test-match.R
In vctrs: Vector Helpers

# ------------------------------------------------------------------------------
# vec_locate_matches() - logicals

test_that("isn't confused by unspecified logical vectors", {
  x <- vec_locate_matches(logical(), NA)
  expect_identical(x$needles, integer())
  expect_identical(x$haystack, integer())

  x <- vec_locate_matches(NA, logical())
  expect_identical(x$needles, 1L)
  expect_identical(x$haystack, NA_integer_)
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - integers

test_that("can match in increasing order", {
  x <- vec_locate_matches(1:2, 1:3)
  expect_identical(x$needles, 1:2)
  expect_identical(x$haystack, 1:2)
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - doubles

test_that("can match doubles", {
  x <- vec_locate_matches(c(1, 2, 5), c(2, 2, 3, 1))

  expect_identical(x$needles, c(1L, 2L, 2L, 3L))
  expect_identical(x$haystack, c(4L, 1L, 2L, NA))
})

test_that("can match Inf and -Inf with all conditions", {
  x <- c(Inf, -Inf)
  y <- c(-Inf, 0, Inf)

  res <- vec_locate_matches(x, y, condition = "==")
  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(3L, 1L))

  res <- vec_locate_matches(x, y, condition = "<")
  expect_identical(res$needles, c(1L, 2L, 2L))
  expect_identical(res$haystack, c(NA, 2L, 3L))

  res <- vec_locate_matches(x, y, condition = "<=")
  expect_identical(res$needles, c(1L, 2L, 2L, 2L))
  expect_identical(res$haystack, c(3L, 1L, 2L, 3L))

  res <- vec_locate_matches(x, y, condition = ">")
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(1L, 2L, NA))

  res <- vec_locate_matches(x, y, condition = ">=")
  expect_identical(res$needles, c(1L, 1L, 1L, 2L))
  expect_identical(res$haystack, c(1L, 2L, 3L, 1L))
})

test_that("NA and NaN don't match numbers with equality conditions", {
  expect_identical(vec_locate_matches(1, NA_real_)$haystack, NA_integer_)
  expect_identical(vec_locate_matches(1, NaN)$haystack, NA_integer_)
  expect_identical(vec_locate_matches(NA_real_, 1)$haystack, NA_integer_)
  expect_identical(vec_locate_matches(NaN, 1)$haystack, NA_integer_)
})

test_that("NA and NaN are the same by default", {
  res <- vec_locate_matches(NA_real_, NaN)
  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, 1L)

  res <- vec_locate_matches(NaN, NA_real_)
  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, 1L)

  res <- vec_locate_matches(c(NaN, NA, NaN), c(NA, NaN, NA), condition = "==")
  expect_identical(res$needles, rep(c(1L, 2L, 3L), each = 3))
  expect_identical(res$haystack, rep(c(1L, 2L, 3L), times = 3))

  res <- vec_locate_matches(c(NaN, NA, NaN), c(NA, NaN, NA), condition = "==", multiple = "first")
  expect_identical(res$needles, c(1L, 2L, 3L))
  expect_identical(res$haystack, c(1L, 1L, 1L))
})

test_that("NA and NaN are distinct if requested", {
  res <- vec_locate_matches(c(NaN, NA, NaN), c(NA, NaN, NA), condition = "==", nan_distinct = TRUE)
  expect_identical(res$needles, c(1L, 2L, 2L, 3L))
  expect_identical(res$haystack, c(2L, 1L, 3L, 2L))

  res <- vec_locate_matches(c(NaN, NA, NaN), c(NA, NaN, NA), condition = "==", multiple = "first", nan_distinct = TRUE)
  expect_identical(res$needles, c(1L, 2L, 3L))
  expect_identical(res$haystack, c(2L, 1L, 2L))
})

test_that("NA and NaN match each other in non-equi conditions by default", {
  res <- vec_locate_matches(c(NaN, NA, NaN, 1), c(NA, NaN, NA), condition = ">=", nan_distinct = FALSE)
  expect_identical(res$needles, c(rep(c(1L, 2L, 3L), each = 3), 4L))
  expect_identical(res$haystack, c(rep(c(1L, 2L, 3L), times = 3), NA))

  res <- vec_locate_matches(c(NaN, NA, NaN, 1), c(NA, NaN, NA), condition = "<=", nan_distinct = FALSE)
  expect_identical(res$needles, c(rep(c(1L, 2L, 3L), each = 3), 4L))
  expect_identical(res$haystack, c(rep(c(1L, 2L, 3L), times = 3), NA))
})

test_that("NA and NaN never match each other in non-equi conditions if treated as distinct", {
  res <- vec_locate_matches(c(NaN, NA, NaN), c(NA, NaN, NA), condition = ">=", nan_distinct = TRUE)
  expect_identical(res$needles, c(1L, 2L, 2L, 3L))
  expect_identical(res$haystack, c(2L, 1L, 3L, 2L))

  res <- vec_locate_matches(c(NaN, NA, NaN), c(NA, NaN, NA), condition = "<=", nan_distinct = TRUE)
  expect_identical(res$needles, c(1L, 2L, 2L, 3L))
  expect_identical(res$haystack, c(2L, 1L, 3L, 2L))
})

test_that("NA and NaN are both considered incomplete no matter the value of `nan_distinct`", {
  res <- vec_locate_matches(c(NA, NaN), c(NA, NaN), incomplete = NA, nan_distinct = FALSE)
  expect_identical(res$needles, 1:2)
  expect_identical(res$haystack, c(NA_integer_, NA_integer_))

  res <- vec_locate_matches(c(NA, NaN), c(NA, NaN), incomplete = NA, nan_distinct = TRUE)
  expect_identical(res$needles, 1:2)
  expect_identical(res$haystack, c(NA_integer_, NA_integer_))
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - complex

test_that("complex can be matched", {
  x <- complex(real = 1, imaginary = c(1, 2))
  y <- complex(real = 1, imaginary = c(1, 1, 3))
  z <- complex(real = 2, imaginary = 1)

  res <- vec_locate_matches(x, y)
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(1L, 2L, NA))

  res <- vec_locate_matches(x, z)
  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(NA_integer_, NA_integer_))
})

test_that("complex order lexicographically", {
  x <- complex(real = 1, imaginary = c(1, 2, 5))
  y <- complex(real = 1, imaginary = c(1, 4, 3))

  res <- vec_locate_matches(x, y, condition = "<")
  expect_identical(res$needles, c(1L, 1L, 2L, 2L, 3L))
  expect_identical(res$haystack, c(2L, 3L, 2L, 3L, NA))
})

test_that("complex incomplete values match correctly", {
  x <- complex(real = c(NA, NA, NaN, NaN), imaginary = c(NA, NaN, NA, NaN))
  y <- complex(real = c(NA, NA, NaN, NaN), imaginary = c(NA, NaN, NA, NaN))

  # Missings can match, and all missing values should be treated equally
  res <- vec_locate_matches(x, y, condition = "==", incomplete = "compare", nan_distinct = FALSE)
  expect_identical(res$needles, rep(1:4, each = 4))
  expect_identical(res$haystack, rep(1:4, times = 4))

  res <- vec_locate_matches(x, y, condition = "==", incomplete = "match", nan_distinct = FALSE)
  expect_identical(res$needles, rep(1:4, each = 4))
  expect_identical(res$haystack, rep(1:4, times = 4))

  # Missings can match, but all combinations are different
  res <- vec_locate_matches(x, y, condition = "==", incomplete = "compare", nan_distinct = TRUE)
  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, 1:4)

  res <- vec_locate_matches(x, y, condition = "==", incomplete = "match", nan_distinct = TRUE)
  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, 1:4)

  # Missings don't match
  res <- vec_locate_matches(x, y, condition = "==", incomplete = NA)
  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, rep(NA_integer_, 4))

