Summary

Impove FlagDependencyExportsPlugin

Parallel collecting exports of dependencies by dep.get_exports.

Since there is a backtracking mechanism in the processing of dependency exports, when re - exporting, if there is a relationship A -> B, after A is computed, when computing B, if B's exports change, then A's exports need to be recomputed.

If the computation queue is [B, A]:

• For serial computation, first compute B's exports and immediately update the B's exports information to the module graph. So when A computes its exports, the module graph contains B's exports information.
• For parallel computation, since A and B will compute their exports simultaneously on an empty module graph, and the subsequent update of A and B's exports to the module graph is sequential. As a result, the state of the module graph when A computes its exports will not match the state when it is updated. Also, because B is updated first and then A, the backtracking and recomputation of A cannot be triggered, leading to incorrect computation results.

If the computation queue is [A, B]:

• For serial computation, first compute A's exports and immediately update the A's exports information to the module graph, but this information is useless. Because later B's exports will be computed and updated to the module graph, then triggering the recomputation of A. At this time, the module graph has B's exports information when A is computing.
• For parallel computation, after A and B compute their exports and update them to the module graph, since B is updated after A, it will cause A to be recomputed. At this time, the module graph has B's exports information when A is computing, so the correct result can be obtained.

Therefore, modules need to be sorted according to their dependency relationships to ensure that the backtracking computation will be triggered, and the results of the first parallel computation are discarded. Although this will result in some wasted computations, since the number of re - exported modules is not particularly large, parallel computation will still significantly improve performance.

Improve FlagDependencyUsagePlugin

parallel collecting referenced exports of dependencies by dep.get_referenced_exports

This plugin has the same issue with FlagDependencyExportsPlugin. But it has a queue internal and has handle the dependencies by recursive calling collect_referenced_export_items. So it is safe to just parallel getting their referenced dependencies and no need to sort them.

Improve SideEffectsFlagPlugin

Just prevent CallExpr being cloned by sending the parent expr node to is_pure_call_expr function.

Related links

Checklist

• Tests updated (or not required).
• Documentation updated (or not required).