I am sorry to reactivate this thread, but I disagree that all mixed densities necessarily are necessarily combinations of the pdf and pmf. A case in point is the already implemented Wiener distribution which is a bivariate distribution defined over pairs of strictly positive RTs and a binary variable. The most natural way for the Wiener would indeed be:
wiener_lpdf(rt, response| alpha, tau, beta, delta)
Currently, the PDF has to be called twice for each response with
-delta for rts hitting the lower bound. This was already discussed on the list some time ago where the original author of the Wiener PDF jumped in and said the same: https://groups.google.com/d/msg/stan-users/-6wJfA-t2cQ/Y0xLTZlZCAAJ
Especially for bivariate PDFs with possibly more than two responses (e.g., the LBA), a specification via a tuple or explicitly with two values before the
| would be ideal.