Starting with version 3.0.0.3010, the C# code completion in SharpDevelop has support for implicitly typed lambda expressions.

Given a variable "IEnumerable<MyClass> items" and the "Select" extension method from LINQ, typing "items.Select(i => i." now shows the members of MyClass. And if the result of the Select call is assigned to an implicitly typed variable, SharpDevelop is now able to infer that the variable has the type IEnumerable<return type of the lambda expression>.

Unlike all other expressions in C#, the type of a lambda expression cannot be inferred just from by at the expression itself (and the variables used by the expression). To resolve lambda type parameters, we also need to look at the context where the lambda is used. Currently, not all contexts are supported by code-completion, you can find the list of known problems in our bugtracker (Component: DOM / Resolver). Should you find anything where code-completion does not work correctly which is not in that list, please file a bug report in our forum.

The most commonly used context for lambda expressions is method calls, and this is also the most difficult thing to support. It's easy when the method has a clear signature like "void M(Func<int, string> f)", since then SharpDevelop can infer the lambda parameter types directly from the delegate type. But most of the time, things aren't that easy. For example, the signature of the Select method is "IEnumerable<R> Select<T, R>(this IEnumerable<T> input, Func<T, R> f)". Here, SharpDevelop needs to first infer what T is, then it can know what the lambda parameter types are, and only after that it can resolve the lambda expression to infer what R is.

But when the method has multiple overloads, things can get even more messy:
When a method has to overloads "void M(Func<string, int> f)" and "void M(Func<int, int> f)", it is valid to call them like this: "F(i=>i.Length)", "F(i=>i+i)". In the first call, i is a string; in the second, it is int. What SharpDevelop needs to do here is to infer the lambda parameter types for each overload separately, infer the lambda's return type; and then check that against the delegate's signature to see which overload was the correct one.

i=>i.Length is a resolve error if i would be int, but returns the expected int if i is string; so i must resolve to string.
i=>i+i returns a string if i would be string, but returns the expected int if i is int; so i must resolve to int.

Note that because there's no way to tell the type of i before the lambda expression is completed, you cannot expect that SharpDevelop gives you correct code completion for it. "Length" will not be included in the code-completion list for i when you type ".", because at that point, the method call is "F(i=>i)", and i is thus an int. But after the expression is written, SharpDevelop will show a tooltip for "Length", and features like "Go to definiton" and "Find references" will work.

In addition to our presence on SourceForge (project page) we now also set up shop at CodePlex (project page). The motivation for this step is to increase awareness about SharpDevelop, and how parts of SharpDevelop can be reused in other applications / scenarios.

In SharpDevelop 1.1, the IClass interface had a property that was used in several places in the code:
Once added to a project content, it was immutable. This was not enforced, not even documented. It just happened that no one changed IClass objects except for the code constructing them. After being added to a project content, a class could be removed or replaced by a new version, but if some code still held a reference to an old instance, it could safely access the members without worrying that an update to the class on another thread changed something.

IClass objects are accessed without locks all over the place on the main thread during code completion, and updates on the parser thread should not interfere with that.

However, because I didn't understand this, I broke it in SharpDevelop 2.0. My implementation of partial classes works as following: each file (compilation unit) contributes a part of the class as IClass object representing that part of the class.
Once multiple files register a part of the class in the project content, the project content creates a compound class. This compound class combines the members of all the parts. When updating a part, only the IClass for that part was recreated, and the existing CompoundClass was updated.

The CompoundClass, which could be in use by multiple threads, changed values. To quote Wes Dyer: "Mutation often seems to just cause problems."

Now, that happened to work correctly for quite some time. Most code iterating through a class' members did this with
foreach (IMethod method in c.Methods) {
}
where c is an IClass (possibly a CompoundClass). An update of the compound class happened to recreate the List<IMethod> behind the c.Methods property, so this code continued to work as expected.

However, in the quick class browser (the two combo boxes above the code window), there was code similar to this:
list.AddRange(c.Methods);
list.Sort();
list.AddRange(c.Properties);
list.Sort(c.Methods.Count, c.Properties.Count); // sort the properties without mixing them up with the methods

Suddenly, due to the addition of partial classes, this became a race condition waiting to happen. But it was found in time for the SharpDevelop 2.0 release, and I fixed the crash.
But I didn't know much about immutability back then, so what I did was the worst fix possible:
lock (c) { ... }
And in CompoundClass, during update of the parts: lock (this) { ... }

Now, this is not only bad because it's fixing the symptom instead of the problem and it leaves the possibility for similar problems elsewhere in the code - though it might have been the only instance of the problem, since no other crashes due to this have been found while SharpDevelop was using the fix (all 2.x releases use it, including the current stable release, SharpDevelop 2.2.1).

