We normally build binaries using garden in the buildkit docker images. However, the remote machines are not particularly quick, when compared to the CI machines we are able to use.
I have set it up such, that on CI, the binaries are build in the CI workers (which is around 3x faster than in-cluster), and those binaries are then passed into the docker context. In the dockerfile, if the binaries are found and the ARG flag is set, it skips over the expensive steps and just uses the CI-built binaries.
So this in principle works, with one exception - the pre-built binaries are of course taken into account when hashing and thus calculating the version of the module. And so it results in two different versions the same code is built in CI (with the binaries) and outside the CI (without the binaries). This is of course wasted work I'd like to avoid (i.e. re-building the module / recompiling binaries from identical code)
Is there a way to make this case work? If not currently (and I admit we're yet to migrate to 0.13.x), this might be a feature request, I guess