  # Missings don't match, but are never considered no-matches
  expect_identical(
    vec_locate_matches(x, y, condition = "==", incomplete = NA, no_match = "error"),
    vec_locate_matches(x, y, condition = "==", incomplete = NA)
  )
})

test_that("complex missing values are always grouped together (#1403)", {
  # Unlike data frames and rcrd types, for complex vectors if either element
  # is missing then the whole observation is normalised to have both components
  # be missing. This means `1+NAi` matches `2+NAi`. It also matches `2+NaNi`
  # unless `nan_distinct = TRUE`.
  x <- complex(real = c(1, 1, 2, 2, 2), imaginary = c(NA, 1, NA, 2, NaN))
  y <- x[-1]

  res <- vec_locate_matches(x, y, condition = ">=")
  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 3L, 4L, 4L, 5L, 5L))
  expect_identical(res$haystack, c(2L, 4L, 1L, 2L, 4L, 1L, 3L, 2L, 4L))
})

test_that("behavior with complex missing values matches base R", {
  skip_if(getRversion() < "3.4.0", message = "`match()` is broken with complex missings")

  x <- complex(real = c(1, 1, 2, 2, 2), imaginary = c(NA, 1, NA, 2, NaN))

  expect_identical(
    vec_locate_matches(x, x, nan_distinct = TRUE, multiple = "first")$haystack,
    match(x, x)
  )
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - characters

test_that("character ordering is done in the C locale", {
  x <- c("a", "A")
  y <- c("a", "A", "b", "B")

  # a < b, but a > A and a > B
  res <- vec_locate_matches(x, y, condition = "<")
  expect_identical(res$needles, c(1L, 2L, 2L, 2L))
  expect_identical(res$haystack, c(3L, 1L, 3L, 4L))
})

test_that("`chr_proxy_collate` can affect the matching process", {
  x <- c("a", "A")
  y <- c("a", "A")

  res <- vec_locate_matches(x, y, condition = "==")
  expect_identical(res$needles, 1:2)
  expect_identical(res$haystack, 1:2)

  res <- vec_locate_matches(x, y, condition = "==", chr_proxy_collate = tolower)
  expect_identical(res$needles, c(1L, 1L, 2L, 2L))
  expect_identical(res$haystack, c(1L, 2L, 1L, 2L))
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - lists

test_that("lists can be matched", {
  x <- list(1, 2, 1, NULL)
  y <- list(1, 1, 3, NULL)

  res <- vec_locate_matches(x, y)

  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 3L, 4L))
  expect_identical(res$haystack, c(1L, 2L, NA, 1L, 2L, 4L))
})

test_that("list incompleteness is detected", {
  res <- vec_locate_matches(list(NULL), list(NULL), incomplete = NA)

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, NA_integer_)
})

test_that("list ordering is by first appearance in `needles` (so non-equi joins don't make much sense)", {
  x <- list(3, 2, 1, NULL)
  y <- list(1, 3, 1, 3)

  res <- vec_locate_matches(x, y, condition = ">")

  # x[1] appears first, so it isn't greater than anything
  # x[2] is greater than x[1] (when x[1] is in y)
  # and so on...
  # NULL still doesn't match anything
  expect_identical(res$needles, c(1L, 2L, 2L, 3L, 3L, 4L))
  expect_identical(res$haystack, c(NA, 2L, 4L, 2L, 4L, NA))

  # With data frame columns containing list-columns
  df1 <- data_frame(col = data_frame(x = x))
  df2 <- data_frame(col = data_frame(x = y))

  expect_identical(vec_locate_matches(x, y, condition = ">"), res)
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - data frame

test_that("can match with 1 column data frames", {
  df1 <- data_frame(x = c(1L, 3L, 1L, 3L))
  df2 <- data_frame(x = c(1L, 3L, 1L))

  expect_identical(
    vec_locate_matches(df1, df2),
    vec_locate_matches(df1$x, df2$x)
  )
})

test_that("can match with >1 column data frames", {
  df1 <- data_frame(x = c(1L, 3L, 1L, 3L), y = c(1L, 4L, 1L, 2L))
  df2 <- data_frame(x = c(1L, 3L, 1L), y = c(1L, 2L, 1L))

  res <- vec_locate_matches(df1, df2, condition = c("==", "=="))

  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 3L, 4L))
  expect_identical(res$haystack, c(1L, 3L, NA, 1L, 3L, 2L))
})

test_that("can match with df-cols of varying types", {
  y <- c(1L, 1L)

  expect_needles <- c(1L, 2L)
  expect_haystack <- c(NA, 1L)

  df1 <- data_frame(x = data_frame(x = c(2L, 1L), y = y))
  df2 <- data_frame(x = data_frame(x = c(1L, 3L), y = y))

  res <- vec_locate_matches(df1, df2)
  expect_identical(res$needles, expect_needles)
  expect_identical(res$haystack, expect_haystack)

  df1 <- data_frame(x = data_frame(x = c(2, 1), y = y))
  df2 <- data_frame(x = data_frame(x = c(1, 3), y = y))

  res <- vec_locate_matches(df1, df2)
  expect_identical(res$needles, expect_needles)
  expect_identical(res$haystack, expect_haystack)

  df1 <- data_frame(x = data_frame(x = c(TRUE, FALSE), y = y))
  df2 <- data_frame(x = data_frame(x = c(FALSE, NA), y = y))

  res <- vec_locate_matches(df1, df2)
  expect_identical(res$needles, expect_needles)
  expect_identical(res$haystack, expect_haystack)

  df1 <- data_frame(x = data_frame(x = c("x", "y"), y = y))
  df2 <- data_frame(x = data_frame(x = c("y", "z"), y = y))

  res <- vec_locate_matches(df1, df2)
  expect_identical(res$needles, expect_needles)
  expect_identical(res$haystack, expect_haystack)

  df1 <- data_frame(x = data_frame(x = complex(real = c(1, 2), imaginary = c(2, 1)), y = y))
  df2 <- data_frame(x = data_frame(x = complex(real = c(2, 3), imaginary = c(1, 1)), y = y))

  res <- vec_locate_matches(df1, df2)
  expect_identical(res$needles, expect_needles)
  expect_identical(res$haystack, expect_haystack)
})

test_that("ensure that matching works if outer runs are present (i.e. `==` comes before non-equi condition)", {
  df1 <- data_frame(x = c(1, 2, 1, 1), y = c(2, 2, 3, 2))
  df2 <- data_frame(x = c(1, 1), y = c(2, 3))

  res <- vec_locate_matches(df1, df2, condition = c("==", "<="))

  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 4L, 4L))
  expect_identical(res$haystack, c(1L, 2L, NA, 2L, 1L, 2L))

  df1$z <- c(1L, 2L, 1L, 3L)
  df2$z <- c(5L, 2L)

  res <- vec_locate_matches(df1, df2, condition = c("==", "==", "<"))

  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, c(1L, NA, 2L, 1L))

  res <- vec_locate_matches(df1, df2, condition = c("==", ">=", "<"))

  expect_identical(res$needles, c(1L, 2L, 3L, 3L, 4L))
  expect_identical(res$haystack, c(1L, NA, 1L, 2L, 1L))
})

test_that("df-cols propagate an NA if any columns are incomplete", {
  df <- data_frame(x = 1, y = data_frame(x = c(1, 1, NA), y = c(1, NA, 2)))

  res <- vec_locate_matches(df, df, incomplete = "compare")
  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, 1:3)

  res <- vec_locate_matches(df, df, incomplete = "match")
  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, 1:3)

  # 2nd and 3rd rows aren't fully complete
  res <- vec_locate_matches(df, df, incomplete = NA)
  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, c(1L, NA, NA))

  res <- vec_locate_matches(df, df, incomplete = "drop")
  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, 1L)
})

test_that("df-cols aren't flattened, so `condition` is applied jointly on the df-col columns", {
  x <- data_frame(a = 1L, b = data_frame(x = 3L, y = 4L))
  y <- data_frame(a = 1L, b = data_frame(x = 2L, y = 5L))

