Scala 2.13.12 fatal warning with -Xsource:3 and private constructor for case class

[vladimirkl]

Reproduction steps

Scala version: 2.13.12

scala> case class Data private (s: String)
                  ^
       error: constructor modifiers are assumed by synthetic `copy` method
       Scala 3 migration messages are errors under -Xsource:3. Use -Wconf / @nowarn to filter them or add -Xmigration to demote them to warnings.
       Applicable -Wconf / @nowarn filters for this fatal warning: msg=<part of the message>, cat=scala3-migration, site=Data

Problem

Same warning is for synthetic apply method. This cannot be fixed without redefining copy and apply explicitly and doesn't help with any Scala 3 migration issue. In all cases I have to suppress it with @nowarn

[guizmaii]

FYI, also reported here: https://discord.com/channels/632150470000902164/632150470000902166/1150748646296518707

[lrytz]

I have to suppress it with @nowarn

You can also use a compiler flag -Wconf:msg=constructor modifiers are assumed:s

[vladimirkl]

I have to suppress it with @nowarn

You can also use a compiler flag -Wconf:msg=constructor modifiers are assumed:s

Sure, this is better than polluting code with @nowarn. But anyway warning looks like a compiler bug.

[lrytz]

But anyway warning looks like a compiler bug.

I don't think it's a bug; the main use case of -Xsource:3 is to help migration to Scala 3, the warning exists because the code is handled differently between Scala 2 and 3.

But I understand that many projects keep -Xsource:3 enabled for a long time, for example libraries that are cross-published; -Xsoruce:3 helps here for example because of the syntax backports. For the warning in question there's nothing one can do, so it needs to be silenced. Maybe that's too inconvenient?

[vladimirkl]

Looks like warning is confusing because it is not clear how this code is handled differently between Scala 2 and 3. Is it possible to add some documentation?

[lrytz]

You're right, "constructor modifiers are assumed by synthetic copy method" is really hard to understand. The difference is this:

scala> case class A private (x: Int); object A { def f = A(1) }
class A
object A

scala> A.f.copy(x = 2)
val res0: A = A(2) // allowed in Scala 2, disallowed in Scala 3 because `copy` is private

I also notice that Scala 2 changes semantics for this example with -Xsource:3 (it makes the copy method private), I'm not sure it should do that; I thought the idea in general was to warn about semantic changes in Scala 3, but not replicate them.

Documenting -Xsource:3 is indeed something we need to do. scala/scala-dev#470 (comment).

Maybe we can also find ways have better defaults. For example focus -Xsource:3on migration (as it is now) and provide -Xsource:3cross for cross-building.

[guizmaii]

I also notice that Scala 2 changes semantics for this example with -Xsource:3 (it makes the copy method private), I'm not sure it should do that; I thought the idea in general was to warn about semantic changes in Scala 3, but not replicate them.

That's one of the main reasons I enable -Xsource:3 in all my projects. I want this behaviour in my Scala2 code.
Not sure we need to emit a warning if the Scala3 behaviour is already applied in Scala2 code using -Xsource:3.
The code will not compile anyway.

[som-snytt]

This is also on my radar since seeing it in std lib scala/scala#10551

-Xsource:3 was always intended to minimize migration, and not just warn: for example, the change in implicit lookup (-Yscala3-implicit-resolution) was retracted only because it was buggy, but it was a behavior change that let you edit code now and minimize change later. It's true that -Xsource:3 is mostly syntax and warnings.

The synthetic apply warning is significant for libraries that must plan around compatibility. I agree there is no great solution, which attests to the severity of the breaking change between the compilers.

I'll try to propose something; maybe another knob.

I noticed the warning about apply is incorrect when user has supplied one, because of an implementation detail. That is a separate bug.

Edit: the change in inference of type of overriding method is an obvious example of behavior change that I think is a win, but also requires a code change if you prefer the old inferred type; and is annoying because it requires -Wconf to silence the warning as an acknowledgment, "Yes, I saw the warning and I like the change." The alternative would be -quickfix to apply the source change.

Oh, maybe -quickfix should be able to emit an edit that either retains old behavior or encodes the new.

[som-snytt]

I will try Lukas's idea of the regular option to warn and the x-tended option to apply new behaviors. Maybe it will make it into 2.13.13 on Friday the 13th. I'll try to document the situation, but I will likely underdeliver, as I am no sjrd. I'll sample the canine cuisine on the stdlib PR.

Edit: current help for -Xsource, because it took over for -future:

"Enable features that will be available in a future version of Scala, for purposes of early migration and alpha testing."

Obviously, the accommodation is to only warn under -Xsource:2.14. I misspoke, I ought to have said -Xsource:2.13, which seems to be a legal setting.

[som-snytt]

Sample improved message

Deadline.scala:32:12: access modifiers for `apply` method are copied from the case class constructor