Remove unused type parameter for Parallel instances

[Billzabob]

No description provided.

[tpolecat]

Looks like changing def to val and/or removing the unused type parameter breaks binary compatibility.

[Billzabob]

@tpolecat How does cats handle binary compatibility breaking changes?

[kailuowang]

@Billzabob we made a promise to no breaking binary compatibility on core until Cats 3.0.
Did it cause any issues for users?
We could fix it in a BC way - we remove the implicit keyword of the original def, make it package private and deprecated, add your implicit val with a BinCompat suffix.

[Billzabob]

@kailuowang No I don't think it causes any issues so it's probably not worth it to do all that. It was just something I noticed and mentioned on Gitter and someone suggested I make PR to fix it. Does that mean this PR just stays up until 3.0?

[kailuowang]

Okay, I will schedule it for 3.0

[johnynek]

isn't this superceded by @travisbrown 's great observation here: #3012

[travisbrown]

@johnynek I hadn't seen this PR, but the problem isn't removing the type parameter, it's the change from def to val and the fact that this is a trait—you end up with the kind of silly-seeming synthetic public setter.

We could go ahead and make this change but without the vals.

[johnynek]

actually, I didn't even look at the diff (like a fool). Sorry, I just read the title and assumed.

[Billzabob]

@travisbrown Can you help me understand what you mean by "synthetic public setter"? I thought zero parameter defs and vals were basically equivalent if you're not side-effecting and aren't using type parameters.

[travisbrown]

@Billzabob They're equivalent in the universal access principle sense—they look the same to callers—but they're implemented differently on the underlying platform, especially in traits, where the initialisation of vals is tricky. The compiler is mapping traits onto JVM interfaces, which don't have instance fields and up until Java 8 didn't even have instance method implementations, and the details of that mapping mean making a def a val breaks binary compatibility (definitely on 2.11 but IIRC also on 2.12 and 2.13).

[travisbrown]

@Billzabob This is a good catch and I'd personally really like to see this change in 2.0.0. Would you mind opening a new PR that only removes the type parameters? Unfortunately it'll have to happen in two places, since these instances are now part of version-specific code, but if you search for the names you'll find them.

[travisbrown]

@Billzabob Hope you don't mind but I took your commit, updated it, and opened #3042, with the hope that we can get this into today's 2.0.0.

[Billzabob]

@travisbrown I don't mind at all! Also, thanks for the explanation.

[codecov-io]

Codecov Report

Merging #2789 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #2789   +/-   ##
=======================================
  Coverage   93.46%   93.46%           
=======================================
  Files         368      368           
  Lines        6979     6979           
  Branches      184      184           
=======================================
  Hits         6523     6523           
  Misses        456      456

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ccd712a...77f5c64. Read the comment docs.

[travisbrown]

Included in #3042, which is now merged.