[Mlir-commits] [mlir] [mlir][tblgen] Expose more of MLIRTableGen as library (PR #113781)

[River Riddle]

River707 wrote:

> > I agree with @joker-eph here, I don't yet understand the goal for all of this code movement. It's not clear to me what this helping to achieve, and the original discussion is still very relevant
> 
> The prior reluctance/apprehension was about this move somehow diluting the canonicalism of ODS:
> 
> > the risk is that ODS loses it's aspect of being a clear spec
> 
> Which does not make sense for upstream to dictate - if someone somewhere (downstream) decides to implement a "custom backend" for ODS, I don't see how that affects _upstream_ backends in the slightest. In addition there's no current barrier today to this anyway because one can simply use the JSON backend and generate therefrom. This change simply enables users to take a slightly more sane approach.

I'm not sure I agree here: upstream does dictate the best practices for interacting with the broader project via ODS, what things are supported and how, and in general the development and evolution of ODS. We may decide to change/remove/evolve in any number of ways that will break downstream significantly (with little to no support). That has been a good thing in general IMO, and prevents divergence in what it means to "define" something in ODS. It doesn't prevent someone from doing something, but it is a reasonable deterrent from going off of the supported path.

https://github.com/llvm/llvm-project/pull/113781