  # In particular `x$b[1,] > y$b[1,]` because `3 > 4` and that breaks the tie
  # before any values of the `x$b$y` column are checked
  res <- vec_locate_matches(x, y, condition = c("==", ">"))
  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, 1L)
})

test_that("must have at least 1 column to match", {
  expect_snapshot(error = TRUE, {
    vec_locate_matches(data_frame(), data_frame())
  })
  expect_snapshot(error = TRUE, {
    vec_locate_matches(data_frame(), data_frame(), error_call = call("foo"))
  })
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - rcrd

test_that("rcrd types can be matched", {
  x <- new_rcrd(list(x = c(1L, 3L), y = c(1L, 4L)))
  y <- new_rcrd(list(x = c(1L, 2L), y = c(1L, 5L)))

  res <- vec_locate_matches(x, y, condition = "<=")
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(1L, 2L, NA))

  # In particular: `(3, 4) > (2, 5)` since the first elt breaks the tie
  res <- vec_locate_matches(x, y, condition = ">")
  expect_identical(res$needles, c(1L, 2L, 2L))
  expect_identical(res$haystack, c(NA, 1L, 2L))
})

test_that("rcrd type matching works with rcrd-cols", {
  x <- data_frame(a = c(1L, 1L), b = new_rcrd(list(x = c(1L, 3L), y = c(1L, 4L))))
  y <- data_frame(a = c(1L, 1L), b = new_rcrd(list(x = c(1L, 2L), y = c(1L, 5L))))

  res <- vec_locate_matches(x, y, condition = c("==", "<="))
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(1L, 2L, NA))

  res <- vec_locate_matches(x, y, condition = c("==", ">"))
  expect_identical(res$needles, c(1L, 2L, 2L))
  expect_identical(res$haystack, c(NA, 1L, 2L))
})

test_that("rcrd type incompleteness is handled correctly", {
  x <- new_rcrd(list(x = c(1L, NA), y = c(NA_integer_, NA_integer_)))
  y <- new_rcrd(list(x = c(1L, 2L, NA), y = c(NA, 5L, NA)))

  # When `incomplete = "compare"`, the types of incompleteness still must
  # match exactly to have a match. i.e. (x=1L, y=NA) doesn't match (x=NA, y=1L).
  # This is the same as the rule for data frames.
  res <- vec_locate_matches(x, y, condition = "==", incomplete = "compare")
  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(1L, 3L))

  res <- vec_locate_matches(x, y, condition = "==", incomplete = "match")
  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(1L, 3L))

  # If any field contains NA, the entire observation is incomplete.
  res <- vec_locate_matches(x, y, condition = "==", incomplete = NA)
  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(NA_integer_, NA_integer_))
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - S3

test_that("S3 types with order proxies that depend on the data are combined before the proxy is taken", {
  # i.e. `bignum:::vec_proxy_order.bignum_biginteger()`

  x <- structure(c(5L, 1L), class = "foo")
  y <- structure(c(8L, 5L), class = "foo")

  local_methods(
    vec_proxy_order.foo = function(x, ...) {
      rank(unclass(x))
    }
  )

  # Can't take the order proxies separately because they are the same!
  expect_identical(vec_proxy_order(x), vec_proxy_order(y))

  res <- vec_locate_matches(x, y)
  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(2L, NA))

  x_df <- data_frame(a = x, b = x)
  y_df <- data_frame(a = y, b = y)

  res <- vec_locate_matches(x_df, y_df)
  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(2L, NA))
})

test_that("Works with base R S3 types we support natively", {
  x <- new_factor(c(1L, 2L), levels = c("x", "y"))
  y <- new_factor(c(3L, 1L, 1L), levels = c("x", "y", "z"))
  res <- vec_locate_matches(x, y)
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(2L, 3L, NA))

  x <- new_ordered(c(1L, 2L), levels = c("x", "y"))
  y <- new_ordered(c(2L, 1L, 1L), levels = c("x", "y"))
  res <- vec_locate_matches(x, y)
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(2L, 3L, 1L))

  x <- new_date(c(1, 2))
  y <- new_date(c(3, 1, 1))
  res <- vec_locate_matches(x, y)
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(2L, 3L, NA))

  x <- new_datetime(c(1, 2))
  y <- new_datetime(c(3, 1, 1))
  res <- vec_locate_matches(x, y)
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(2L, 3L, NA))

  x <- as.POSIXlt(new_datetime(c(1, 2)))
  y <- as.POSIXlt(new_datetime(c(3, 1, 1)))
  res <- vec_locate_matches(x, y)
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(2L, 3L, NA))
})

test_that("Works with classed data frame columns", {
  x_col <- new_data_frame(list(a = c(1L, 2L), b = c(2, 3)), class = "foo")
  y_col <- new_data_frame(list(a = c(1L, 1L, 1L), b = c(2, 4, 2)), class = "foo")

  x <- new_data_frame(list(c = c(1L, 1L), d = x_col))
  y <- new_data_frame(list(c = c(1L, 1L, 1L), d = y_col))

  res <- vec_locate_matches(x, y)
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(1L, 3L, NA))
})

test_that("AsIs types are combined before order proxies are taken (#1557)", {
  x <- I(list(5, 1))
  y <- I(list(8, 5, 5))

  res <- vec_locate_matches(x, y)
  expect_identical(res$needles, c(1L, 1L, 2L))
  expect_identical(res$haystack, c(2L, 3L, NA))
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - ptype2 / casting

test_that("common type of `needles` and `haystack` is taken", {
  x <- 1
  y <- "a"

  expect_snapshot(error = TRUE, {
    vec_locate_matches(x, y)
  })
  expect_snapshot(error = TRUE, {
    vec_locate_matches(x, y, needles_arg = "x", error_call = call("foo"))
  })
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - missing values

test_that("integer missing values match with `==`, `>=`, and `<=` when `incomplete = 'compare'", {
  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "==")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "<=")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = ">=")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "<")

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, NA_integer_)

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = ">")

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, NA_integer_)
})

test_that("integer missing values can match with any condition when `incomplete = 'match'`", {
  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "==", incomplete = "match")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "<=", incomplete = "match")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = ">=", incomplete = "match")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "<", incomplete = "match")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = ">", incomplete = "match")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))
})

test_that("integer missing values report all matches even with a `filter`", {
  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "<=", filter = "min")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "<", filter = "min", incomplete = "match")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = ">=", filter = "max")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = ">", filter = "max", incomplete = "match")

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(2L, 4L))
})

test_that("integer missing value matches can be limited by `multiple`", {
  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "<=", multiple = "first")

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, 2L)

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "<=", multiple = "last")

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, 4L)

  res <- vec_locate_matches(NA_integer_, c(1L, NA, 2L, NA), condition = "<=", multiple = "any")

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, 2L)
})

test_that("missing values match within columns", {
  df1 <- data_frame(x = c(1L, 2L, 1L), y = rep(NA_integer_, 3))
  df2 <- data_frame(x = c(2L, 1L, 1L), y = c(1L, NA, NA))

  res <- vec_locate_matches(df1, df2, condition = c("==", "=="))

  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 3L))
  expect_identical(res$haystack, c(2L, 3L, NA, 2L, 3L))

  expect_identical(
    vec_locate_matches(df1, df2, condition = c("<=", ">=")),
    vec_locate_matches(df1, df2, condition = c("==", "=="))
  )

  res <- vec_locate_matches(df1, df2, condition = c("<", ">"))

  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, rep(NA_integer_, 3))

  res <- vec_locate_matches(df1, df2, condition = c("<=", ">"), incomplete = "compare")

  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, rep(NA_integer_, 3))

  res <- vec_locate_matches(df1, df2, condition = c("<=", ">"), incomplete = "match")

  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 3L))
  expect_identical(res$haystack, c(2L, 3L, NA, 2L, 3L))
})

test_that("missing values being 'match'ed hands off correctly to next column", {
  df1 <- data_frame(x = c(NA, NA, 1L, 2L, NA), y = c(2, 3, 0, 1, NA))
  df2 <- data_frame(x = c(NA, NA, NA, 3L), y = c(2, 1, NA, 0))

  res <- vec_locate_matches(df1, df2, condition = c("<", ">"), incomplete = "match")

  expect_identical(res$needles, c(1L, 2L, 2L, 3L, 4L, 5L))
  expect_identical(res$haystack, c(2L, 1L, 2L, NA, 4L, 3L))
})

test_that("integer needles can't match NAs in the haystack", {
  # At the C level, 1L > NA_integer_ (INT_MIN),
  # but we are careful to work around this
  res <- vec_locate_matches(1L, c(1L, NA, 2L, NA), condition = ">=")

