Unlike as_theta_matrix(), which reshapes whatever it is given, this errors
rather than guess. A bare vector is read as a single row and must therefore
have exactly d entries; a length that does not match used to be recycled
into a matrix of the right width, which turned one wrong-length parameter
vector into several parameter sets and a plausible-looking answer.
Arguments
- value
The user's value: a numeric vector, matrix or data frame.
- d
Required number of columns, or
NULLto accept any width.- arg
Name of the argument, as it appears in the user's call.
- what
Optional phrase describing what a column means, e.g.
"one parameter per column". Shown in parentheses.- min_rows
Smallest number of rows accepted.
0(the default) accepts an empty matrix, which is right for most callers –within_support()on zero draws is a legitimate, if odd, question. An observation argument is not:resolve_obs()passesmin_rows = 1so a zero-rowobs/x(a real-data filter that happened to drop every row) gets a named error here instead of reachingx[1, , drop = FALSE]and failing with a bare, unnamed "subscript out of bounds" (#349).