When compiling a pattern match like
```
[] -> ..
[_] -> ..
[_, ..] -> ..
```
to a decision tree, we must make sure that the last test (len >= 1)
does not touch the branch reached by the second test (len == 1). It is
enough to ban (len >=) tests from ever touching exact-sized list
patterns, because a spread test (len >=) can never reach an exact-sized
test.
On the other hand, an exact-sized test can reach a spread pattern,
because in
```
[_, _] -> ..
[..] -> ..
```
the last branch generates tests for patterns `[]` and `[_]`, and we would
like those patterns to be covered by the spread test (len >= 0)!
Closes#4685
In #3352 an optimization to transform `ra = .field` into
```
ra = \#rcd -[ra]-> #rcd.field
```
rather than
```
__ra1 = \#rcd -[__ra1] -> #rcd.field
ra = LambdaSet { __ra1 }
```
was introduced. However, this optimization is not correct when `ra =
.field` is defined as a toplevel thunk, for in such situations we
indeed want the thunk `ra` to return the lambda set it resolves to,
rather than repointing at itself.
Besides reverting this change, another option would be to convert
accessors into closures before translation of Can to IR. However, this
complicates the translation algorithm more than it already is, and I'd
like to avoid additional special-cases.
Closes#4606
Previously, a program like
```
main =
f =
n = 1
\{} -[#lam]-> n # suppose lambda set = #lam
f {}
```
would be transformed to
```
main =
n = 1
f = \{} -[#lam]-> n
f {}
```
However, the IR lowering procedure is such that we would then associate
`f` as definining the procedure given the lambda set `#lam`. This is not
correct, as `f` is really a function pointer in this circumstance,
rather than the definer of `#lam`.
Instead, the transformation we want to perform is
```
main =
n = 1
#lam = \{} -[#lam]-> n
f = #lam
f {}
```
Which is what this patch does
Closes#2403
Makes sure that we turn on `debug-symbols` in mono tests, and that this
feature is fully respected in symbol generation, so that output in
release + debug builds are the same.
Closes#4435
Currently things like `1 / 200` lead to a miscompilation because we type
`200` (and as a result, both `1` and the division result) as a ranged
number with width >= U8. During mono that forces the number to become an
`I64` because our logic was that a ranged number can only become a float
if it's at least as wide as an I8. But this is incorrect; as long as the
type is wrapped in `Frac` constructor and it's a ranged number (and not
a ranged int), it should become a fractional type.
```
» 1 / 200
0.005 : Float *
```
Closes#4047
Even if there are no changes to alias arguments, and no new variables were
introduced, we may still need to unify the "actual types" of the alias or opaque!
The unification is not necessary from a types perspective (and in fact, we may want
to disable it for `roc check` later on), but it is necessary for the monomorphizer,
which expects identical types to be reflected in the same variable.
As a concrete example, consider the unification of two opaques
P := [Zero, Succ P]
(@P (Succ n)) ~ (@P (Succ o))
`P` has no arguments, and unification of the surface of `P` introduces nothing new.
But if we do not unify the types of `n` and `o`, which are recursion variables, they
will remain disjoint! Currently, the implication of this is that they will be seen
to have separate recursive memory layouts in the monomorphizer - which is no good
for our compilation model.
Closes#3653
