Ian McLerran (Apr 22 2024 at 00:08): Ian McLerran (Apr 22 2024 at 00:08):Can any tell me why can't I pass an Int Unsigned16 into a function which requires Int *?
This 2nd argument to addDays has an unexpected type:
109│ addDays { year: date.year + 1, month: 1, dayOfMonth: 1, dayOfYear: 0 } (newDayOfYear - daysInYear)
^^^^^^^^^^^^^^^^^^^^^^^^^
This sub call produces:
Int Unsigned16
But addDays needs its 2nd argument to be:
Int *
Can you share the function addDays please?
Ian McLerran (Apr 22 2024 at 00:09):addDays : Date, Int * -> Date
addDays = \date, days ->
daysInYear = if isLeapYear date.year then 366 else 365
newDayOfYear = date.dayOfYear + Num.toU16 days
if newDayOfYear > daysInYear then
addDays { year: date.year + 1, month: 1, dayOfMonth: 1, dayOfYear: 0 } (newDayOfYear - daysInYear)
else
fromYd date.year newDayOfYear
Date : { year: I64, month: U8, dayOfMonth: U8, dayOfYear: U16 }
Luke Boswell (Apr 22 2024 at 00:10):Does it work if you change it to a instead of *?
Ian McLerran (Apr 22 2024 at 00:11):Nope, no difference.
Luke Boswell (Apr 22 2024 at 00:11):I'm not sure I can help much... it seems strange to me
Brendan Hansknecht (Apr 22 2024 at 00:11):Int _ should work, but so should Int *
Ian McLerran (Apr 22 2024 at 00:12):Ahah! Int _ does work!
Luke Boswell (Apr 22 2024 at 00:12):Maybe it's something to do with Num.toU16... what if you change that to Num.intCast
Brendan Hansknecht (Apr 22 2024 at 00:12):But yeah, 99% sure that Int * should work. Good bug to file.
Ian McLerran (Apr 22 2024 at 00:13):Changing to intCast does not fix, but yes, the Int _ does work.
Ian McLerran (Apr 22 2024 at 00:14):I can definitely file a bug report for that
Luke Boswell (Apr 22 2024 at 00:17):Reproduction
interface Test
exposes []
imports []
expect add 1 == 1
add : Int * -> _
add = \in -> add (Num.toU16 (in - 1))
edit even smaller
Ian McLerran (Apr 22 2024 at 00:19):Awesome. Thanks for finding that minimum reproduction.
Ian McLerran (Apr 22 2024 at 00:20):Surprised I haven't run into that bug sooner... I've been using Int *s left and right
Richard Feldman (Apr 22 2024 at 00:39):this isn't a bug actually!
Richard Feldman (Apr 22 2024 at 00:39):if you accept an Int * argument, that means you can pass any size int into the function and it will work
Richard Feldman (Apr 22 2024 at 00:39):so let's say I pass 1i32 in, for example (because Int * says I'm allowed to do that)
Richard Feldman (Apr 22 2024 at 00:39):now the body of the function is 1i32 + 1u16 which is a type mismatch
Richard Feldman (Apr 22 2024 at 00:42):so I think a more helpful error message would say something like:
This type annotation says the function can accept any
Int, but the implementation only works withU16.If you used any integer other than
U16here, the implementation would have a type mismatch, but
the type annotation incorrectly says other integer types would work.Hint: Try changing
Int *toU16, or changing the implementation to no longer useU16wherever it currently uses it.
Brendan Hansknecht (Apr 22 2024 at 01:07):I don't think this is the same @Richard Feldman
This is U16 + (Num.toU16 (Int *))
Brendan Hansknecht (Apr 22 2024 at 01:07):So should always type check, right?
Brendan Hansknecht (Apr 22 2024 at 01:08):Cause no matter the input int type, you convert it to U16
Brendan Hansknecht (Apr 22 2024 at 01:08):As long as the function can specialize multiple times for different Int *, it should be totally fine.