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, 1L)

  res <- vec_locate_matches(1L, c(1L, NA, 2L, NA), condition = ">")

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, NA_integer_)
})

test_that("double needles can't match NAs or NaNs in the haystack", {
  # At the C level, our helpers assumg NA and NaN are the smallest values,
  # so we are careful to avoid including them with >= and > conditions
  res <- vec_locate_matches(1, c(1, NA, 2, NaN), condition = ">=")

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, 1L)

  res <- vec_locate_matches(1, c(1, NA, 2, NaN), condition = ">")

  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, NA_integer_)
})

test_that("NA and NaN match correctly with non-equi conditions and `nan_distinct`", {
  res <- vec_locate_matches(c(NA, NaN), c(1L, NA, 2L, NaN), condition = "<=", nan_distinct = TRUE)

  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(c(NA, NaN), c(1L, NA, 2L, NaN), condition = "<", nan_distinct = TRUE)

  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(NA_integer_, NA_integer_))

  res <- vec_locate_matches(c(NA, NaN), c(1L, NA, 2L, NaN), condition = "<", nan_distinct = TRUE, incomplete = "match")

  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(2L, 4L))

  res <- vec_locate_matches(c(NA, NaN), c(1L, NA, 2L, NaN), condition = ">=", nan_distinct = FALSE)

  expect_identical(res$needles, c(1L, 1L, 2L, 2L))
  expect_identical(res$haystack, c(2L, 4L, 2L, 4L))

  res <- vec_locate_matches(c(NA, NaN), c(1L, NA, 2L, NaN), condition = ">", nan_distinct = FALSE, incomplete = "match")

  expect_identical(res$needles, c(1L, 1L, 2L, 2L))
  expect_identical(res$haystack, c(2L, 4L, 2L, 4L))
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - `incomplete`

test_that("can handle incomplete needles with `incomplete = <integer>`", {
  x <- c(1L, NA, 2L)
  y <- c(NA, 1L, 1L)

  res <- vec_locate_matches(x, y, condition = "==", incomplete = NA)

  expect_identical(res$needles, c(1L, 1L, 2L, 3L))
  expect_identical(res$haystack, c(2L, 3L, NA, NA))

  res <- vec_locate_matches(x, y, condition = "<=", incomplete = 0L)

  expect_identical(res$needles, c(1L, 1L, 2L, 3L))
  expect_identical(res$haystack, c(2L, 3L, 0L, NA))

  res <- vec_locate_matches(x, y, condition = ">=", incomplete = -1L)

  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 3L))
  expect_identical(res$haystack, c(2L, 3L, -1L, 2L, 3L))
})

test_that("can drop incomplete needle rows with `incomplete = 'drop'", {
  x <- c(1L, NA, 2L)
  y <- c(NA, 1L, 1L)

  res <- vec_locate_matches(x, y, condition = "==", incomplete = "drop")
  expect_identical(res$needles, c(1L, 1L, 3L))
  expect_identical(res$haystack, c(2L, 3L, NA))
})

test_that("if `incomplete = <integer>`, an NA in any column results in the value", {
  df1 <- data_frame(x = c(1L, NA, 2L, 1L, 1L), y = c(2L, 2L, NA, 1L, 1L))
  df2 <- data_frame(x = c(1L, 1L, 2L), y = c(1L, 1L, NA))

  res <- vec_locate_matches(df1, df2, condition = c("==", "=="), incomplete = NA)

  expect_identical(res$needles, c(1L, 2L, 3L, 4L, 4L, 5L, 5L))
  expect_identical(res$haystack, c(NA, NA, NA, 1L, 2L, 1L, 2L))

  res <- vec_locate_matches(df1, df2, condition = c(">=", ">="), incomplete = NA)

  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 4L, 4L, 5L, 5L))
  expect_identical(res$haystack, c(1L, 2L, NA, NA, 1L, 2L, 1L, 2L))
})

test_that("`incomplete = <integer> / 'drop'` still handles NAs in future columns when an earlier column has no matches", {
  df1 <- data_frame(x = c(1, 1, 2, 3), y = c(1, NA, NA, 4))
  df2 <- data_frame(x = c(1, 3), y = c(1, 5))

  # The 2 in row 3 of df1 has no match, but the NA in the 2nd column still propagates
  res <- vec_locate_matches(df1, df2, incomplete = NA, no_match = -1L)

  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, c(1L, NA, NA, -1L))

  res <- vec_locate_matches(df1, df2, incomplete = "drop", no_match = -1L)

  expect_identical(res$needles, c(1L, 4L))
  expect_identical(res$haystack, c(1L, -1L))

  # The 1 in row 1 and 2 of df1 have no match, but the NA in row 2 of the 2nd column propagates
  res <- vec_locate_matches(df1, df2, incomplete = NA, no_match = -1L, condition = ">")

  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, c(-1L, NA, NA, 1L))

  res <- vec_locate_matches(df1, df2, incomplete = "drop", no_match = -1L, condition = ">")

  expect_identical(res$needles, c(1L, 4L))
  expect_identical(res$haystack, c(-1L, 1L))
})

test_that("`incomplete` can error informatively", {
  expect_snapshot({
    (expect_error(vec_locate_matches(NA, 1, incomplete = "error")))
    (expect_error(vec_locate_matches(NA, 1, incomplete = "error", needles_arg = "foo")))
    (expect_error(vec_locate_matches(NA, 1, incomplete = "error", needles_arg = "foo", error_call = call("fn"))))
  })
})

test_that("`incomplete` error is classed", {
  expect_error(vec_locate_matches(NA, 1, incomplete = "error"), class = "vctrs_error_matches_incomplete")
})

test_that("`incomplete` is validated", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1, 2, incomplete = 1.5)))
    (expect_error(vec_locate_matches(1, 2, incomplete = c("match", "drop"))))
    (expect_error(vec_locate_matches(1, 2, incomplete = "x")))
    # Uses internal call
    (expect_error(vec_locate_matches(1, 2, incomplete = "x", error_call = call("fn"))))
  })
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - `condition`

test_that("multiple matches from a non-equi condition are returned in first appearance order", {
  res <- vec_locate_matches(0L, c(1L, 0L, -1L, 0L), condition = "<=")

  expect_identical(res$needles, rep(1L, 3))
  expect_identical(res$haystack, c(1L, 2L, 4L))

  # Checking equi for good measure
  res <- vec_locate_matches(0L, c(1L, 0L, -1L, 0L), condition = "==")

  expect_identical(res$needles, rep(1L, 2))
  expect_identical(res$haystack, c(2L, 4L))
})

test_that("multiple matches from a non-equi condition are returned in first appearance order when the matches are in different nesting containers", {
  df <- data_frame(x = 0, y = 0)
  df2 <- data_frame(x = 2:1, y = 1:2)

  res <- vec_locate_matches(df, df2, condition = c("<=", "<="))

