areaDetector and motors encode build-rules and a dependency chain that is useful for building multiple modules under the same command, but they bring a disadvantage by forcing specific versions of the submodules they contain. We can (and do) override these versions using git checkouts, which IMO is inefficient and goes against the principle of using tarballs and eventually checksumming them ( #80 ).
For one version of this, we can download a tarball of the supermodules for their build rules, and then individual tarballs for each submodule as we decide to include them. No more using git as part of the build process. Is this a good idea? Would we like for base packages (e.g. ADCore, ADSupport) to be tied to the areaDetector version?
Alternatively, we can keep git usage, perform shallow clones for modules we don't wish to override, and only override those modules we need to, and here we can evaluate switching to tarball downloads instead of deep clones.
I'm in favor of the first proposal without special treatment even for base packages; especially because I don't think we can ignore the possibility of needing to patch one of the base packages ourselves during some intermediate situation, which would be made more complicated by treating them differently.
areaDetector and motors encode build-rules and a dependency chain that is useful for building multiple modules under the same command, but they bring a disadvantage by forcing specific versions of the submodules they contain. We can (and do) override these versions using git checkouts, which IMO is inefficient and goes against the principle of using tarballs and eventually checksumming them ( #80 ).
For one version of this, we can download a tarball of the supermodules for their build rules, and then individual tarballs for each submodule as we decide to include them. No more using git as part of the build process. Is this a good idea? Would we like for base packages (e.g. ADCore, ADSupport) to be tied to the areaDetector version?
Alternatively, we can keep git usage, perform shallow clones for modules we don't wish to override, and only override those modules we need to, and here we can evaluate switching to tarball downloads instead of deep clones.
I'm in favor of the first proposal without special treatment even for base packages; especially because I don't think we can ignore the possibility of needing to patch one of the base packages ourselves during some intermediate situation, which would be made more complicated by treating them differently.