I'm working on UWP after working for a long time in WPF and compilation is abysmally slow.
I understand why release compilation is slow (net native compilation) but that's disabled in debug yet it still takes a lot of time between F5 and the application being displayed on screen, even for a blank application.
Is there any way to speed this up (even at the cost of runtime performance)? It's really hard to work with when you're used to C# giving you extremely fast compile times and testing after every small changes.
For exemple just a right click and rebuild on a very simple (4 pages, each < 200 line of xaml, and pretty much 0 C#) uwp project takes almost exactly 20 seconds in debug (without .net native) with no project references. Meanwhile a much larger WPF application (dozens of windows, thouthand of lines of codes) takes a few seconds and most of that time is copying child projects!
As suggested you can find a minimal example for download here :
https://www.dropbox.com/s/0h89qsz66erba3x/WPFUWPCompileSpeed.zip?dl=0
It's just a solution with a blank wpf & blank uwp app. Compilation time for WPF is just over 1 second, for UWP 12 seconds, in each case solution was cleaned and the single project i was testing was right clicked and rebuilt. This is in debug (without .net Native compilation)
I definitely agree that UWP compilation is significantly slower than WPF. I've always had to break apart my WPF assemblies whenever I reached about 5-6 dozen xaml windows or views, in order to keep the incremental compilation times down at 10-20 seconds. The way things are looking, my UWP assemblies will probably grow to only about 2 or 3 dozen items before they take way too long to compile. Any assembly that takes over 10 seconds to compile is problematic when you are trying to do iterative coding & debugging.
Here are two recommendations, if you haven't tried them. The first thing to do is go to the build tab and always remember to uncheck "compile with .NET native tool chain". That is pretty useless for normal debug/test iterations.
The next thing to do is to monitor your build operations with procmon. That will surface any issues that may be specific to your workstation (like virus scanning, other apps that may be interfering).
Here are the factors I noticed that would slow down UWP compilation compared to WPF:
Lots of compilation passes (see them in the build output window):
> MarkupCompilePass1:
> XamlPreCompile:
> MarkupCompilePass2:
> GenerateTargetFrameworkMonikerAttribute:
> CoreCompile:
> CopyGeneratedXaml:
> CopyFilesToOutputDirectory:
> ComputeProcessXamlFiles:
> CustomOutputGroupForPackaging:
> GetPackagingOutputs:
> _GenerateProjectPriConfigurationFiles:
> _GenerateProjectPriFileCore:
Core/Nuget dependencies
Unlike with the .Net Framework, all your dependencies for UWP are coming from .Net core and nuget. You should be able to use procmon and see tons of reads against the directory where VS keeps the stuff: C:\Program Files (x86)\Microsoft SDKs\UWPNuGetPackages. Note that the dependencies themselves aren't of a totally different quality than the .Net framework dependencies (in WPF). But there is a big difference in how those dependencies are gathered and how they are pushed around in your output directories during compile operations (eventually ending up in the final Appx directory).
No "shared output directory" for optimization.
With WPF we could send class library targets individually to a shared output directory (by checking them for build, while unchecking other libraries and the application exe itself). Then we could launch the debugger and without compiling a whole ton of stuff that wasn't necessarily changing. However, UWP requires us to build the entry-level application, regardless of whether we try to configure a shared output directory.
The new "Deploy" step is required in UWP
WPF didn't have the new "deploy" step that is required every time you build a UWP app. You will see the checkbox in the build configuration, and it applies to the entry-level application. If you don't deploy, you won't see all your changes while debugging.
UWP is still actively changing (unlike WPF)
They keep changing the UWP compilation operation. Soon we are going to start to see something called the "WinUI" library that will introduce another dependency from NuGet for UWP apps. This is probably not going to help matters where compilation performance is concerned.