But multi-threading (without immutability) is not hard, it's really, really hard. So it's not really surprising that some day, I found this code to deadlock.

So where's the deadlock?
First I must tell you the other lock we're colliding with: every project content has a lock that it uses to ensure that GetClass() calls are not happening concurrently when the list of classes is updated. So the parser thread acquires the project content lock and then the CompoundClass lock to update the CompoundClass.

But why would the AddRange / Sort code deadlock with this? The comparer used for list.Sort() sorts the members alphabetically using their language-specific conversion to string. In the case of methods, this includes the parameter list, including the parameter types.

What you need to know here is that type references (IReturnType objects) are not immutable - they need to always reference the newest version of the class, as we cannot afford rebuilding all IReturnType objects from all classes in the solution whenever any class changes. Now remember that C# allows code like "using Alias = System.String; class Test { void Method(Alias parameter) {} }".

In this case, the quick class browser correctly resolves the alias and reports "Method(string parameter)". This means that our Sort() call actually sometimes needs to resolve types! And resolving types works using IProjectContent.SearchType, which locks on the project contents' class list lock. And that's our deadlock.

I think it's near impossible to find this kind of deadlock until it occurs and you can see the call stacks of the two blocked threads.
Remember that the actual Method->string conversion and the type resolving is language-specific; it may or may not happen to take a lock for other language binding AddIns.

I fixed the deadlock on trunk (SharpDevelop 3.0) a few months ago by removing the lock on CompoundClass and instead doing this:
list.AddRange(c.Methods);
list.Sort();
int count = list.Count;
list.AddRange(c.Properties);
list.Sort(count, list.Count - count);

It's still a hack, but this doesn't have any side effects (like taking a lock). And it works correctly under our (new) rule (undocumented rule, aka. assumption) that multiple c.get_Methods calls may return different collections, but the collection's contents don't change.

"foreach (IMethod method in c.Methods) { ... }" is safe, but "for (int i = 0; i < c.Methods.Count; i++) { IMethod method = c.Methods[i]; ... }" can crash.

But that's quite a difficult rule compared to "IClass never changes". So after reading Erip Lippert's series on immutability, I finally decided to make IClass immutable again.

It's "popsicle immutability", that means IClass instances are mutable, but when the Freeze() method is called, they become immutable. And this time, immutability is enforced, trying to change a property of a frozen IClass will cause an exception. Adding an IClass to a project content will cause it to freeze if it isn’t already frozen, so it's guaranteed that IClass objects returned by GetClass or by some type reference are immutable.

A change that happened rather early in the development process of SharpDevelop 3.0 (revision 2658, 8/13/2007) was that we replaced NDoc (a stalled open source project) with Sandcastle Help File Builder (SHFB). SHFB looks and feels similar to NDoc, however, it builds on top of Sandcastle, a documentation generation tool by Microsoft.

Our build server (revision 2913 and higher) contains SHFB 1.6.0.4, which itself builds on top of the January 2008 release of Sandcastle (both are at the moment the latest releases of their respective projects). SharpDevelop 3.0 setup ships with SHFB, however, Sandcastle itself does not and must be installed separately! (this might change in the future once the final license of Sandcastle is known)

Because of the regular releases of Sandcastle and SHFB, the distribution of SharpDevelop 3.0 might ship with older versions of SHFB than currently available for download. Thanks to Eric Woodruff, the maintainer of SHFB, this is not a big deal - he is offering a download specific for SharpDevelop:

All you have to do is first delete the contents of the following installation directory (including subdirectories):

Then simply unzip the SHFB distribution archive into this folder and presto - you are now using the latest version of SHFB from inside SharpDevelop!

This question came up for example in the thread SharpDev 2.2.x on BuilderServer (this thread was started because we ceased to automatically build v2.2 on 1/11/2008).

The answer is Yes. Development of SharpDevelop 2.2 stopped with revision 2675 (8/28/2007), which is three revisions higher than the officially shipping version of SharpDevelop 2.2 (Download, 8/8/2007). The three non-shipping commits are:

• 2673: Improved CSharpCodeCompletion sample: add tool tip support, show only one entry for overloaded methods
• 2674: Fixed some off-by-one bugs in the CSharpCodeCompletion example (caused by the different line counting in the parser and the text editor).
• 2675: CSharpCodeCompletionSample: show xml documentation

All three were (of course) merged into Montferrer (SharpDevelop 3.0), this merge happened in revision 2679. However, those commits did not merit a release of a new setup because those were all changes to a sample shipping only in the source download.

Since releasing v2.2.1 all work stopped on the 2.x series of SharpDevelop. Our efforts went (and still go) into SharpDevelop 3.

