RFR: 8306543: GHA: MSVC installation is failing when component is already installed

[Christoph Langer]

On Thu, 20 Apr 2023 11:56:37 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

> Currrent GHA fails on MSVC installation. Looks like the required versions are already installed in the runner image, and this is why the command fails. Our configure scripts look for the specific minor version, and the builds still pass if I drop the VC installation block completely. This tells me the appropriate minor version is already installed. 
> 
> For safety, I think we can just ignore the errors from the MSVC installation. In case the runner images moves on to another minor version later, the dependencies would start to install successfully again.
> 
> See the GHA run. Both Windows x86_64 and Windows AArch64 select `1429~1.301`:
> 
> 
> * Toolchain:      microsoft (Microsoft Visual Studio 2019)
> * C Compiler:     Version 19.29.30148 (at /c/progra~2/micros~2/2019/enterp~1/vc/tools/msvc/1429~1.301/bin/hostx64/x64/cl.exe)
> * C++ Compiler:   Version 19.29.30148 (at /c/progra~2/micros~2/2019/enterp~1/vc/tools/msvc/1429~1.301/bin/hostx64/x64/cl.exe)
> 
> 
> The last successful GHA run I had selected the same versions:
> 
> 
> Tools summary:
> * Toolchain:      microsoft (Microsoft Visual Studio 2019)
> * C Compiler:     Version 19.29.30148 (at /c/progra~2/micros~2/2019/enterp~1/vc/tools/msvc/1429~1.301/bin/hostx64/x64/cl.exe)
> * C++ Compiler:   Version 19.29.30148 (at /c/progra~2/micros~2/2019/enterp~1/vc/tools/msvc/1429~1.301/bin/hostx64/x64/cl.exe)
> 
> 
> Additional testing:
>  - [x] Ad-hoc GHA runs pass the configure/initial-build
>  - [x] Full GHA runs pass (this PR)

Hm, I did some more testing... The actual error comes from the path in option --installPath in this [line](https://github.com/openjdk/jdk/blob/8346ae2bc1152f13bc77c643252d84e2043ffe0b/.github/workflows/build-windows.yml#L105). Seems like the Windows installer now needs a path notation with backslash () instead of (/). If you correct that, the installation will work again. So that would be the more correct fix, I guess.

On the other hand, the installation takes some 5 to 15 minutes, for something that's obviously already available in the runner. I think we should tweak that and somehow check whether an installation is necessary beforehand at all.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/13558#issuecomment-1517254394