Richard Feldman (Apr 22 2024 at 02:03):sorry, the original one looks like it should work (so there's a bug there) but I wouldn't expect the minimal repro to work (although it's been edited since my previous response)
Richard Feldman (Apr 22 2024 at 02:04):I totally understand why this looks like it should work:
add : Int * -> _
add = \in -> add (Num.toU16 (in - 1))
because in the mental model of "Int * means it accepts any integer" it should totally work
Richard Feldman (Apr 22 2024 at 02:05):however, that's not quite what Int * means
Richard Feldman (Apr 22 2024 at 02:05):it's very close to what it means, but not quite
Richard Feldman (Apr 22 2024 at 02:05):and I think the fact that it's recursively calling itself with a narrower type than Int * (namely U16) would be an error in a perfect implementation of Roc's type system
Richard Feldman (Apr 22 2024 at 02:09):but @Ayaz Hafiz would know better (and probably be able to explain better) than I would - I might be wrong about that!
Ayaz Hafiz (Apr 22 2024 at 02:33):yeah, the reason
add : Int * -> _
add = \in -> add (Num.toU16 (in - 1))
doesn't work is that it demand polymorphic recursion - regardless of what you pass in for the first add, the recursive call to add must be U16. And Roc does not support polymorphic recursion. That is, a recursive call must be as generic as the initial call, so in this case the recursive call to add must be made with the same type as in.
Luke Boswell (Apr 22 2024 at 02:39):This was my intuition for the issue and why I wrote the repro that way. Is it the same issue as the original report?
Luke Boswell (Apr 22 2024 at 02:40):The original is also passing a U16 as the second argument to addDays
Ayaz Hafiz (Apr 22 2024 at 02:45):yes, I think so
Richard Feldman (Apr 22 2024 at 03:38):ahh I missed the recursion in the original because it's further down than where the conversion to U16 happens - makes sense now!
Ian McLerran (Apr 22 2024 at 13:56):Thanks for taking a look at this, all. Good to know its just PEBCAK! :sweat_smile:
I want my API to be flexible on argument types, so I rewrote with a helper method to get out of the polymorphic recursion issue.
Richard Feldman (Apr 22 2024 at 15:02):this definitely seems like an opportunity for more helpful error messaging though!
Brendan Hansknecht (Apr 22 2024 at 15:05):Why don't we have polymorphic recursion here? Feels like after the first call it will settle into a base case
Brendan Hansknecht (Apr 22 2024 at 15:05):I feel like polymorphic recursion makes sense to support.
Brendan Hansknecht (Apr 22 2024 at 15:06):But maybe it leads to pains due to tail recursion
Richard Feldman (Apr 22 2024 at 15:59):I'm not sure how feasible it is to support without a totally different type system - that's another @Ayaz Hafiz question :big_smile:
Ayaz Hafiz (Apr 22 2024 at 16:13):yeah polymorphic recursion is not a good idea
Ayaz Hafiz (Apr 22 2024 at 16:14):I don't think the use cases are really there and it makes a ton of code generation really hard
Ayaz Hafiz (Apr 22 2024 at 16:15):it also requires a certain expansion to the type system
Brendan Hansknecht (Apr 22 2024 at 17:22):Side question: could the initial function be made to work by adding a Num.intCast to go from U16 back to the original Int * type?
Ayaz Hafiz (Apr 23 2024 at 02:06):I think yes
Ian McLerran (Apr 25 2024 at 17:58):Tested and confirmed, intCast will allow the original recursive form.
Notification Bot (Apr 25 2024 at 17:58):Ian McLerran has marked this topic as resolved.
Luke Boswell (Apr 25 2024 at 20:48):I wonder if we can capture this somehow? Maybe an advanced section in Tutorial on recursion?
Last updated: Jul 06 2025 at 12:14 UTC