In continuing to list changes to SharpDevelop 3, we are going to talk about the code coverage addin in SharpDevelop 3 "Montferrer" in this blog post.

Previously, the addin used NCover for calculating code coverage (this is a metric you gain by writing unit tests). However, recently NCover was turned into a commercial product. Because we only include / support tools that are free to use for anyone (commercial or open source / hobby development), we switched to a different tool - PartCover.

This change happened in rev 2744 on 19th of November last year. The addin has retained its original functionality, however, for end users there is an important change: you no longer need to download and install a separate package (as it was the case with NCover), PartCover is part of the SharpDevelop setup. You are good to go right after installation!

An issue that initially came up in 2006 (Unable to compile #develop: access denied) "resurfaced" on our contributors mailing list because one of our developers ran into this very problem that McAfee blocks access to our Main\StartUp folder:

The problem and a workaround is described in McAfee VirusScan and the Startup folder. However, this developer doesn't have the administrative rights to change this setting of McAfee, so he asked whether we are going to rename the folder.

My take on this issue is that this is a bug in McAfee's software, because willy-nilly disabling anything that resides in a random folder called StartUp isn't a security feature.

Over the past months we made a couple of feature changes in SharpDevelop 3 (currently alpha status). One major change is that we moved NAnt and Mono support from the binary distribution (aka "setup") to the source code distribution. You can find the addins ready to build in the \samples directory:

What was the reasoning behind this decision? For NAnt, we had taken this decision a long time ago because SharpDevelop itself no longer uses NAnt as the primary build solution, and as such, the addin wasn't actively enhanced and tested. But it still is a great tool as well as a good sample for building addins with SharpDevelop.

The decision to "relegate" Mono from production to sample status has been based on multiple factors. For one, we only support basic compilation for Mono, no debugger nor any kind of visual designers (like GTK#). We got lots of support questions regarding these, and the honest answer had to be "we won't support that, sorry". Then in December Miguel announced that MonoDevelop will come to Windows (MonoDevelop is a fork of SharpDevelop), which meant that an IDE would come to Windows that fully supports all the things in Mono we don't have.

That's why we decided to make Mono an addin for people who know how to deal with source code, all the features are still there. And now that it is separate, it also makes a great sample addin because of the deepness of integration with low-level features of SharpDevelop.

To sum it up - we didn't remove anything, we just trimmed the setup to include features that are targeted at the Microsoft .NET platform.

Microsoft just released Visual Studio 2008 and the .NET Framework 3.5.

Download .NET Framework 3.5

To develop applications for the .NET Framework 3.5, download SharpDevelop 3.0 from the build server.

Note that SharpDevelop 3.0 are not release quality builds, especially the integrated debugger is very unstable. As currently no one is working on the debugger (hint: you can help!), I disabled it in SharpDevelop 3.0.0.2745 so that people trying SharpDevelop 3.0 do not immediately get exceptions when they try run their programs. To re-enable the debugger, open C:\Program Files\SharpDevelop\3.0\AddIns\AddIns\Misc\Debugger\Debugger.AddIn.addin in a text editor and remove the <DisableAddIn .../> node.

Daniel is currently working on areas such as project subsystem or code completion (aside from fixing bugs). Therefore, work on the WPF designer is stalled. But Daniel put together a list of tasks (hard, medium & easy) that you could be helping us with. So if you are interested in helping, and want to be part of writing an open source WPF designer, please do get in touch with me at christophw at icsharcode.net!

WPF Designer mini tasks

Here are some jobs to do on the WPF Designer broken into small parts.

Apart from this list, there are also big jobs to do: Data Binding support, support using Styles, support defining resources.

Components to be used by multiple designer parts

Create “Choose Class” dialog

The data binding UI will need a dialog that allows choosing a class from the referenced assemblies. This will be implemented in WpfDesigner.AddIn and made available to WpfDesign.Designer through a service. The referenced assemblies should be inspected using ICSharpCode.SharpDevelop.Dom

This will be used by the “Content” property editor and by the data binding UI.

Support ‘Virtual Design Root Element’

This means that instead of displaying the designed component, the designer is able to display another component as if it was the root component. The designer should provide a “Back” button on the design surface to allow the user to go back to the real root component.

This feature is required for designing elements that cannot be designed directly on the main design surface, e.g. Tooltips

Write menu designer

Priority: High

The menu designer might be a dialog box that allows editing the menu in a tree view; or it might be an in-place menu designer like the Windows.Forms designer.

Property editing support

Here I’m listing properties that could need improved editing support.

When implementing property editors, take care that the editor should also work when multiple components with different property values are selected, so you always need a way to represent the ‘ambiguous’ value.

