Drill used to expose both scalar and aggregate functions correctly to Optiq. I'm confused on why it no longer does.
See here: https://github.com/apache/incubator-drill/blob/master/exec/java-exec/src/main/java/org/apache/drill/exec/expr/fn/DrillFunctionRegistry.java#L79 Yash, can you debug Optiq to see why this isn't already working? J On Mon, Sep 22, 2014 at 8:19 AM, Jinfeng Ni <[email protected]> wrote: > There are some differences between Drill simple functions and aggregate > functions. > > For Drill simple function, optiq will check Drill function registry whether > a function exists. So, all you need to is to provide the function template, > the function converter will automatically add the function template into > function registry. > > Aggregate functions are different in that simple function could appear > anywhere in the query, while aggregate function could be only in certain > clause ( no aggregation function in where clause, etc). Optiq have to know > one certain function is aggregate function, in order to validate such > logic. Seems we either need modify optiq to let it know new function > "regr_avgx" > is an aggregate, or Drill need provide a way to let optiq know "regr_avgx" > is aggregate function, not simple function. > > > > On Mon, Sep 22, 2014 at 7:42 AM, Jason Altekruse <[email protected] > > > wrote: > > > Hi Yash, > > > > I did not work on the code for this, but if the functions are just acting > > on columns to provide aggregates I believe they are just supported by > > Drills default mechanism for tying into the parser. We ended up modifying > > the parser with some FreeMarker hooks that allows for consumers of optiq > to > > provide custom syntax that extends the parser without forking the whole > > thing. > > > > I believe that this works together with our function registry to provide > > default support for UDFs (which our system functions are defined in the > > same way as UDFs) without having to modify the parser itself. I haven't > had > > any trouble just adding new DrillFuncs and having them be discovered by > > Drill and work after a new build. I believe I have only written scalar > > functions and not aggregations, but you can look at the SumFunctions > class > > for some reference on the differences. > > > > -Jason > > > > On Mon, Sep 22, 2014 at 7:50 AM, Yash Sharma <[email protected]> wrote: > > > > > Hi All, > > > I am planning to add new drill functions to optiq syntax for parsing. > > > We have new functions like regr_avgx/regr_avgy etc and existing ones > like > > > covar_samp,covar_pop, correlation/corr etc. > > > > > > Going through this old thread[1] I see that we need to add the > > definitions > > > in the CombinedParser.jj[2] in Optiq. > > > > > > Is that the way we still use or its handled differently now ? > > > > > > Thanks. > > > > > > 1: https://groups.google.com/forum/#!topic/optiq-dev/dkkxsHh2MnE > > > 2. > > > > > > > > > https://github.com/yssharma/optiq/blob/retired/core/src/main/codegen/templates/CombinedParser.jj > > > > > >
