On Mon Jul 10 02:57:02 UTC 2017, Nikita Churaev wrote:

...

You can maybe fix that up with FunDeps or TypeFamilies, but then you lose most of TDNR's benefits (esp not coping with existential quanted fields , which is typically a requirement to cope with OOP-style).

What benefits exactly are lost?

Hi Nikita, I'm not sure I (or anybody) has the energy for another round of the "records debate". The TDNR (ex-)proposal seems so attractive because it just doesn't consider in detail all the complexities. They're magically swept under the carpet with the special dot syntax. So you might expect TDNR would support all of the functionality with H98 field names, plus having multiple fields of the same name. But how? Note the comment on TDNR section 'The Proposal' "TDNR is not "overloading" in the sense of type classes." And there were plenty of complaints that it's confusingly similar-but-different to type classes. You could look at SPJ's re-working of TDNR [2010] as SORF [2013]: https://ghc.haskell.org/trac/ghc/wiki/Records/OverloadedRecordFields/SORF SORF took advantage of the new Symbol Kinds (not available in 2010), and tried to be more type-class-like, with the `Has` class. One of the motivations with SORF is "Records with polymorphic fields are very important in practice, so it would be a major wart not to support them." TDNR just didn't discuss higher-ranked types. It turned out that was just too hard, even for SORF. (Or rather for OverloadedRecordFields, which is what's being delivered from SORF.) SORF tries to tackle record update, for which TDNR said "However I propose not to apply it [TDNR] to record update. " I don't think our cousins in OOP would be impressed by a records system in which you can't write polymorphic record updates. (I.e. a method to update the `firstName` in any record that has that field.) Actually delivering improvements in GHC around records is being painfully slow. `DuplicateRecordFields` has eased the pain a little. Ironically `OverloadedRecordFields` doesn't tackle higher-ranked types, nor record updates. And needs special syntax `#firstName`. So perhaps TDNR's "benefits" were never achievable? ANtC