[hack pipes] Inline topic token when possible #14278
Conversation
nicolo-ribaudo
added
Spec: Pipeline Operator
PR: Output optimization 🔬
|
|
||
| return _ref4 = n, (_ref3 = Math.abs(_ref4), (_ref2 = Promise.resolve(_ref3), (_ref = await _ref2, triple(_ref)))); | ||
| return _ref = Math.abs(n), triple(await Promise.resolve(_ref)); |
There was a problem hiding this comment.
This isn't fully inlined because Promise.resolve might be an impure getter. Since Babel assumes unmodified builtins, in Babel 8 we could expand this "implicit assumptions" to also mark this as pure and transform it to
return triple(await Promise.resolve(Math.abs(n)));
| @@ -1,7 +1,7 @@ | |||
| let i = 0; | |||
| let sum = 0; | |||
|
|
|||
| while (i |> (i = ^ + 1) |> ^ <= 10) | |||
| while (i |> (i = 2 * ^ - ^ + 1) |> ^ <= 10) | |||
There was a problem hiding this comment.
This is just to keep the "spirit" of the test, otherwise it was completely inlined.
|
Build successful! You can test your changes in the REPL here: https://babeljs.io/repl/build/51308/ |
| !state.sideEffectsBeforeFirstTopicReference && | ||
| !path.isPure() | ||
| ) { | ||
| state.sideEffectsBeforeFirstTopicReference = true; |
There was a problem hiding this comment.
Is sideEffectsBeforeFirstTopicReference invalidated only when side effects occur before the first topic reference in an RHS? For example, will it be set to true for x |> (f(x), g(x))x |> (f(#), g(#))?
There was a problem hiding this comment.
I assume you mean x |> (f(#), g(#)) and that f/g are defined and not reassigned. It will not be set to true because there are no side effects before the first #. However, since there are multiple #s we won't inline anything.
I only tracked the side effects before the first one, so that x |> (#, SIDE_EFFECT) doesn't prevent inlining.
| } | ||
| } | ||
| }, | ||
| "ClassBody|Function"(_, state) { |
There was a problem hiding this comment.
Class body does not encapsulate all expressions under class, for example: class decorators.
There was a problem hiding this comment.
That's exactly why I used ClassBody: class decorators evaluation is not deferred!
There was a problem hiding this comment.
If so then we have to exclude computed class element keys, too.
There was a problem hiding this comment.
The problem is that exactly determining what runs "now" is hard. The list includes (but is not limited to):
- class keys
- static fields
- iifes
To avoid making the transform too complex, I'd prefer to leave it as-is and deoptimize when they happen.
github-actions
bot
added
the
outdated
I discovered that "removing unnecessary tmp vars" is quite satisfying 😁
When a topic token is only used once and either:
|>is pure|>before evaluating the topic token are purethe pipeline operator can be compiled without temporary variables, by directly inlining the lhs in the rhs.
cc @js-choi if you have time to review, since you initially implemented this plugin!