remove static dispatch from PrintWriter in VM

[davidhewitt]

I'm not convinced that &mut PrintWriter is called in a hot enough loop that inlining potential of static dispatch is worth the huge cost in bloat from additional monomorphizations of all VM methods for all P.

[codspeed-hq]

Merging this PR will not alter performance

✅ 13 untouched benchmarks

---

_{Comparing dh/dyn-print-writer (0c215eb) with main (50cd46d)}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[samuelcolvin]

Ok, but does this actually improve build time or binary size relative to Ty?

[davidhewitt]

No (in general it looks like each binary we build is typically using only a single PrintWriter implementation).

But I was going to experiment with passing &mut VM around in a bunch of places where &mut Heap is currently passed around to implement the recursion guards, and I didn't fancy the visual clutter of the additional generic.

[davidhewitt]

Correction: the Python bindings use two different PrintWriter implementations depending on whether print_callback is passed, and the javascript bindings also will after #146

I checked the Python bindings, and the release build goes from 8.8MB to 8.6MB on this branch. If we make more functions that take the VM, that size saving will increase further.

[samuelcolvin]

we definitely want to add a variant of print_callback that collects the output into stdout and stderr strings.

So probably worth merging this, currently has conflicts.