Properties of type Brush (e.g. Control.Background / Control.BorderBrush)

Priority: High

Change ICSharpCode.WpfDesign.Designer.Controls.TypeEditors.BrushEditor to include a little drop down button, the drop down should allow to choose the brush type (SolidColorBrush, etc., we don't need to support all of them at the beginning) and allow the user to edit the brush according to the chosen type.

Properties of type BitmapEffect / BitmapEffectInput (e.g. UIElement.BitmapEffect / UIElement.BitmapEffectInput)

Priority: Low

Implement a TypeEditor (similar to BrushEditor) for the types BitmapEffect and BitmapEffectInput. Could be as simple as a combo box with the most commonly used effects.

Properties of type ICommand (e.g. ButtonBase.Command)

Priority: Medium

There should be a way to choose the command to use from some kind of list.

ContentControl.Content

Priority: High

ICSharpCode.WpfDesign.Designer.Controls.TypeEditors.ContentEditor
The “C” button should be made a drop down button (like the one used for the Brush editor), it should present “null”, “string”, and menu items for creating commonly used child configurations (e.g. StackPanel with Image and Text when the parent is a button), and “choose class” to create arbitrary objects. This depends on “Create Choose Class Dialog”.

Properties of type ContextMenu (e.g. FrameworkElement.ContextMenu)

Priority: Medium

Provide a way to create and edit a context menu inside the designer.

Properties of type InputScope

Priority: Low

Research what an InputScope is and if/how we should allow the user to edit it.

Properties of type Transform

Priority: Low

Similar to the BrushEditor, provide a drop down to choose from the different available transforms and allow editing the transform properties.

FrameworkElement.ToolTip

Priority: Medium (setting string tool tips), Low (designing complex tool tips)

Should be editable similar to ContentControl.Content, but has to allow the user to design complex tooltips. Depends on “Support ‘Virtual Design Root Element’”

Properties of type FontStretch / FontStyle / FontWeight

Priority: Low

Provide a drop down with the available settings

Properties of type FontFamily

Priority: Low

Use a dialog to allow the user to choose the font to use.

Properties of type Nullable<bool> (e.g. ToggleButton.IsChecked)

Priority: Low

ItemsControl.ItemsSource

Priority: Low

There are two main usage scenarios: this property is specified using data binding (this doesn’t need to be handled by the type editor), or there are some hard coded values.

Write a type editor to support entering string values.

Label.Target

Priority: Low

Label.Target is set to the control described by the label (the control getting focus when Alt+Access Key is pressed). This is done using data binding, but choosing the target from the data binding dialog is too tedious – drag’n’drop of a crosshair on the target control would be much easier. Depends on data binding support.

Properties of type ImageSource (e.g. Image.Source)

Priority: High

Allow choosing a “Resource” element already part of the project, or choose a file and it will get added to the project. This TypeEditor would be implemented in WpfDesign.AddIn and not WpfDesign.Designer because it needs access to the project in SharpDevelop

Properties of type ViewBase (e.g. ListView.View)

Priority: Low

Provide a drop down with the most commonly used views.

SD2-1234, titled "Create common way to handle in-memory representations of files that have multiple views" is the issue tracker entry behind a major refactoring of the IViewContent interface.

The major new feature introduced is that now it is possible to open a file in multiple view contents at the same file (using "Open with"). Both IViewContent instances will edit the same underlying file - when you switch between them, one view content will display the unsaved changes of the other; pressing Ctrl+S in one of them will save both.

Now how is this possible? There must be some data structure shared by both view contents. Since the view contents might not know anything about each other (both could be independently developed AddIns), this data structure must work on the lowest possible level: bytes. It's simply a MemoryStream.

Of course the view contents cannot serialize their content on every change; so a view content may have local changes in a higher-level data structure (text editor buffer, list of entries in resource file, etc). To ensure that such changes are correctly synchronized between view contents showing the same file, I introduced the concept of an active view content. The active view content is the only view content that may have such local changes. Thus, the active view content always has the most recent version of the file, and it is the only view content that is guaranteed to have the most recent version.

To represent files, I created the OpenedFile class. This class knows which view contents have opened it and knows which of them is the active view content. Usually every OpenedFile can have a different active view content; though it is possible for multiple OpenedFiles to have the same active view content if that view content opens multiple files - the refactoring also added support for view contents that are editing multiple files at the same time.

Now a quick overview of the SharpDevelop UI: The SharpDevelop main window is composed of a main menu, a tool bar, and the DockPanel of Weifen Luo's DockPanel Suite. In the dockable area, there are pads, which can be docked in groups at the sides of the SharpDevelop window; or can float somewhere (e.g. on a second monitor).
In the remaining space (not occupied by pads), workbench windows are displayed. The active workbench window can be chosen using the tabs at the top. There can be multiple visible workbench windows if a user drags a tab and docks it to the side of the remaining space.
Finally, each workbench window contains at least one view content. The active view content can be chosen using the tabs at the bottom.