I think once the pace of change starts to slow down, Microsoft may decide to focus on finding ways to improve the performance of compiles. It seems pretty clear when reading the procmon output that less than half of the work done by devenv.exe is especially being done for me. The rest of it is pulling in all the dependencies so that my code can compile against it. There has to be a way to optimize that repetitive processing.
Ronan the (twelve-second?) compile time for an empty project does seem a bit high... Mine generally runs for under five seconds (when empty).
You might want to watch what is going on with CPU. The compilation of a single project should be entirely cpu-bound, especially on the second or third attempt, after the file system has cached your source code (and all the nuget dependencies are downloaded). You may want to keep an eye on task-manager and make sure that you are running at full speed on a single core (ie 25% if you have four cores, 12% for eight cores, etc). If you see CPU drop down too far then something is going wrong. Also make sure to check that CPU is only being used by the usual things that you would expect, eg devenv.exe, VBCSCompiler.exe and MSBuild.exe.
For those who claim to be compiling UWP projects so much faster than everyone else, it might be interesting to hear their benchmarks for the windows community toolkit. (https://github.com/windows-toolkit/WindowsCommunityToolkit/tree/rel/5.0.0) The most interesting compile times would be for projects that are most heavy on xaml. Here are my times:
10 seconds for Microsoft.Toolkit.Uwp.UI.Controls
8 seconds Microsoft.Toolkit.Uwp.UI.Controls.Graph
8 seconds Microsoft.Toolkit.Uwp.UI.Controls.DataGrid
The basic Van Arsdel inventory app is another thing to benchmark. https://github.com/Microsoft/InventorySample Here is my result:
14 seconds for Inventory.App
The VanArsdel sample app would be another good one to benchmark. (Careful with this; you may have to be on a Windows insider build and it may mess up your VS like it did mine). The url is here: https://github.com/microsoft/vanarsdel Here is my result:
13 seconds for VanArsdel project
Remember that projects which are heavily xaml-oriented are also usually the slowest because they involve more processing work during the various compilation passes.
Using Diagnostic Output
While the total elapsed time to build a project is critical, it may also be helpful to understand where the time is coming from. You can get a summary of that if you go to Tools->Options->Build-and-Run and then turn up the project build output verbosity to Diagnostic. Then compile the project in question. At the end you will see a Task Performance Summary that looks like the following. Note that CompileXaml should take the most time. These five items seem to be pretty common.
Task Performance Summary:
100 ms ValidateAppxManifest 1 calls
400 ms ExpandPriContent 1 calls
400 ms Csc 2 calls
600 ms ResolveAssemblyReference 1 calls
7000 ms CompileXaml 2 calls
If you see anything else (eg GenerateAppxManifest) taking up a few seconds every time you compile then that is probably a corruption problem within your project. You should be able to troubleshoot by searching for the word "completely", as in "Building target _GenerateCurrentProjectAppxManifest completely". When you find that, it should tell you why it is doing the extra work.
Min Version Targeting
I noticed that changing the min version on the targeting tab will cut out three seconds out of my "CompileXaml" time.
See below that changing min version to 15063 helps cut down on compile time. I suspect this is related to bloat in the related dependencies.
15063 (creators update) [4 seconds]
16299 (fall creators) [7 seconds]
17134 (version 1803) [ 7.5 seconds]
Its not always an option to target the creators update, so this may not be general-purpose fix. But it is interesting to know about the impact of getting an updated Windows 10 SDK.
Related
I need help with re-hosted designer of WF4. It should be used for design
very complex workflow but there is limit of the nested activities. (It's
around 40th nested activities in one branch). If is that limit exceeded, an
System.StackOverflowException occurred in PresentationCore.dll.
Is there any way, how to increase limit for System.StackOverflowException?
Here is the code example for download. After building and executing application, move the
scrollbars to last activity with number 40 and exception should occur.
I can run it without error and scroll to the very bottom - though it does get a bit slow - (64 bit machine, 16 gb memory).
You can allocate stack size on creation of a new thread, but I don't know how you could change the size for the default UI thread in your application...and if you can I'm not sure it'd be a good idea.
