Fix Dockerfile.gpu: COPY sources before install_deps.sh

[haas-ac]

Problem

Dockerfile.gpu currently runs install_deps.sh and ln -s /var/www/node.sh before COPY. Both files live in this repository, so a GPU image build from source fails (install_deps.sh: No such file or directory).

A later RUN npm install --production also fails (npm: not found) because Node/npm come from nvm inside install_deps.sh and are not on PATH in a new layer.

The CPU Dockerfile already does this correctly: COPY first, then install_deps.sh and node --powercycle in the same RUN.

Change

• Copy the tree into /var/www before installing deps
• Keep install_deps.sh, tool symlinks, and node index.js --powercycle in one RUN (install_deps.sh already runs npm install --production)
• Keep the CUDA ldconfig step and the odm user
• Add ARG FROM_REPO=opendronemap/odm:gpu so a custom ODM GPU image can be used with --build-arg FROM_REPO=…

Test

PR CI only builds the CPU Dockerfile, not Dockerfile.gpu.

Built locally with a custom ODM GPU base:

bash
docker build -f Dockerfile.gpu \
  --build-arg FROM_REPO=<odm-gpu-image> \
  -t nodeodm:gpu-test .
docker run --rm --gpus all -p 3000:3000 nodeodm:gpu-test
curl -s http://127.0.0.1:3000/info

Container starts and /info responds.

Notes

This is a small bugfix (NodeODM README: PR directly). I used an AI coding assistant to help draft the reorder; the failure mode and the intended order come from GPU builds against current master.

[maurerle]

This harmonizes the dockerfile.gpu with the other Dockerfile and reduces the unique changes 😊

[maurerle]

This is useful when cross checking other states of the underlying ODM image.
Similar as here: https://github.com/WebODM/NodeODX/blob/a480583ff8a4d7aaa23f890433b78727ebe750d0/Dockerfile.gpu#L1-L2