So if you create a new Windows Application and open Program.cs and MainForm.cs, there are two workbench windows titled "Program.cs" and "MainForm.cs", but three view contents - "Program.cs - Source", "MainForm.cs - Source" and "MainForm.cs - Design".

Note that in the new IViewContent architecture, there are no secondary view contents. Now all view contents are equal, and they shouldn't care if they share a workbench window or not. Whether they will share a workbench window or not depends on how the view contents were created - there is still the difference between primary and secondary display bindings, which are responsible for creating view contents.

Now how do changes get from one view content to the other? It's quite simple: Whenever the user activates a view content (by clicking a tab at the top, by clicking a tab at the bottom, or by setting focus into another workbench window after the user docked a window to the side), that view content becomes the active view content for all files it has opened. The old active view content will be asked to save its content to a MemoryStream and the new active view content will be asked to load from that MemoryStream. This way, unsaved changes are transferred from one view content to another.

When the active view content for a file is closed, but there are other open view contents for that file, SharpDevelop will not ask the user to save the file. Instead, it will save the data from the view content being closed into a MemoryStream. After that, the OpenedFile has no active content. Only when one of the other view contents that have opened that file get activated by the user, that view content will load from the MemoryStream and thus will preserve unsaved changes from the closed view content. However, if the user closes all other view contents for that file without making them active (by middle-clicking, or using Window>Close all), SharpDevelop will ask the user if the file should be saved and write the MemoryStream content to disk if required.

The system sounds simple for view contents: they just have to be able to load and save; and it'll just work.

But it isn't that easy. The view contents must be able to load and save reliably at any time.

The user just did something invalid which cannot be saved and then switches to another view of the file? The view content is forced to save. It's not possible say "I don't want to save".

The user loads a .resx file in the text editor, changes something by hand that renders the file invalid XML, opens it in the resource editor, gets an error message, switches back to the text editor. Here the file is saved by the text editor, loaded in the resource editor, the user gets the error and switches back, the resource editor must save and the text editor will load again. The resource editor view content must support loading and saving invalid files unless you want this kind of round-trip to result in loss of user data.

If your view content is editing multiple files, it gets even more complicated: you must support loading and saving individual files reliably at any time, in any order. Sounds fun, right?

Well, if you don't get this right, the user looses data only when editing a file in multiple views simultaneously. In SharpDevelop 2.x, view contents were simply overwriting each other's data; in SharpDevelop 3.0 there's at least a chance that it works if all view contents are implemented correctly.

However, be warned that I didn't have the time to update all view contents in SharpDevelop to make use of the new model. There's still a class AbstractSecondaryViewContent that implements Load and Save so that they run through an underlying "primary" view content, so existing secondary view contents do not have to be completely rewritten (although they still need several changes). The text and resource editors are fine; use them to see how it should work. The forms designer does not yet use the new model, it uses AbstractSecondaryViewContent and still touches the Designer.cs file directly, resulting in bugs like SD2-1175.

But if you are writing a new view content, try to design it so that you can support loading and saving at any time. The AbstractViewContentHandlingLoadErrors class (which both the resource editor and the WPF designer use) can help you handling invalid files.

If your view content edits multiple files, it can get tricky to support loading and saving those independently. But if it is likely that a user will want to edit one that files separately while also using your multi-file view content, you will have to do it. It is possible in SharpDevelop 3.0, so that's an improvement over SharpDevelop 2.x.

By the way: the reason for all this is the settings designer (still not implemented): it edits both a .settings XML file and app.config, and it's very likely that the user has opened the app.config at the same time.

Post by Daniel Grunwald (we use the category to mark the author on this blog, but I'll repeat it from now on at the bottom of the post because some feed readers like Google Reader don't show the category)

Visual Studio 2008 uses MSBuild 3.5 which supports multi-targeting: you can use it to compile applications for .NET 2.0, .NET 3.0 and .NET 3.5.