Besides, increasing the limit is only hiding the overall issue - which is why you need this in the first place - is anyone really going to nest 40 layers of complexity in a workflow? It would be completely unwieldy and incredibly difficult to support. Couldn't the logic be split into sub workflows, etc?
I solved the problem by increasing the size of the stack by using utility EDITBIN from VisualStudio
editbin /STACK:6291456 "WpfApplication1.exe"
Unfortunately it doesn't work in VisualStudio by adding into Post-Build section in project properties.
So I'd created the bat file, which is necessary to execute after build.
It is well known that
If compiling takes even 15 seconds, programmers will get bored while the compiler runs and switch over to reading The Onion, which will suck them in and kill hours of productivity.
Our MonoTouch app takes 40 seconds to compile on Macbook Air in Debug/Simulator configuration.
We have about 10 assemblies in the solution.
We're also linking against some native libraries with gcc_flags.
I'm sure there are ways to optimize compilation time that I'm not aware of, which might have to do with references, linker, whatever.
I'm asking this question in hope that someone with better knowledge than me will compile (no pun intended) a list of tips and things to check to reduce MonoTouch compilation time for debug builds.
Please don't suggest hardware optimizations or optimizations not directly related to MonoTouch.
Build Time Improvements in Xamarin.iOS 6.4
Xamarin.iOS 6.4 has significant build time improvements, and there is now an option to only send updated bits of code to the device. See for yourself:
(source: xamarin.com)
Read more and learn how to enable incremental build in Rolf's post.
Evolve 2013 Video
An updated and expanded version of this content can be seen in the video of the Advanced iOS Build mechanics talk I gave at Evolve 2013.
Original Answer
There are several factors affecting build speed. However most of them have more impact on device builds, including the use of the managed linker that you mentioned.
Managed Linker
For devices then Link all is the fastest, followed by Link SDK and (at the very end) Don't link. The reason is that the linker can eliminate code faster than the AOT compiler can build it (net gain). Also the smaller .app will upload faster to your devices.
For simulator Don't link is always faster because there's no AOT (the JIT is used). You should not use other linking options unless you want to test them (it's still faster than doing a device build).
Device tricks
Building a single architecture (e.g. ARMv7) is faster than a FAT binary (e.g. ARMv7 + ARMV7s). Smaller applications also means less time to upload to device;
The default AOT compiler (mono) is a lot faster than using LLVM compilers. However the later will generate better code and also supports ARMv7s, Thumb2;
If you have large assets bundled in your .app then it will take time to deploy/upload them (every time since they must be signed) with your app. I wrote a blog post on how you can workaround this - it can save a lot of time if you have large assets;
Object file caching was implemented in MonoTouch 5.4. Some builds will be a lot faster, but others won't be (when the cache must be purged) faster (but never slower ;-). More information why this often happens here).
Debug builds takes longer because of symbols, running dsymutil and, since it ends up being larger, extra time to upload to devices.
Release builds will, by default (you can turn it off), do a IL strip of the assemblies. That takes only a bit of time - likely gained back when deploying (smaller .app) to the device.
Simulator tricks
Like said earlier try to avoid linking since it will take more time and will require copying assemblies (instead of symlinking them);
Using native libraries is slower because we cannot reuse the shared simlauncher main executable in such cases and need to ask gcc to compile one for the application (and that's slow).
Finally whenever in doubt time it! and by that I mean you can add --time --time to your project extra mtouch arguments to see a timestamp after each operation :-)
This is not really meant as an answer, rather a temporary placeholder until there is a better one.
I found this quote by Seb:
Look at your project's build options and make sure the "Linker
behavior" is at the default "Link SDK assemblies".
If it's showing "Don't link" then you'll experience very long build
time (a large part of it in dsymutil).
I don't know if it is still relevant though, because MonoDevelop shows a warning sign when I choose this option, and it doesn't seem to affect performance much.
