy of your blog using ATOM, however BlogEngine.NET (and other tools) speak BlogML.

I've just released a command line tool that takes the ATOM format Blogger export and converts it to BlogML. You can download Blogger2BlogML from GitHub. The tool uses .NET 4.0 (client profile) so you'll need to install this if you don't already have it. If you give Blogger2BlogML a try let me know how you get on. 

I've just released a small update for my ESRI Shapefile Reader project on GitHub. The only change is a patch from SolutionMania that fixes a problem when the shapefile name is also a reserved name in the metadata database. The patch escapes the name preventing an exception from being thrown.

Catfood.Shapefile.dll is a .NET 2.0 forward only parser for reading an ESRI Shapefile. Download 1.20 from GitHub.

Links

The Volume Shadow Copy Service (VSS) takes a snapshot of an NTFS drive at a point in time. The clever thing about VSS is that it doesn't copy anything — it starts with the assumption that nothing has changed and then keeps track of every change to the snapshot so only changes need to be stored.

From Windows Vista on it's possible to mount a shadow copy as a drive letter or share. ShadowTask is a command line tool that creates a VSS copy, mounts it as a drive and then runs a program or batch file. For example:

ShadowTask C V dostuff.bat

Creates a copy of C:, mounts it as V: and then runs dostuff.bat.

Let's say you want to copy a locked file — maybe some outlook personal folders. Dostuff.bat could contain:

copy V:\Users\You\AppData\Local\Microsoft\Outlook.pst C:\Users\You\Desktop\OutlookBackup.pst

Bingo, you have a copy of your PST without shutting down Outlook.

Download: ShadowTask.zip (50.86 kb)

The ZIP contains both 32 and 64 bit versions of the tool. You must use the version that matches your platform. ShadowTask supports Windows Vista and 7. XP doesn't support mounting a shadow copy so ShadowCopy will fail if you try to use it on XP. ShadowCopy must run as admin (elevated).

PdfScan is a simple tool for scanning pages into a PDF file. You can scan single pages from a flatbed scanner or several pages from a document feeder. The page size applies to both the scan and the page(s) added to the PDF.

I wrote PdfScan because I know I'm going to be scanning a lot of documents over the next couple of weeks. Previously I used a tool called ScanToPDF from O Imaging but their licensing pissed me off so much that I'd rather waste time reinventing the wheel than pay them for another copy.

This is a beta — it works with my scanner and my documents. There's no installer, so extract the ZIP file and run the EXE to use it. PdfScan requires the .NET 4.0 Framework. If you get an error when you run PdfScan.exe try installing .NET 4 and then run it again.

If enough people use this I'll make it a bit more friendly, add an installer and release it through Catfood. If you like it leave a comment below. If it doesn't work for you leave a comment or email me and I'll try to help.

(Update September 12, 2010: I've tided PdfScan up and released it through Catfood Software. Download from Catfood PdfScan.)

PdfScan uses PDFsharp from empira Software. Thanks chaps!

I've been going nuts trying to scan from the document feeder on my Canon imageClass MF4150. Everything worked as expected from the flatbed, no dice trying to persuade the ADF to kick in. I found some sample code but it was oriented towards devices that can detect when a document is available in the feeder. Evidently my Canon doesn't expose this and so needs to be told the source to use.

The way to do this is to set the WIA_DPS_DOCUMENT_HANDLING_SELECT property to FEEDER. You then read WIA_DPS_DOCUMENT_HANDLING_STATUS to check that it's in the right mode and initiate the scan. This did not work for toffee.

After much experimentation I discovered a solution. I had been setting device properties and then setting item properties before requesting the scan. Switching the order - item then device - made everything work.

Here's the function to scan one page:

A few notes — XImage is a type from PDFSharp. I wrote this as part of a PDF scanner that I'll post next so the scanned images are saved and then loaded into an XImage for rendering to the PDF document. The magic numbers come from WiaDef.h in the Platform SDK. If the ADF is out of pages this method sets the return image to null and eats the exception. This is because the function is called repeatedly to scan in pages until the ADF is empty if _adf is true (otherwise it grabs one image from the flatbed). 

If you've been banging your head against a wall trying to get WIA to work with a document feeder I hope this helps.

Updated 2015-05-20: Full source code at https://github.com/abfo/pdfscan

After floundering a bit with the WPF Dispatcher I've come up with a simple way to make sure an event handler executes on the UI thread without paying the overhead of always invoking a delegate.

This has the benefit (for me at least) of being very easy to remember. Hook up the event handler and then if there's a chance it could be called from a different thread wrap it using the pattern above. It's easier to read than an anonymous delegate and much faster than defining a specific delegate for the event in question.

I haven't tested the various methods to see which is the fastest yet… will get round to this at some point.

5. Validation

The metadata required to use the like button looks like this:

But the property attribute isn't valid html or xhtml. The “Open” Graph Protocol says that it's inspired by Dublin Core. DC manages to get by using the name attribute like any other meta tag - why can't Open Graph? It's not the worst problem but it just seems needlessly irksome. Facebook has published a presentation describing their design decisions. This would be great, but it's in that Lessig one word per slide style and so it's attractive but completely useless without the presenter.

4. Fragility

Facebook's documentation is frustratingly sparse. For example you need to specify the owner of the page using a Facebook ID, and once you've chosen a name for your profile this is hard to find. The information vacuum has been filled with many erroneous blog posts saying to use the name, or some number from a shared photo (the best source is http://graph.facebook.com/robert.ellison, substituting your own username). Once you've got the admin ID wrong, you can't correct it - the first admin specified is fixed forever. What happens if a site is hacked and a bad actor sets themselves up as the admin? Surely something like the Google Webmaster Tools authentication scheme could have been used instead?

3. Pages with more than one object

Describing the object being liked in the head element limits you to one object per page. For some sites this is perfect, but what about a blog where you have many posts on the home page? It would be useful to have a like button per post, pointing at the permalink for the post in question. I've worked around this by having a like button for the blog on the home page, and a like button for each post on the post pages. Not ideal. I'm using the iframe version of the gadget, possibly there's some more flexibility with the XBML variant.

2. Duplicating existing pages

Let's say you've spent the past couple of years building up a Facebook page for your site/band/blog/movie and have thousands of fans. When you click your new like button for the first time you create a whole new page. There's no way to tell the like button about the existing page or the existing page about the like button. You now have at least two pages to worry about managing and potentially many, many more. You're also starting from scratch on the ‘like’ count, so even if your brand is already popular on Facebook it's back to Billy no-mates for you.

I can't believe this won't be fixed at some point. As with admin authentication above there must be a better way to establish ownership of various objects in the social graph.

1. Vocabulary

For better or worse Facebook has the inexorable pull to start making the semantic web a reality. Given this, and that there are something like twenty-four thousand verbs in the English language it's time for more expressiveness than ‘like’. You also can't comment on the ‘liked’ item in your stream (yet) so no clarification or discussion is possible.

--

Having said all that, if you enjoyed this post please click the ‘like’ button above ;)

Links