But what happens if you open an MSBuild 2.0 project (Visual Studio 2005 or SharpDevelop 2.x project) in Visual Studio 2008?
Answer: The “Visual Studio Conversion Wizard” will pop up and “convert” the project to MSBuild 3.5. Though it does generate a fancy upgrade report that looks like source files have been converted, all it does for a normal Windows Application is the following:

• In the .sln file, the solution version is set from 9.00 to 10.00, and the “# Visual Studio 2005” line is replaced with “# Visual Studio 2008”. This causes the Visual Studio loader to run VS 2008 instead of VS 2005 when the .sln is double-clicked.
• In the .csproj file, the attribute ToolsVersion=”3.5” is inserted. This causes MSBuild to use the C# 3.0 compiler.
• The property “<OldToolsVersion>2.0</OldToolsVersion>” is inserted, and two properties FileUpgradeFlags and UpgradeBackupLocation are introduced, but not set to a value. This has no effect on compilation.
• In the .csproj file, the Resources.Designer.cs file got marked as “<DesignTime>True</DesignTime>”, which has no effect on compilation.
• All files generated by a custom tool (Resources.Designer.cs and Settings.Designer.cs) get regenerated.
• All other source files remain unchanged.

Note that even though the C# 3.0 compiler is used, the project is still compiled for .NET 2.0. The new C# 3.0 languages features are available, but LINQ cannot be used because it requires referencing System.Core.dll, which comes with the .NET Framework 3.5.
So the converter did only minor changes, but they prevent working with both VS 2005 and VS 2008 on the same project.

Currently, SharpDevelop does not use the multi-targeting model of VS 2008, but continues to use the model of SharpDevelop 2.x: The “Target Framework” setting in the project options allows you to choose the compiler to work with. If you choose C# 2.0 / .NET 2.0, SharpDevelop will not use ToolsVersion=”3.5” on the project, and will create a solution file version 9. Only if you choose the C# 3.0 compiler, it will make the project an MSBuild 3.5 project and mark the solution as version 10.

However, this introduces several problems:
C# 2.0 does not support embedding manifest files in assemblies, but C# 3.5 embeds a default manifest file automatically. Simply changing the target framework changes the manifest behavior.

MSBuild 2.0 does not check whether the assemblies you referenced are available in the target framework you have chosen. SharpDevelop can fix this by doing the check on its own.

The target framework is an option that normally can be dependent on the chosen configuration/platform. However the ToolsVersion attribute in MSBuild cannot be dependent on configuration/platform. If one configuration uses C# 3.0, all configurations will; even if they are explicitly set to C# 2.0.

WPF Applications created in SharpDevelop 2.2 use a different way to support XAML and resource compilation than MSBuild 3.5 uses. Such applications can be edited in SharpDevelop 3.0 and will continue to work even though they are set to C# 2.0 / .NET 2.0 in the options.  Setting these projects to C# 3.0 causes problems that can only be fixed by manually editing the .csproj file.

So what should we do?

1. Keep it as it is: works for usual projects, has the problems mentioned above
2. Do it like VS 2008, drop support for the C# 2.0 compiler and force conversion of the project
3. Disable the “Target framework” combo box until the user runs a conversion tool for that project (which could be a button next to the target framework combo box)

1. “Keep it broken” has the advantage that it’s zero work for us
2. Would be the easy to implement, but means that you cannot use SharpDevelop 3.0 if your co-worker still uses SharpDevelop 2.x or Visual Studio 2005
3. Would be more work than 2, but is the most flexible solution

So do you think the ability to edit VS 2005 projects is worth the effort of implementing 3 instead of 2?

From .NET 1.x times, you might remember .manifest files as the way to enable XP visual styles support in your application.

Application manifests are important in Vista. If your application does not have a manifest, Vista will treat it as legacy application and enable file and registry virtualization. This allows the program to write to its own application directory even if it does not have write access to it - the writes are stored in some hidden folder in the user profile. This redirection can cause a lot of confusions when users try to save files using your program where they don't have write access, and instead of getting an error message the files get saved somewhere else!

Vista also has to determine if programs need administrative rights and should cause a UAC prompt. To do that, it uses a heuristic that looks at the file name and string resources to guess if it should elevate. Using a manifest, you can tell Vista if your program needs elevation and avoid the risk of Vista guessing incorrectly.

Previously, you had to distribute the .manifest file along with the .exe; or you had to embed it in the .exe using a post-build action calling mt.exe.

The C# 3.0 compiler has built-in support for embedding manifests, and actually creates a default manifest that turns virtualization off and tells Vista not to elevate if you don't explicitly tell the compiler not to embed a manifest.

Support for embedding manifests is now available in the SharpDevelop project options:

You can choose to embed the default manifest, not embed any manifest, or to embed a custom manifest.
If you select <Create...>, SharpDevelop will automatically create a file called "app.manifest" in your project. The app.manifest created by SharpDevelop is similar to the default manifest the C# compiler uses; it just adds some comments so you don't have to look up how to control Vista's UAC.

Yesterday I wrote about how I made compiling 30% faster by using multi-threading to let my dual-core CPU work with both cores. A problem was that I had to use serialize build events to send them back to the host process.

