Fix td_formula covariate parsing (tests + docs; fix via barry#26)

DESCRIPTION

@@ -1,7 +1,7 @@
 Package: defm
 Type: Package
 Title: Estimation and Simulation of Multi-Binary Response Models
-Version: 0.2.2.0
+Version: 0.2.2.9000
 Authors@R: c(
     person("George", "Vega Yon", role=c("aut", "cre"),
     email="g.vegayon@gmail.com", comment = c(ORCID = "0000-0002-3171-0844")),
@@ -22,7 +22,6 @@ License: MIT + file LICENSE
 Encoding: UTF-8
 LazyData: true
 Roxygen: list(markdown = TRUE)
-RoxygenNote: 7.3.3
 LinkingTo: Rcpp, barry (>= 0.2.2)
 Imports: 
     Rcpp,
@@ -34,3 +33,4 @@ Suggests:
     texreg, tinytest, barry
 Remotes:
     USCbiostats/barryr
+Config/roxygen2/version: 8.0.0

NEWS.md

@@ -1,3 +1,20 @@
+# defm 0.2.2.9000 (development version)
+
+* Building against `barry` (>= 0.2.2.9000), currently available from GitHub
+  (USCbiostats/barryr), fixes the `td_formula()` parser (USCbiostats/barry#26).
+  The CRAN version of `barry` still has these bugs:
+  - Covariate names containing `y<digit>` (e.g., `Day1`) were read as motif
+    terms, silently changing the model: `"{y0} > {y0} x Day1"` became
+    `Motif {a+}>{a+, b+} x Day1`.
+  - `"{...}x Covar"` (no space before `x`) silently dropped the covariate.
+  - Transitions such as `"{y0_1} > {y1}"`, with a current-time term on the
+    left-hand side, were silently turned into intercept motifs; they now
+    throw an error.
+
+* The same `barry` version also fixes wrong draws in `sim_defm()` caused by a
+  support-vs-array index confusion (USCbiostats/barry#25).
+
+
 # defm 0.2.2.0
 
 * Requires `barry` (>= 0.2.2), which fixes a hash-collision bug that made

R/RcppExports.R

@@ -432,6 +432,10 @@ td_generic <- function(m, mat, covar = "") {
 #' - Intercept effect: `{...} x Hispanic` interacts with the Hispanic covar.
 #' - Transition effect: `{...} > {...} x Hispanic` Same.
 #'
+#' The `x` must be followed by a space. Only the terms inside the curly
+#' brackets are part of the motif, so covariate names such as `Day1` are
+#' fine.
+#'
 #' ## Intercept effects
 #'
 #' Intercept effects only involve a single set of curly brackets. Using the

inst/tinytest/test_formula_parser.R

@@ -0,0 +1,59 @@
+# Regression tests for the formula parser (USCbiostats/barry#26): covariate
+# names are not motif terms, `}x Covar` keeps the covariate, and the LHS of a
+# transition cannot include the current time.
+set.seed(1)
+n  <- 40
+id <- rep(1:10, each = 4)
+Y  <- matrix(
+  rbinom(n * 2, 1, .5), ncol = 2,
+  dimnames = list(NULL, c("a", "b"))
+  )
+X  <- cbind(Day1 = runif(n), Female = rbinom(n, 1, .5))
+
+new_model <- function() new_defm(id, Y, X, order = 1)
+
+# The fix lives in barry's headers (LinkingTo), i.e., it is compiled into
+# defm. Instead of checking barry's version, we probe the parser defm was
+# built with: the unfixed parser reads `y1` from `Day1` and throws
+# "The term y1 shows more than once in the formula."
+parser_fixed <- tryCatch({
+  td_formula(new_model(), "{y0} > {y1} x Day1")
+  TRUE
+}, error = function(e) FALSE)
+
+if (!parser_fixed)
+  exit_file("defm was built against a barry without the formula parser fix.")
+
+# Covariate names containing y<digit>
+m <- new_model()
+td_formula(m, "{y0} > {y0} x Day1")
+td_formula(m, "{y0} > {y1} x Day1")
+td_formula(m, "{y0, 0y1} x Day1")
+expect_equal(
+  names(m),
+  c(
+    "Motif {a+}>{a+} x Day1",
+    "Motif {a+}>{b+} x Day1",
+    "Motif {a+, b-} x Day1"
+  )
+)
+
+# The statistic is weighted by the covariate
+init_defm(m)
+S    <- get_stats(m)
+prev <- rbind(NA, Y[-n, ])
+prev[c(TRUE, id[-1] != id[-n]), ] <- NA
+expect_equal(
+  unname(S[, 1]),
+  ifelse(is.na(prev[, 1]), NA, (prev[, 1] == 1 & Y[, 1] == 1) * X[, "Day1"])
+)
+
+# No space before `x` keeps the covariate
+m <- new_model()
+td_formula(m, "{y0} > {y1}x Female")
+expect_equal(names(m), "Motif {a+}>{b+} x Female")
+
+# The LHS cannot include the current time
+m <- new_model()
+expect_error(td_formula(m, "{y0_1} > {y1}"), "LHS")
+expect_error(td_formula(m, "{y0_5} > {y1}"), "out of range")

src/terms.cpp

@@ -168,6 +168,10 @@ SEXP td_generic(
 //' - Intercept effect: `{...} x Hispanic` interacts with the Hispanic covar.
 //' - Transition effect: `{...} > {...} x Hispanic` Same.
 //'
+//' The `x` must be followed by a space. Only the terms inside the curly
+//' brackets are part of the motif, so covariate names such as `Day1` are
+//' fine.
+//'
 //' ## Intercept effects
 //'
 //' Intercept effects only involve a single set of curly brackets. Using the