  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(1L, 2L))
})

test_that("`condition` is validated", {
  expect_error(vec_locate_matches(1, 2, condition = 1), "`condition` must be a character vector")
  expect_error(vec_locate_matches(1, 2, condition = "x"), 'must only contain "==", ">", ">=", "<", or "<="')
  expect_error(vec_locate_matches(1, 2, condition = c("==", "==")), "must be length 1, or the same length as the number of columns of the input")
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - `multiple`

test_that("can get all matches", {
  x <- vec_locate_matches(c(1L, 3L), c(1L, 3L, 1L, 3L), multiple = "all")

  expect_identical(x$needles, c(1L, 1L, 2L, 2L))
  expect_identical(x$haystack, c(1L, 3L, 2L, 4L))
})

test_that("can get first match", {
  x <- vec_locate_matches(c(1L, 3L), c(1L, 3L, 1L, 3L), multiple = "first")

  expect_identical(x$needles, 1:2)
  expect_identical(x$haystack, 1:2)
})

test_that("can get last match", {
  x <- vec_locate_matches(c(1L, 3L), c(1L, 3L, 1L, 3L), multiple = "last")

  expect_identical(x$needles, 1:2)
  expect_identical(x$haystack, 3:4)
})

test_that("can get any match", {
  x <- vec_locate_matches(c(1L, 3L), c(1L, 3L, 1L, 3L), multiple = "any")

  expect_identical(x$needles, 1:2)
  expect_identical(x$haystack, 1:2)
})

test_that("duplicate needles match the same haystack locations", {
  x <- vec_locate_matches(c(1L, 3L, 1L, 3L), c(1L, 3L, 1L), multiple = "all")

  expect_identical(x$needles, c(1L, 1L, 2L, 3L, 3L, 4L))
  expect_identical(x$haystack, c(1L, 3L, 2L, 1L, 3L, 2L))
})

test_that("correctly gets all matches when they come from different nesting containers", {
  needles <- data_frame(
    a = c(1, 8),
    b = c(2, 9)
  )
  haystack <- data_frame(
    a = c(6, 5),
    b = c(6, 7)
  )

  expect_identical(
    vec_locate_matches(needles, haystack, condition = "<", multiple = "all"),
    data_frame(needles = c(1L, 1L, 2L), haystack = c(1L, 2L, NA))
  )
})

test_that("correctly gets first/last/any match when they come from different nesting containers", {
  needles <- data_frame(
    a = c(1, 8),
    b = c(2, 9)
  )
  haystack <- data_frame(
    a = c(6, 5, 0),
    b = c(6, 7, 1)
  )

  expect_identical(
    vec_locate_matches(needles, haystack, condition = "<", multiple = "first"),
    data_frame(needles = c(1L, 2L), haystack = c(1L, NA))
  )
  expect_identical(
    vec_locate_matches(needles, haystack, condition = "<", multiple = "last"),
    data_frame(needles = c(1L, 2L), haystack = c(2L, NA))
  )
  expect_identical(
    vec_locate_matches(needles, haystack, condition = "<", multiple = "any"),
    data_frame(needles = c(1L, 2L), haystack = c(2L, NA))
  )
  expect_identical(
    vec_locate_matches(needles, haystack, condition = "<", multiple = "first", remaining = NA_integer_),
    data_frame(needles = c(1L, 2L, NA, NA), haystack = c(1L, NA, 2L, 3L))
  )
  expect_identical(
    vec_locate_matches(needles, haystack, condition = "<", multiple = "last", remaining = NA_integer_),
    data_frame(needles = c(1L, 2L, NA, NA), haystack = c(2L, NA, 1L, 3L))
  )
  expect_identical(
    vec_locate_matches(needles, haystack, condition = "<", multiple = "any", remaining = NA_integer_),
    data_frame(needles = c(1L, 2L, NA, NA), haystack = c(2L, NA, 1L, 3L))
  )
})

test_that("`multiple` is validated", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1, 2, multiple = 1.5)))
    (expect_error(vec_locate_matches(1, 2, multiple = c("first", "last"))))
    (expect_error(vec_locate_matches(1, 2, multiple = "x")))
    # Uses internal error
    (expect_error(vec_locate_matches(1, 2, multiple = "x", error_call = call("fn"))))
  })
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - `multiple` (deprecated)

test_that("`multiple` can error informatively", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1L, c(1L, 1L), multiple = "error")))
    (expect_error(vec_locate_matches(1L, c(1L, 1L), multiple = "error", needles_arg = "foo")))
    (expect_error(vec_locate_matches(1L, c(1L, 1L), multiple = "error", needles_arg = "foo", error_call = call("fn"))))
    (expect_error(vec_locate_matches(1L, c(1L, 1L), multiple = "error", needles_arg = "foo", haystack_arg = "bar")))
  })
})

test_that("`multiple` can warn informatively", {
  expect_snapshot({
    (expect_warning(vec_locate_matches(1L, c(1L, 1L), multiple = "warning")))
    (expect_warning(vec_locate_matches(1L, c(1L, 1L), multiple = "warning", needles_arg = "foo")))
    (expect_warning(vec_locate_matches(1L, c(1L, 1L), multiple = "warning", needles_arg = "foo", error_call = call("fn"))))
    (expect_warning(vec_locate_matches(1L, c(1L, 1L), multiple = "warning", needles_arg = "foo", haystack_arg = "bar")))
  })
})

test_that("warning falls back to 'all'", {
  expect_warning(
    result <- vec_locate_matches(c(1L, 3L, 1L, 3L), c(1L, 3L, 1L), multiple = "warning"),
    class = "vctrs_warning_matches_multiple"
  )

  expect_identical(
    result,
    vec_locate_matches(c(1L, 3L, 1L, 3L), c(1L, 3L, 1L), multiple = "all")
  )
})

test_that("errors on multiple matches that come from different nesting containers", {
  df <- data_frame(x = 0, y = 0)
  df2 <- data_frame(x = 1:2, y = 2:1)

  expect_snapshot(error = TRUE, {
    vec_locate_matches(df, df2, condition = c("<=", "<="), multiple = "error")
  })
})

test_that("errors when a match from a different nesting container is processed early on", {
  # Row 1 has 2 matches
  # Row 2 has 0 matches
  needles <- data_frame(
    a = c(1, 8),
    b = c(2, 9)
  )

  # Rows 1 and 2 end up in different nesting containers
  haystack <- data_frame(
    a = c(5, 6),
    b = c(7, 6)
  )

  # needles[1,] records the haystack[1,] match first, which is in the 1st
  # value of `loc_first_match_o_haystack`, then records the haystack[3,] match
  # which is in the 3rd value of `loc_first_match_o_haystack` even though it
  # is processed 2nd (i.e. we need to use `loc` rather than `i` when detecting
  # multiple matches)
  expect_snapshot(error = TRUE, {
    vec_locate_matches(needles, haystack, condition = "<", multiple = "error")
  })
})

test_that("`multiple = 'error'` doesn't error errneously on the last observation", {
  expect_error(res <- vec_locate_matches(1:2, 1:2, multiple = "error"), NA)
  expect_identical(res$needles, 1:2)
  expect_identical(res$haystack, 1:2)
})

test_that("`multiple = 'error' / 'warning'` throw correctly when combined with `relationship`", {
  x <- c(1, 2, 2)
  y <- c(2, 1, 2)

  # `multiple` error technically fires first
  expect_snapshot({
    (expect_error(vec_locate_matches(x, y, relationship = "one-to-one", multiple = "error")))
  })

  # Works when warning is also requested
  expect_snapshot({
    (expect_error(vec_locate_matches(x, y, relationship = "warn-many-to-many", multiple = "error")))
  })
  # Both warnings are thrown if applicable
  expect_snapshot({
    vec_locate_matches(x, y, relationship = "warn-many-to-many", multiple = "warning")
  })
  # Both warning and error are thrown if applicable
  expect_snapshot(error = TRUE, {
    vec_locate_matches(x, y, relationship = "one-to-one", multiple = "warning")
  })

  x <- c(1, 2)
  y <- c(2, 1, 2)

  expect_snapshot({
    (expect_error(vec_locate_matches(x, y, relationship = "warn-many-to-many", multiple = "error")))
  })
  # Only `multiple` warning is applicable here
  expect_snapshot({
    vec_locate_matches(x, y, relationship = "warn-many-to-many", multiple = "warning")
  })
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - `relationship`

test_that("`relationship` handles one-to-one case", {
  # No error
  expect_identical(
    vec_locate_matches(1:2, 2:1, relationship = "one-to-one"),
    vec_locate_matches(1:2, 2:1)
  )

  # Doesn't care about the zero match case
  expect_identical(
    vec_locate_matches(1:2, 3:4, relationship = "one-to-one"),
    vec_locate_matches(1:2, 3:4)
  )

  expect_snapshot({
    (expect_error(vec_locate_matches(c(2, 1), c(1, 1), relationship = "one-to-one")))
    (expect_error(vec_locate_matches(c(1, 1), c(1, 2), relationship = "one-to-one")))
  })
})

test_that("`relationship` handles one-to-many case", {
  # No error
  expect_identical(
    vec_locate_matches(c(1, 2), c(1, 2, 2), relationship = "one-to-many"),
    vec_locate_matches(c(1, 2), c(1, 2, 2))
  )

