Aman, I think we should be able to pretty easily be able to abstract some of the redundant aspects of the plans well, but I think the hardest part would be designing the methods in the builders in such a way as to avoid re-creating the same problem in code that we currently have in the json plans.
I'm thinking that we would create a pretty standard plan builder at first, but as we decide the change the plan format, rather than adjust all of the calls to the previous builder methods, we would just write a transformation from the old inputs, to the new plan syntax (except for where we absolutely need more info, in which case we have to update the method calls). We would mark the old method in the builder deprecated to prevent future use, but allow all of the previously written tests to use it. -Jason Altekruse On Fri, Mar 7, 2014 at 8:34 PM, Timothy Chen <[email protected]> wrote: > I originally added the builder interface to logical operators so optic > generating logical nodes is exposed through that. > > I think it makes sense to add for physical operators too. > > +1 > > Tim > > Sent from my iPad > > > On Mar 7, 2014, at 6:28 PM, Aman Sinha <[email protected]> wrote: > > > > Some sort of programmatic generation of physical plans would be useful > > although it is not clear how much effort it would save ... that would > > depend on how the interface is implemented. > > > > Aman > > > > > > On Fri, Mar 7, 2014 at 3:10 PM, Jason Altekruse < > [email protected]>wrote: > > > >> Hello Drillers, > >> > >> I was working on rebasing my work for logical plan -> optiq conversion > and > >> found an issue with some tests for the parquet reader. The tests that > were > >> in error were not making the build failed because I had marked them > ignore, > >> as they depended on binary files we had not decided on a good place to > >> store. This issue should hopefully be solved soon with the publishing of > >> our 'official' test dataset. > >> > >> That being said, I know that most if not all of the test plans had to be > >> amended due to the recent changes with storage engines. I wanted to see > >> what everyones' thoughts would be regarding the possibility of > generating > >> test plans though some programmatic interface, which would allow for > easy > >> changes to the code inside of the plan builder to replace the task of > >> manually re-writing plans as the syntax changes. I know this should not > >> happen often, so it may be more work than it is worth. > >> > >> -Jason Altekruse > >> >
