Issue with msbuild custom build rule
I have an issue with a custom build rule that is installed as part of the CUDA toolkit. Installation of the CUDA toolkit works corretly. This is done via a powershel install script. Appveyor however reports the build as failing with the following error on the FLAMEGPU2 project.
C:\Program Files
(x86)\MSBuild\Microsoft.Cpp\v4.0\V120\BuildCustomizations\CUDA
8.0.targets(514,9): error : The CUDA Toolkit v8.0 directory '' does
not exist. Please verify the CUDA Toolkit is installed properly or
define the CudaToolkitDir property to resolve this error
This is usually the case when an installation has not worked correctly however I have enabled access tot the remote build worker to debug this. Before building in the remote session the boost environment variable must be set. e.g.
$env:BOOST_DIR = "C:\Libraries\boost_1_59_0"
Build can then take place explicityly listing the configuration and platform (normally specified in appveyor.yml)
msbuild FLAMEGPU2.sln /p:configuration=BoostTest
/p:platform="x64"
This works correctly and has no issue with the custom build rule.
What is AppVeyor doing differently in the build processes to when I login via RD and trigger the build?
Comments are currently closed for this discussion. You can start a new one.
Keyboard shortcuts
Generic
? | Show this help |
---|---|
ESC | Blurs the current field |
Comment Form
r | Focus the comment reply box |
---|---|
^ + ↩ | Submit the comment |
You can use Command ⌘
instead of Control ^
on Mac
1 Posted by Ilya Finkelshte... on Jan 03, 2017 @ 10:29 PM
Hi Paul,
I cloned your repro and was trying to find logical explanation to the issue, but I could not. After googling for
CudaToolkitDir
I found that mostly people fixing it with re-installation of CUDA toolkit. Taking also into account that it worked from interactive RDP, I am speculating that this is something related to CUDA toolkit installation process. I tried to reboot build VM after install finished and it helped. I did that with the following in YAML:Sorry, this is probably not best solution, but at least it should be good enough workaround for now.
--ilya.
Ilya Finkelshteyn closed this discussion on Aug 25, 2018 @ 02:10 AM.