C++ analysis for variants built in Docker

[tiolan (Timo)]

I have a follow up question on

C++/CMake Analysis with Code Variants

Hi all, I am building an open-source project with Travis and followed the guides to also scan it with SonarCloud. The project is written in C++ and CMake. There are basically two variants of the software, that are chose by a CMake define (cmake -D…). Depending on this define, different Source Files are build. What I managed to do is to run two jobs in Travis consecutively (one for each variant) and wrap the build with the sonar scanner. In the Travis logs both variants complete and it seem…

but it was already closed.

I want to analyze C++ source-code that has multiple variants. Meaning parts of the code is flagged with #ifdef.
With the provided solution in above mentioned question, I am able to create a build-script, build all the variants inside the script, wrapped with the build-wrapper and then start the analysis.

Now I want to build each of those variants inside a Docker container, using docker build followed by docker run.
When invoking docker commands inside the build-script, the build inside Docker is ignored - which is understandable.

Is there any solution for this?
Something like combining the output of two build-wrappe’d builds and then scan them together.

Thanks!

[mpaladin (Massimo Paladin)]

Hi @tiolan,

@tiolan

Is there any solution for this?
Something like combining the output of two build-wrappe’d builds and then scan them together.

unfortunately there is not, analysis and build should happen at the same OS level, if two builds requires 2 different environments inside 2 different containers it is not going to work.

[sebastian-freitag]

This is a pity. We are facing the same challenge. We build the software in parallel in 16 different containers. 4 different customers, 2 different operating systems, 2 different CPU architectures (x64 and arm).
(4x2x2 = 16).

It would be really helpful, if I could pass a “build id” as a parameter to sonar-scanner and then the reports would be merged server-side. Or a local export-and-merge and then a second run of the scanner that does no scan but sends the merged data to the cloud.

Is such feature planned?

[mpaladin (Massimo Paladin)]

Hi @sebastian-freitag

thank you for your feedback, we are aware of the use-case.

For now, we have nothing planned regarding the multi-configuration analysis.