  # Doesn't care about the zero match case
  expect_identical(
    vec_locate_matches(1:2, 3:4, relationship = "one-to-many"),
    vec_locate_matches(1:2, 3:4)
  )

  expect_snapshot({
    (expect_error(vec_locate_matches(c(1, 2, 2), c(2, 1), relationship = "one-to-many")))
  })
})

test_that("`relationship` handles many-to-one case", {
  # No error
  expect_identical(
    vec_locate_matches(c(1, 2, 2), c(1, 2), relationship = "many-to-one"),
    vec_locate_matches(c(1, 2, 2), c(1, 2))
  )

  # Doesn't care about the zero match case
  expect_identical(
    vec_locate_matches(1:2, 3:4, relationship = "many-to-one"),
    vec_locate_matches(1:2, 3:4)
  )

  expect_snapshot({
    (expect_error(vec_locate_matches(c(1, 2), c(1, 2, 2), relationship = "many-to-one")))
  })
})

test_that("`relationship` handles many-to-many case", {
  # No error
  expect_identical(
    vec_locate_matches(c(1, 2, 2), c(1, 2), relationship = "many-to-many"),
    vec_locate_matches(c(1, 2, 2), c(1, 2))
  )

  # No error
  expect_identical(
    vec_locate_matches(c(1, 2), c(1, 2, 2), relationship = "many-to-many"),
    vec_locate_matches(c(1, 2), c(1, 2, 2))
  )

  # No error
  expect_identical(
    vec_locate_matches(c(1, 1, 2), c(1, 2, 2), relationship = "many-to-many"),
    vec_locate_matches(c(1, 1, 2), c(1, 2, 2))
  )

  # Doesn't care about the zero match case
  expect_identical(
    vec_locate_matches(1:2, 3:4, relationship = "many-to-many"),
    vec_locate_matches(1:2, 3:4)
  )
})

test_that("`relationship` handles warn-many-to-many case", {
  # No warning
  expect_identical(
    expect_silent(
      vec_locate_matches(c(1, 2, 2), c(1, 2), relationship = "warn-many-to-many")
    ),
    vec_locate_matches(c(1, 2, 2), c(1, 2))
  )

  # No warning
  expect_identical(
    expect_silent(
      vec_locate_matches(c(1, 2), c(1, 2, 2), relationship = "warn-many-to-many")
    ),
    vec_locate_matches(c(1, 2), c(1, 2, 2))
  )

  # Doesn't care about the zero match case
  expect_identical(
    expect_silent(
      vec_locate_matches(1:2, 3:4, relationship = "warn-many-to-many")
    ),
    vec_locate_matches(1:2, 3:4)
  )

  # Specifically designed to ensure we test both:
  # - Finding multiple `needles` matches before multiple `haystack` matches
  # - Finding multiple `haystack` matches before multiple `needles` matches
  expect_snapshot({
    (expect_warning(vec_locate_matches(c(1, 2, 1), c(1, 2, 2), relationship = "warn-many-to-many")))
    (expect_warning(vec_locate_matches(c(1, 1, 2), c(2, 2, 1), relationship = "warn-many-to-many")))
  })
})

test_that("`relationship` considers `incomplete` matches as possible multiple matches", {
  x <- c(1, NA, NaN)
  y <- c(NA, 1)

  expect_snapshot({
    (expect_error(vec_locate_matches(x, y, relationship = "one-to-many")))
  })

  # No error
  expect_identical(
    vec_locate_matches(x, y, relationship = "one-to-many", incomplete = NA),
    vec_locate_matches(x, y, incomplete = NA)
  )

  # No error
  expect_identical(
    vec_locate_matches(x, y, relationship = "one-to-many", nan_distinct = TRUE),
    vec_locate_matches(x, y, nan_distinct = TRUE)
  )
})

test_that("`relationship` errors on multiple matches that come from different nesting containers", {
  df <- data_frame(x = 0, y = 0)
  df2 <- data_frame(x = 1:2, y = 2:1)

  expect_snapshot({
    (expect_error(vec_locate_matches(df, df2, condition = c("<=", "<="), relationship = "many-to-one")))
  })
})

test_that("`relationship` errors when a match from a different nesting container is processed early on", {
  # Row 1 has 2 matches
  # Row 2 has 0 matches
  needles <- data_frame(
    a = c(1, 8),
    b = c(2, 9)
  )

  # Rows 1 and 2 end up in different nesting containers
  haystack <- data_frame(
    a = c(5, 6),
    b = c(7, 6)
  )

  # needles[1,] records the haystack[1,] match first, which is in the 1st
  # value of `loc_first_match_o_haystack`, then records the haystack[3,] match
  # which is in the 3rd value of `loc_first_match_o_haystack` even though it
  # is processed 2nd (i.e. we need to use `loc` rather than `i` when detecting
  # multiple matches)
  expect_snapshot({
    (expect_error(vec_locate_matches(needles, haystack, condition = "<", relationship = "many-to-one")))
  })
})

test_that("`relationship` doesn't error errneously on the last observation", {
  expect_error(res <- vec_locate_matches(1:2, 1:2, relationship = "many-to-one"), NA)
  expect_identical(res$needles, 1:2)
  expect_identical(res$haystack, 1:2)
})

test_that("`relationship` doesn't error if `multiple` removes multiple matches", {
  out <- vec_locate_matches(c(1, 2), c(1, 1), multiple = "any", relationship = "one-to-one")
  expect_identical(out$needles, c(1L, 2L))
  expect_identical(out$haystack, c(1L, NA))

  out <- vec_locate_matches(c(1, 2), c(1, 1), multiple = "first", relationship = "one-to-one")
  expect_identical(out$needles, c(1L, 2L))
  expect_identical(out$haystack, c(1L, NA))

  out <- vec_locate_matches(c(1, 2), c(1, 1), multiple = "last", relationship = "one-to-one")
  expect_identical(out$needles, c(1L, 2L))
  expect_identical(out$haystack, c(2L, NA))
})

test_that("`relationship` can still detect problematic `haystack` relationships when `multiple = first/last` are used", {
  expect_snapshot({
    (expect_error(vec_locate_matches(c(3, 1, 1), c(2, 1, 3, 3), multiple = "first", relationship = "one-to-one")))
    (expect_error(vec_locate_matches(c(3, 1, 1), c(2, 1, 3, 3), multiple = "first", relationship = "one-to-many")))
  })
})

test_that("`relationship` and `remaining` work properly together", {
  expect_snapshot({
    out <- vec_locate_matches(
      c(1, 2, 2),
      c(2, 3, 1, 1, 4),
      relationship = "warn-many-to-many",
      remaining = NA_integer_
    )
  })
  expect_identical(out$needles, c(1L, 1L, 2L, 3L, NA, NA))
  expect_identical(out$haystack, c(3L, 4L, 1L, 1L, 2L, 5L))
})

test_that("`relationship` errors if `condition` creates multiple matches", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1, c(1, 2), condition = "<=", relationship = "many-to-one")))
  })
})

test_that("`relationship` doesn't error if `filter` removes multiple matches", {
  out <- vec_locate_matches(1, c(1, 2), condition = "<=", filter = "min", relationship = "many-to-one")
  expect_identical(out$needles, 1L)
  expect_identical(out$haystack, 1L)

  out <- vec_locate_matches(1, c(1, 2), condition = "<=", filter = "max", relationship = "many-to-one")
  expect_identical(out$needles, 1L)
  expect_identical(out$haystack, 2L)
})

test_that("`relationship` still errors if `filter` hasn't removed all multiple matches", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1, c(1, 2, 1), condition = "<=", filter = "min", relationship = "many-to-one")))
  })

