Grouping multiple 'in_parallel' tasks using 'do' steps

[marcransome]

I'm currently investigating running groups of serial steps in parallel, for example:

- in_parallel:
    - do:
        - task: generate-output-1
          ..
        - task: use-output-1
          ..
    - do:
        - task: generate-output-2
          ..
        - task: use-output-2
          ..

Both generate-output-* tasks generate an output of the same name. Similarly, both use-output-* tasks expect an input of the same name. My concern is whether there is any likely crossover between the outputs from the generate-output-* tasks in terms of what the subsequent use-output-* tasks see?

Syntactically the pipeline is valid and it runs as expected—exposing the correct outputs to the individual use-output-* tasks—but my question is wether it is generally considered safe to use do in this scenario and are there any caveats to this approach?

[vito]

@marcransome This is indeed a gotcha - outputs are effectively in a global namespace through a build, so it's very likely for this to result in a race condition (last write wins). I would recommend using input_mapping and output_mapping to give them unique names.

It would be nice for these to be lexically scoped or something. 🤔