MSBuild provides these events to loggers (number of occurrences during
the test shown in parenthesis):


MessageRaised   (21156)

ErrorRaised      (0)

WarningRaised   (18)

BuildStarted      (47)

BuildFinished     (47)

ProjectStarted  (449)

ProjectFinished (449)

TargetStarted (2910)

TargetFinished (2910)

TaskStarted (4308)

TaskFinished (4308)

CustomEventRaised (0)

All these events must be send to the SharpDevelop process, where logger AddIns can process them. Each one is serialized separately, and includes information about the full name of the project file being build and some other highly redundant information. Each event is about 1 KB in serialized form. Half of the events occurred in the worker running in the SharpDevelop process, but the other half occurred in the worker that ran in its own process; so half of the events had to be transferred, so we tranferred a total of 18 MB during the 25-second build. Remember the main cost is the serialization (.NET BinaryFormatter), not the data transfer itself.

Obviously, most of these events are not interesting for SharpDevelop, and the best way to improve performance is to ignore uninteresting events in the worker instead of ignoring them in the host.
So which events does SharpDevelop need? Potentially all, since AddIns can add loggers; but if those AddIns tell SharpDevelop which
events are required before the build starts, we can optimize for the common case that all AddIns are only interested in specific events - usually only in errors and warnings. Events that SharpDevelop currently does not use anywhere are MessageRaised, TargetStarted and TargetFinished - all are called quite frequently.
To our benchmark results from yesterday, I added an entry "Less events" where MessageRaised, TargetStarted and TargetFinished are deactivated:

Core Duo Notebook, 1 GB RAM, Vista Home Premium, Aero disabled:
 Rebuild:
   One worker:  35.2, 34.9, 35.2, 35.0, 34.3 : Average 35.1 seconds
   Two workers: 25.6, 23.9, 24.2, 23.5, 23.8 : Average 24.2 seconds
   Less events: 23.7, 23.5, 23.0, 22.4, 21.2 : Average 22.8 seconds

 Build (no changes):
   One worker:  7.4, 7.1, 7.2, 7.8, 7.1 : Average 7.3 seconds
   Two workers: 6.9, 6.8, 6.7, 6.6, 6.2 : Average 6.6 seconds
   Less events: 5.1, 5.0, 6.2, 5.6, 5.1 : Average 5.4 seconds

More than 1 second saved! (see yesterdays' post for details about the benchmark)

Now why does SharpDevelop need TaskStarted and TaskFinished? Take a look at the code:
void OnTaskStarted(object sender, TaskStartedEventArgs e)
{
  activeTaskName = e.TaskName;
  if (MSBuildEngine.CompileTaskNames.Contains(e.TaskName.ToLowerInvariant())) {
    AppendText("${res:MainWindow.CompilerMessages.CompileVerb} " + Path.GetFileNameWithoutExtension(e.ProjectFile));
  }
}

TaskStarted is just there to let the user know that building a project resulted in a recompilation - when doing a Rebuild, SharpDevelop
shows "Compiling ProjectName" for each project, whereas for a Build (no changes), no such lines are shown in the Output view.
But in fact, all consumers of TaskStarted probably look for specific task names. And all consumers of TaskFinished are also listening to TaskStarted and are also only interested in some specific types of tasks.
If we pass the build worker a list of task names we are interested in, it can send us only those events and we don't spend time serializing the tons of events about the tasks <Copy>, <Message> etc.
Together with the "Less events" changes, this decreases the event count from over 36000 to less than 1000.

I implemented this in revision 2696.

Now let's extend our benchmark:
Core Duo Notebook, 1 GB RAM, Vista Home Premium, Aero disabled, two workers, new event handling:
 Rebuild: 20.3, 20.9, 19.9, 20.2, 20.0 : Average 20.3 seconds
 Build (no changes): 4.3, 4.6, 5.0, 4.4, 4.2 : Average 4.5 seconds

That's better than "Less events" because we're sending even less events now, but logger AddIns can tell SharpDevelop that they need to receive additional events, so AddIns did not lose any functionality (but if you install such an AddIn, of course you would lose the performance benefits).
So we're now down from formerly 35.1 seconds (using only one core) to 20.3 seconds - that's 43% faster.
For "Build (no changes)", the reduction of the event count now makes using two workers efficient; from originally 7.3 seconds to 4.5 seconds is a 39% decrease in build time.

[Warning: long post. Short version: Building in SharpDevelop goes multi-threaded, scroll down for benchmark results]

SharpDevelop 3.0.0.2694 contains a rewritten build system. It has some new features that might be interesting for AddIn developers, but one new feature is directly useful for users, or more specifically, users with multi-core CPUs.

