I think there are two aspects to workspace support:
(a) the set of crates being monitored
(b) the set of crates being tested / benched / etc.

What @malbarbo suggested are these two modes:

(a) monitor the full set of crates, but (b) only test the single crate that changed
(a) monitor a fixed set of crates, and (b) test exactly that same set of crates

Strategy (1) would be a good default since workspaces can be big.

On the other hand it is possible that changes have side effects on other crates within the same workspace. So it's definitely interesting to have a way to pass --all and monitor + test the full set of crates. This could be achieved with strategy (2).

.. however

I suggest to adopt this alternative strategy:

cargo watch could always watch the entire workspace. If the user passes a package flag (--package, -p or --all) it would still watch the entire workspace, but only test for side effects in the specified packages.

The difference is this:
Say we call cargo watch -p foo 'test' and then a crate bar is changed within the same workspace. Within this alternative suggestion foo would be recompiled/tested nevertheless, because it can be affected by that change. In @malbarbo s original suggestion the change to bar would be ignored.

Support for workspaces #52

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions