Hi, thanks for sharing this issue.

This is very odd. Are you saying that the build error messages themselves are in Japanese?

Something worth checking is the encoding of the source file inside the NCrunch workspace. When NCrunch builds your project, it makes copies of all the source files and places them inside an isolated workspace, where the project can then be built. There is some complexity in identifying the correct encoding of the files, as when they are open in the IDE, it is difficult for NCrunch to know how they are actually encoded (since the text just gets pulled out of the IDE's memory, rather than read from file).

If you right-click on the failed build, then choose Advanced->Browse to workspace, you should be able to inspect the workspace NCrunch has built for the project. It's worth finding the source file giving the error inside this workspace and examining it in an editor that can detect its encoding. Compare the encoding with the original file back in your foreground solution. Are they the same?

Is there any chance you can whittle this down to a small code sample (a project that builds) and submit it through the contact form on this site? If I can get a clean sample to reproduce the problem, I stand a good chance of being able to implement a fix.

This makes sense. When the file isn't open in VS, NCrunch can just straight copy the file from disk in whatever encoding it happens to be.

When the file is open in VS, NCrunch needs to obtain the file content from VS itself rather than the file. In this situation, there is no information available about the encoding of the text in the file. To work around this, NCrunch tries to auto-detect the encoding by reading the file on disk, then writing the data from VS into the workspace using the detected encoding. There is no way this can ever be 100%, but it catches almost every case. You've managed to find a situation where the encoding for some reason isn't being detected correctly or the text provided by VS isn't encoded the way it should be. In either case, I'll need to get a sample that can reproduce the problem so that I can recreate it and assess our options.

One way you may be able to limit the impact of the problem is by moving the Japanese text into a separate source file, away from the rest of your code. In this way, you can leave the file closed while you work so NCrunch doesn't need to try and obtain its data from VS.

Thanks for sharing the sample code. I've managed to reproduce the problem. It didn't fail for me in the same way as you've described, but it still failed badly enough for me to figure out what was happening :)

NCrunch is failing to correctly detect the encoding of BOM-less UTF8 source files that are open in the IDE. It's working fine for resource files, just not source files.

To work around the problem, make sure that any UTF8 source files containing non-ASCII text always have a BOM header. There are a number of tools that let you set the encoding of a source file (I use Notepad++).

A fix for this will be out in the next version of NCrunch.