Here a quick history on build systems in SharpDevelop:
SharpDevelop 2.0 (the first version to use MSBuild), compiled solutions by telling MSBuild to build the .sln file.
SharpDevelop 2.1 and 2.2 compile solutions by telling MSBuild to build the projects individually.

When writing the build engine for SharpDevelop 2.1, I tried to make it multi-threaded so that two projects that are not dependent on each other can be built in parallel. However, this did not work: MSBuild makes use Environment.CurrentDirectory; and if two MSBuild instances run in the same process at the same time, they overwrite the CurrentDirectory and cause the other instance to search files in the wrong location.

Now I solved this problem by moving the build into worker processes. However, this means that all MSBuild events are serialized and remoted to the host process, where SharpDevelop AddIns need to have access to them - e.g. the CodeAnalysis AddIn registers itself as MSBuild logger, waits for the TaskStarted, ErrorRaised, WarningRaised and TaskFinished events and fixes certain FxCop error messages (some FxCop errors are not associated with a line number, but with a class/member name). Other SharpDevelop AddIns might decide that certain MSBuild events should be displayed in some form, e.g. in the output window. The only way to give AddIns access to all MSBuild events like they had before is to serialize all events and send them to the SharpDevelop process. When doing a build where there's relatively few work but lots of events (e.g. "Clean solution"), the worker is sending up to 1 MB/s of serialized events to the host.

This had a noticeable impact on performance, so I've kept the first worker in the SharpDevelop process, and only additional workers get their own process. So if you use only one worker, you won't see any worker process and should get the same build performance as SharpDevelop 2.x.

So even though building a project has multiple parts, of which some are waiting for the disk and others are waiting for the CPU; the overhead of using a worker process will cause the build to get slower. (MSBuild has to check if input files changed, start the compiler; the compiler will load input files, actually compile, write the output assembly; and some other stuff happens in the obj\ directory that you probably don't want to hear about)

But if you have a dual core processor, the CPU overhead doesn't hurt: a little overhead for one core is better than not using that core at all (this is assuming the compiler you use is not multithreaded).

But enough talking, let's get some benchmark results:

I used SharpDevelop 3.0.0.2694 to compile the source code for SharpDevelop 2.2.1:
AMD 3500+ (single core), 2 GB RAM, Vista Ultimate (32 bit), Areo enabled:
 Rebuild:
   One worker: 34.3, 34.1, 34.6, 34.7, 34.3 : Average 34.4 seconds
   Two workers: 39.6, 38.7, 38.1, 37.0, 36.6 : Average 38.0 seconds

 Build (no changes, so it's not compiling, but just checking everything for changes):
   One worker:  8.5,  8.0,  8.3,  8.0,  8.0 : Average  8.2 seconds
   Two workers: 13.4, 13.9, 14.0, 12.8, 13.1 : Average 13.4 seconds

The number of MSBuild events (Target started, Task started, Log Message...) is not much different between Rebuild and Build (no changes), so we get a constant overhead of about 5 seconds for the SharpDevelop solution on my desktop computer with a single core processor.

Here finally is the interesting part: Test results from my dual core notebook:

Core Duo T2300, 1 GB RAM, Vista Home Premium (32 bit), Aero disabled:
 Rebuild:
   One worker: 35.2, 34.9, 35.2, 35.0, 34.3 : Average 35.1 seconds
   Two workers: 25.6, 23.9, 24.2, 23.5, 23.8 : Average 24.2 seconds

 Build (no changes):
   One worker: 7.4, 7.1, 7.2, 7.8, 7.1 : Average 7.3 seconds
   Two workers: 6.9, 6.8, 6.7, 6.6, 6.2 : Average 6.6 seconds

For build (no changes), the overhead plays a big role, but using two workers still made the build a little faster.

But for the real job when there's something to compile, using two workers really rocks: My build is now 30% faster.

SharpDevelop 3.0 allows you to set the number of workers under Tools > Options > Project and solution > Number of projects to build in parallel. The default value for single-core processors is 1, for other processors it's 2. I don't know if going above 2 will have benefits for quad-core users as the overhead (both CPU for sending events to the host and memory because of the additional processes) increases with the number of workers.

Please leave a comment here with your test results - I'm looking for feedback from other multi-core users.

And if you're doing your own benchmark, some notes on how I did my benchmark:
- make sure enough RAM is free so that not only the system does not swap, but Windows has enough room to keep the files you are compiling in memory (this is generally a good idea if you want your system to be fast)
- don't count the first test run as the files are not yet in memory there
- reuse the same SharpDevelop instance for all test runs
- if you leave less than 60 seconds between test runs, SharpDevelop will re-use the worker process, otherwise it is shut down and has to be restarted. I did my tests re-using a single worker process. It might be interesting to see if the overhead of restarting the worker process is worth keeping it longer than 60 seconds.