You cannot expect your compiler to be lightninng quick without understanding everything that it is required to do. Larger applications will naturally take longer. Different languages or different compilers of the same language can make a huge difference on how long it takes to compile your code.
We have a project that will take almost 2 minutes to compile. Your best solution is to figure out a way to reduce the number of times you compile your code.
Instead of trying to fix 1 line of code and rebuilding, over and over again. Get a group of people together to discuss the problem. Or create a list of 3 or 4 things you want to work on, complete them all then test.
These are just some suggestions and they will not work in all cases.
I noticed that sometimes a .net 4.0 c# application takes a long time to start, without any apparent reason. Can can I determine what's actually happening, what modules are loaded? I'm using a number of external assemblies. Can putting them into the GAC improve performances?
Is .NET 4 slower than .NET 2?
.NET programs have two distinct start-up behaviors. They are called cold-start and warm-start. The cold-start is the slow one, you'll get it when no .NET program was started before. Or when the program you start is large and was never run before. The operating system has to find the assembly files on disk, they won't be available in the file system cache (RAM). That takes a while, hard disks are slow and there are a lot of files to find. A small do-nothing Winforms app has to load 51 DLLs to get started. A do-nothing WPF app weighs in at 77 DLLs.
You get a warm start when the assembly files were loaded before, not too long ago. The assembly file data now comes from RAM instead of the slow disk, that's zippedy-doodah. The only startup overhead is now the jitter.
There's little you can do about cold starts, the assemblies have to come of the disk one way or another. A fast disk makes a Big difference, SSDs are especially effective. Using ngen.exe to pre-jit an assembly actually makes the problem worse, it creates another file that needs to be found and loaded. Which is the reason that Microsoft recommends not prejitting small assemblies. Seeing this problem with .NET 4 programs is also highly indicated, you don't have a lot of programs that bind to the version 4 CLR and framework assemblies. Not yet anyway, this solves itself over time.
There's another way this problem automatically disappears. The Windows SuperFetch feature will start to notice that you often load the CLR and the jitted Framework assemblies and will start to pre-load them into RAM automatically. The same kind of trick that the Microsoft Office and Adobe Reader 'optimizers' use. They are also programs that have a lot of DLL dependencies. Unmanaged ones, the problem isn't specific to .NET. These optimizers are crude, they preload the DLLs when you login. Which is the 'I'm really important, screw everything else' approach to working around the problem, make sure you disable them so they don't crowd out the RAM space that SuperFetch could use.
The startup time is most likely due to the runtime JIT compiling assembly IL into machine code for execution. It can also be affected by the debugger - as another answerer has suggested.
Excluding that - I'll talk about an application ran 'in the wild' on a user's machine, with no debugger etc.
The JIT compiler in .Net 4 is, I think it's fair to say, better than in .Net 2 - so no; it's not slower.
You can improve this startup time significantly by running ngen on your application's assemblies - this pre-compiles the EXEs and DLLs into native images. However you lose some flexibility by doing this and, in general, there is not much point.
You should see the startup time of some MFC apps written in C++ - all native code, and yet depending on how they are linked they can take just as long.
It does, of course, also depend on what an application is actually doing at startup!
I dont think putting your assemblies in GAC will boot the performance.
If possible do logging for each instruction you have written on Loading or Intialize events which may help you to identify which statement is actually taking time and with this you can identify the library which is taking time in loading.
It seem that when I re run my .net application , it became much faster than before , why ?
Also is there anyway for my software to be run faster on startup ?
regards
If it's the first .NET application running in your system, then the first time you run it, all the .NET libraries and the CLR have to be loaded from physical disk. The second time you run, everything will be in the file system cache, so it'll be loading it from memory. There may well be other caching effects in play beyond the file system cache, but that's the most obvious one.
The same is true of your specific application, although that's likely to be a lot smaller than the framework itself.