  # But not here
  out <- vec_locate_matches(c(1, 1), c(1, 2, 1), condition = "<=", filter = "max", relationship = "many-to-one")
  expect_identical(out$needles, c(1L, 2L))
  expect_identical(out$haystack, c(2L, 2L))
})

test_that("`relationship` errors when we have >1 size 1 matches across containers (tidyverse/dplyr#6835)", {
  # Carefully designed to ensure we get 2 nested containment groups that split
  # up the rows of `y`, but each of the nested containment groups contain exactly
  # 1 match, so `size_match` in `expand_compact_indices()` won't ever be >1
  x <- data_frame(a = 1L, b = 5L)
  y <- data_frame(a = c(1L, 2L), b = c(4L, 3L))

  expect_snapshot(error = TRUE, {
    vec_locate_matches(
      x,
      y,
      condition = c("<=", ">="),
      filter = c("none", "none"),
      relationship = "one-to-one"
    )
  })
})

test_that("`relationship` doesn't error when the first match from a different container gets filtered out (tidyverse/dplyr#6835)", {
  # Carefully designed to ensure we get 2 nested containment groups that split
  # up the rows of `y`. Row 1 (processed first) doesn't hold the minimum `b`
  # value, so it gets filtered out. Row 2 is in the "extra" matches section
  # but is actually the first (and only) real match, so we don't want to error
  # on it.
  x <- data_frame(a = 1L, b = 5L)
  y <- data_frame(a = c(1L, 2L), b = c(4L, 3L))

  out <- vec_locate_matches(
    x,
    y,
    condition = c("<=", ">="),
    filter = c("none", "min"),
    relationship = "one-to-one"
  )
  expect_identical(out$needles, 1L)
  expect_identical(out$haystack, 2L)

  # Similar to the above example, but with a `max` filter. Row 1 doesn't hold
  # the max `c` value so it is filtered out even though it is a `>=` match.
  x <- data_frame(a = 1L, b = 5L, c = 3L)
  y <- data_frame(a = c(1L, 2L), b = c(4L, 3L), c = c(1L, 2L))

  out <- vec_locate_matches(
    x,
    y,
    condition = c("<=", ">=", ">="),
    filter = c("none", "none", "max"),
    relationship = "one-to-one"
  )
  expect_identical(out$needles, 1L)
  expect_identical(out$haystack, 2L)
})

test_that("`relationship` errors respect argument tags and error call", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1L, c(1L, 1L), relationship = "one-to-one", needles_arg = "foo", haystack_arg = "bar", error_call = call("fn"))))
    (expect_error(vec_locate_matches(c(1L, 1L), 1L, relationship = "one-to-one", needles_arg = "foo", haystack_arg = "bar", error_call = call("fn"))))
    (expect_error(vec_locate_matches(c(1L, 1L), 1L, relationship = "one-to-many", needles_arg = "foo", haystack_arg = "bar", error_call = call("fn"))))
    (expect_error(vec_locate_matches(1L, c(1L, 1L), relationship = "many-to-one", needles_arg = "foo", haystack_arg = "bar", error_call = call("fn"))))
  })
})

test_that("`relationship` warnings respect argument tags and error call", {
  expect_snapshot({
    (expect_warning(vec_locate_matches(c(1L, 1L), c(1L, 1L), relationship = "warn-many-to-many", needles_arg = "foo", haystack_arg = "bar", error_call = call("fn"))))
    (expect_warning(vec_locate_matches(c(1L, 1L), c(1L, 1L), relationship = "warn-many-to-many", needles_arg = "foo", error_call = call("fn"))))
    (expect_warning(vec_locate_matches(c(1L, 1L), c(1L, 1L), relationship = "warn-many-to-many", haystack_arg = "bar", error_call = call("fn"))))
  })
})

test_that("`relationship` is validated", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1, 2, relationship = 1.5)))
    (expect_error(vec_locate_matches(1, 2, relationship = c("one-to-one", "one-to-many"))))
    (expect_error(vec_locate_matches(1, 2, relationship = "x")))
    # Uses internal error
    (expect_error(vec_locate_matches(1, 2, relationship = "x", error_call = call("fn"))))
  })
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - `no_match`

test_that("can control `no_match`", {
  x <- vec_locate_matches(1:3, 1L)
  expect_identical(x$haystack, c(1L, NA, NA))

  x <- vec_locate_matches(1:3, 1L, no_match = 0L)
  expect_identical(x$haystack, c(1L, 0L, 0L))
})

test_that("can drop unmatched needles", {
  x <- vec_locate_matches(1:3, 2L, no_match = "drop")
  expect_identical(x$needles, 2L)
  expect_identical(x$haystack, 1L)
})

test_that("can drop unmatched missings when `incomplete = 'match'`", {
  x <- vec_locate_matches(c(NaN, 2, NA), 2, no_match = "drop")
  expect_identical(x$needles, 2L)
  expect_identical(x$haystack, 1L)

  x <- vec_locate_matches(c(NaN, 2, NA), NA, no_match = "drop", nan_distinct = FALSE)
  expect_identical(x$needles, c(1L, 3L))
  expect_identical(x$haystack, c(1L, 1L))

  x <- vec_locate_matches(c(NaN, 2, NA), NA, no_match = "drop", nan_distinct = TRUE)
  expect_identical(x$needles, 3L)
  expect_identical(x$haystack, 1L)
})

test_that("can differentiate between `no_match` and `incomplete`", {
  res <- vec_locate_matches(c(1, NA), 2, incomplete = NA, no_match = -1L)

  expect_identical(res$needles, 1:2)
  expect_identical(res$haystack, c(-1L, NA))
})

test_that("`no_match` can error informatively", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1, 2, no_match = "error")))
    (expect_error(vec_locate_matches(1, 2, no_match = "error", needles_arg = "foo")))
    (expect_error(vec_locate_matches(1, 2, no_match = "error", needles_arg = "foo", error_call = call("fn"))))
    (expect_error(vec_locate_matches(1, 2, no_match = "error", needles_arg = "foo", haystack_arg = "bar")))
  })
})

test_that("`no_match = 'error'` doesn't error on handled incomplete values", {
  res <- vec_locate_matches(c(NA, NaN, NA, 1), c(NA, 1), incomplete = NA, no_match = "error")

  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, c(rep(NA, 3), 2L))
})

test_that("`no_match = 'drop'` doesn't drop handled incomplete values", {
  res <- vec_locate_matches(c(NA, NaN, NA, 1), c(NA, 1), incomplete = NA, no_match = "drop")

  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, c(rep(NA, 3), 2L))
})

test_that("errors with the right location on unmatched needles when different nesting containers are present", {
  df <- data_frame(x = 2:1, y = 2:1)
  df2 <- data_frame(x = 1:2, y = 2:1)

  # i.e. should be location 2
  expect_snapshot(
    (expect_error(vec_locate_matches(df, df2, condition = ">=", no_match = "error")))
  )
})

test_that("`no_match` is validated", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1, 2, no_match = 1.5)))
    (expect_error(vec_locate_matches(1, 2, no_match = c(1L, 2L))))
    (expect_error(vec_locate_matches(1, 2, no_match = "x")))
    # Uses internal call
    (expect_error(vec_locate_matches(1, 2, no_match = "x", error_call = call("fn"))))
  })
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - `remaining`

test_that("`remaining` can retain `haystack` values that `needles` didn't match", {
  res <- vec_locate_matches(1, 0:2, remaining = NA)
  expect_identical(res$needles, c(1L, NA, NA))
  expect_identical(res$haystack, c(2L, 1L, 3L))

  res <- vec_locate_matches(1, 0:2, remaining = NA, condition = ">=")
  expect_identical(res$needles, c(1L, 1L, NA))
  expect_identical(res$haystack, c(1L, 2L, 3L))

  res <- vec_locate_matches(1, 0:2, remaining = NA, condition = "<")
  expect_identical(res$needles, c(1L, NA, NA))
  expect_identical(res$haystack, c(3L, 1L, 2L))
})

test_that("`incomplete` affects `needles` but not `haystack`", {
  # Matches NA to NA, so nothing remaining
  res <- vec_locate_matches(c(1, NA), c(NA, 1), incomplete = "compare", remaining = NA)
  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(2L, 1L))

