Spread props cause unnecessary changes

[Rich-Harris]

Describe the bug

Inline expressions are re-evaluated when setting spread props on a component.

To Reproduce

https://svelte.dev/repl/38ab6eb9233a4d8c97e67d576e063109?version=3.6.8

Expected behavior

Waggling the slider should only cause a to show up as having changed. With the spread version, c also appears to have changed.

Severity

Mildly annoying. Could result in incorrect behaviour (as opposed to mere redundant work)

Additional context

This could be fixed like so:

var nested1_changes = (changed.x || changed.a || changed.b) ? get_spread_update(nested1_spread_levels, [
  (changed.x) && ctx.x,
  (changed.a) && { a: ctx.a },
  (changed.b) && { b: ctx.b },
-  { c: [1] },
-  { d: "string" }
+  nested1_spread_levels[3],
+  nested1_spread_levels[4]
]) : {};
nested1.$set(nested1_changes);

[Rich-Harris]

Looking at that code again, I'm wondering if there's a good reason not to do this instead:

if (changed.x || changed.a || changed.b) {
  nested1.$set(get_spread_update(nested1_spread_levels, [
    (changed.x) && ctx.x,
    (changed.a) && { a: ctx.a },
    (changed.b) && { b: ctx.b },
    nested1_spread_levels[3],
    nested1_spread_levels[4]
  ]));
}

[shavyg2]

Hmm kinda makes sense that it changes, aren't you creating a new array ever time the component is called? I'm kinda thinking that if c should be const shouldn't it be declared in the script at the top as a const?

In react pretty sure c would change on each render and it kinda makes sense

[Rich-Harris]

No, this is a philosophical difference between React and Svelte. The React ideal is to recreate the universe from scratch on every state change (in practice, it falls short of that for performance reasons), whereas the Svelte ideal is to only ever update things that are known to have changed (in practice, it falls short of that for correctness/simplicity reasons). There's no reason to recreate b in <Foo a={x} b={[1]}/> — no possible value can come of it — so we shouldn't. This is just about aligning the spread prop behaviour with the non-spread-prop behaviour.

[shavyg2]

I see what your saying at compile time you see there is no reason to change, what if some one spread a value that contains b as a value tho?

[orlov-vo]

@Rich-Harris Is it related with this bug REPL?