One option to try to bootstrap this is to have a small no-op application (e.g. a WinForms app that never actually launches a window) which runs on startup. Of course, this will slow down the rest of your startup a bit - and if the computer doesn't run any .NET applications for a long time, the framework will be ejected from the cache eventually.
The first time you run your .NET app the following happens:
1) Loading of your application, the runtime, and the framework from hard disk (which is slow) to the memory (which is much faster)
2) Then your application and the associated libraries are just-in-time JIT compiled to native code...as needed. This native code stays around in the memory, and the runtime infrastructure keeps a record of the code that it has compiled to native code.
3) Only in the third step does this native code actually executed by the processor.
If you dont shut down your computer and rerun your application. The following happens:
1) When the run time encounters your managed code that has already been compiled to native by the JIT compiler , it does not recompile it. It simply executes the already compile native in memory.
2) Only the code that was not JIT compiled to native in the first run is now compile from managed to native...and thats only if needed.
So on a second run of your application two things get real fast:
1) loading either doesnt happen at all or its far smaller than the first one.
2) compilation from managed to native either doesnt happen or its minimal
Thats why your second run of the application is almost always faster then the first run.
This is almost certainly because the OS has loaded needed DLLs which stay in memory (unless the memory is needed elsewhere) after your application exits.
You can run your program in a special mode that just loads and exits) so that those DLLs will load up and this is a trick used by a few applications (MS Office and OpenOffice.org are two that spring to mind immediately).
Some people will run their programs at startup to make their first invocation seem faster but it's my opinion that this should be left to the user. It is their machine after all. By all means show them how they can do it (e.g., add yourprogram.exe /loadandexit to your startup folder) but leave it up to them.
I, for one, don't want every application I run slowing down my boot time.
My project is developed in Visual Studio 08, in C#. It's a standalone desktop application, about 60k lines of code.
Once upon a time I loved working on this software - now that the compliation time has grown to approx 2 minutes, it becomes a far less enjoyable experience...
I think that my lack of experience in C# may be a factor; I have developed everything under one namespace for example - would having a well structured codebase enable the compiler to recompile only the necessary parts of the code when changes are made? Or do I need to separate sections into separate projects/DLLs to force this to happen?
How much of a difference would upgrading to the latest quad-core processor make?
The other thought is, perhaps this is a typical thing for programmers to deal with - is a long compile time like this simply something that must be managed?
Thanks in advance.
Things that increase compile time:
The number of projects in a solution makes more difference than the number of files in a particular project.
Custom build tasks can make a huge difference, especially if they are generating code or running post-build analysis (FxCop, StyleCop, Code Contracts).
Native code projects take longer to build.
A single project containing 60K lines of C# code with no special build features enabled should compile in seconds on any machine made in the past 5+ years.
I'm surprised that 60k lines of code take 2 minutes to compile. I have an application that is 500,000 lines of code, and it only takes about a minute and a half. Make sure you are not doing full rebuilds each time, and make sure you are not cleaning the solution between builds. A normal build should perform an incremental build, only recompiling code that has changed since the last build (along with anything affected by that change.)
Perhaps some other factors might include heavy use of large resources (images?), broad-sweeping changes in the lowest level libraries (i.e. those used by everything else), etc. Generally speaking, on a relatively modern machine, compiling 60,000 lines of C# code should take less than a minute on average, unless you are rebuilding the entire solution.
There is this thread about hardware to improve compile time. Also this really excellent blog post from Scott Guthrie on looking at hard drive speed for performance.
Splitting your project up into multiple projects will help. Only those projects that have changes (and projects that depend on it) will need recompilation.
A single namespace however, shouldn't affect compile time. However, if you do split up your project into multiple projects/assemblies, then a single namespace is definitely not a good idea.
Upgrading to a faster CPU will probably help, but you might find that faster I/O (better disks, RAID, etc will be more useful).
And yes, avoiding long compile times are one of the things developers need to take care of. When it comes to productivity, do whatever you can (better tools, bigger screens, faster machines, etc...)