Box unboxed-rep values reaching value targets at more use-sites

[mzagozen]

This closes #2942. The initial fix was only limited to the crash I found in one of the applications:

 fixassign env t e
   | Just t' <- unboxedRepType t    = forceUnbox env t' e
+  | Box{} <- e                     = e
+  | Just rt <- exprUnboxedRep env e = Box rt e
   | otherwise                      = e

But then my robot continued the search and came up with more edge cases

Unboxable primitives (u64/int/float) are now carried unboxed; each use-site that needs a boxed value must insert the Box. That insertion existed for fixarg/fixreturn/boxValueExpr but was missing elsewhere, so a raw int reached a boxed slot and the first virtual access dereferenced it, causing a runtime crash.

• fixassign: box a raw unboxed-rep RHS into a value/?value target. Missing box when assigning fixed-size int to value #2942
• Cond: box then/else branches when the join type is not unboxable.
• IsInstance: box the operand ($ISINSTANCE derefs $class).
• BinOp And/Or: box both operands ($AND/$OR call bool on them).
• exprUnboxedRep: treat a processed And/Or as already boxed, else a fallback re-boxes it (toB_u64(B_u64), a compile error).

[mzagozen]

@sydow can you please take a look? I also included regression programs that crash on main.

[sydow]

Good morning Marko, did you get a review? I thought I submitted one but now I see no signs of it on the PR page at GitHub. I hate git… Anyhow, the short form is: go ahead and merge! Björn

…

On 18 Jun 2026, at 07:24, Marko Zagožen ***@***.***> wrote: mzagozen left a comment (actonlang/acton#2943) @sydow can you please take a look? I also included regression programs that crash on main. — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS and Android. Download it today! You are receiving this because you were mentioned.Message ID: ***@***.***>