  # Matches NA to NA, so nothing remaining
  res <- vec_locate_matches(c(1, NA), c(NA, 1), incomplete = "match", remaining = NA)
  expect_identical(res$needles, c(1L, 2L))
  expect_identical(res$haystack, c(2L, 1L))

  # Doesn't match NA to NA, so `haystack` is left with remaining values
  res <- vec_locate_matches(c(1, NA), c(NA, 1), condition = "<", incomplete = "compare", remaining = NA)
  expect_identical(res$needles, c(1L, 2L, NA, NA))
  expect_identical(res$haystack, c(NA, NA, 1L, 2L))

  # Matches NA to NA, so only remaining value is for `1`
  res <- vec_locate_matches(c(1, NA), c(NA, 1), condition = "<", incomplete = "match", remaining = NA)
  expect_identical(res$needles, c(1L, 2L, NA))
  expect_identical(res$haystack, c(NA, 1L, 2L))

  # `needles` NA value is propagated, so `haystack` is left with a remaining value
  res <- vec_locate_matches(c(1, NA), c(NA, 1), incomplete = NA, remaining = NA)
  expect_identical(res$needles, c(1L, 2L, NA))
  expect_identical(res$haystack, c(2L, NA, 1L))

  # `needles` NA value is dropped, so `haystack` is left with a remaining value
  res <- vec_locate_matches(c(1, NA), c(NA, 1), incomplete = "drop", remaining = NA)
  expect_identical(res$needles, c(1L, NA))
  expect_identical(res$haystack, c(2L, 1L))
})

test_that("`remaining` combined with `multiple = 'first/last'` treats non-first/last matches as remaining", {
  x <- c(1, 2)
  y <- c(1, 2, 2)

  res <- vec_locate_matches(x, y, remaining = NA, multiple = "first")
  expect_identical(res$needles, c(1L, 2L, NA))
  expect_identical(res$haystack, c(1L, 2L, 3L))

  res <- vec_locate_matches(x, y, remaining = NA, multiple = "last")
  expect_identical(res$needles, c(1L, 2L, NA))
  expect_identical(res$haystack, c(1L, 3L, 2L))

  res <- vec_locate_matches(x, y, remaining = NA, multiple = "any")
  expect_identical(res$needles, c(1L, 2L, NA))
  expect_identical(res$haystack, c(1L, 2L, 3L))
})

test_that("`remaining` combined with the haystack reordering retains appearance order", {
  x <- data_frame(a = 1, b = 4)
  y <- data_frame(a = c(2, 1, 0), b = c(2, 1, 0))

  # Appearance order for the haystack locations
  res <- vec_locate_matches(x, y, condition = c("<=", ">="))
  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(1L, 2L))

  # Retain that appearance order of the matches, with remaining values appended
  res <- vec_locate_matches(x, y, condition = c("<=", ">="), remaining = NA)
  expect_identical(res$needles, c(1L, 1L, NA))
  expect_identical(res$haystack, c(1L, 2L, 3L))
})

test_that("`remaining` can error informatively", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1, 2, remaining = "error")))
    (expect_error(vec_locate_matches(1, 2, remaining = "error", needles_arg = "foo")))
    (expect_error(vec_locate_matches(1, 2, remaining = "error", needles_arg = "foo", error_call = call("fn"))))
    (expect_error(vec_locate_matches(1, 2, remaining = "error", needles_arg = "foo", haystack_arg = "bar")))
  })
})

test_that("`remaining` is validated", {
  expect_snapshot({
    (expect_error(vec_locate_matches(1, 2, remaining = 1.5)))
    (expect_error(vec_locate_matches(1, 2, remaining = c(1L, 2L))))
    (expect_error(vec_locate_matches(1, 2, remaining = "x")))
    # Uses internal call
    (expect_error(vec_locate_matches(1, 2, remaining = "x", error_call = call("fn"))))
  })
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - filter

test_that("simple `filter`s work", {
  needles <- c(1, 2, 4)
  haystack <- c(2, 1, 3, 0)

  res <- vec_locate_matches(needles, haystack, condition = "<", filter = "max")
  expect_identical(res$haystack, c(3L, 3L, NA))

  res <- vec_locate_matches(needles, haystack, condition = "<", filter = "min")
  expect_identical(res$haystack, c(1L, 3L, NA))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max")
  expect_identical(res$haystack, c(2L, 1L, 3L))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "min")
  expect_identical(res$haystack, c(4L, 4L, 4L))
})

test_that("haystack duplicates are preserved", {
  needles <- c(1, 2, 4)
  haystack <- c(2, 1, 2, 3, 0, 1, 0)

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max")
  expect_identical(res$needles, c(1L, 1L, 2L, 2L, 3L))
  expect_identical(res$haystack, c(2L, 6L, 1L, 3L, 4L))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "min")
  expect_identical(res$needles, c(1L, 1L, 2L, 2L, 3L, 3L))
  expect_identical(res$haystack, c(5L, 7L, 5L, 7L, 5L, 7L))
})

test_that("haystack duplicates can be controlled by `multiple`", {
  needles <- c(1, 2, 4)
  haystack <- c(2, 1, 2, 3, 0, 1, 0)

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max", multiple = "first")
  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, c(2L, 1L, 4L))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max", multiple = "last")
  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, c(6L, 3L, 4L))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max", multiple = "any")
  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, c(2L, 1L, 4L))
})

test_that("`filter` works when valid matches are in different nesting containers", {
  needles <- data_frame(x = 0L, y = 1L, z = 2L)
  haystack <- data_frame(x = c(1L, 2L, 1L, 0L), y = c(2L, 1L, 2L, 3L), z = c(3L, 3L, 2L, 2L))

  info <- compute_nesting_container_info(haystack, c("<=", "<=", "<="))
  haystack_order <- info[[1]]
  container_ids <- info[[2]]

  # Rows 1 and 2 of haystack are in different nesting containers, but
  # both have the "max" filter value of `z=3` so both should be in the result.
  # Row 4 is in its own container, so it will be considered the "max"
  # of its group, but it is less than rows 1 and 2 so it will ultimately be
  # filtered out.
  expect_identical(container_ids, c(1L, 2L, 1L, 0L))
  expect_identical(haystack_order, c(4L, 3L, 1L, 2L))

  res <- vec_locate_matches(needles, haystack, condition = c("<=", "<=", "<="), filter = c("none", "none", "max"))
  expect_identical(res$needles, c(1L, 1L))
  expect_identical(res$haystack, c(1L, 2L))

  res <- vec_locate_matches(needles, haystack, condition = c("<=", "<=", "<="), filter = c("none", "none", "max"), multiple = "first")
  expect_identical(res$haystack, 1L)

  res <- vec_locate_matches(needles, haystack, condition = c("<=", "<=", "<="), filter = c("none", "none", "max"), multiple = "last")
  expect_identical(res$haystack, 2L)

  res <- vec_locate_matches(needles, haystack, condition = c("<=", "<=", "<="), filter = c("none", "none", "max"), multiple = "any")
  expect_identical(res$haystack, 1L)
})

test_that("single filter is applied to all columns", {
  needles <- data_frame(x = 5L, y = 8L, z = 4L)
  haystack <- data_frame(x = c(1L, 3L, 2L, 2L), y = c(1L, 3L, 2L, 3L), z = c(1L, 2L, 3L, 3L))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max")
  expect_identical(res$haystack, 2L)

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "min")
  expect_identical(res$haystack, 1L)
})

test_that("different `filter`s can be used per column", {
  needles <- data_frame(x = c(0, 2, 1, 1), y = c(2, 0, 0, 4))
  haystack <- data_frame(x = c(2, 2, 2, 1, 1), y = c(1, 1, 2, 2, 1))

  res <- vec_locate_matches(needles, haystack, condition = c(">=", "<"), filter = c("max", "min"))
  expect_identical(res$needles, c(1L, 2L, 2L, 3L, 4L))
  expect_identical(res$haystack, c(NA, 1L, 2L, 5L, NA))
})

test_that("`filter` works with incomplete values", {
  needles <- c(1, NA, 4, NA)
  haystack <- c(NA, 1, NA, 1, 3)

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max", incomplete = "compare")
  expect_identical(res$needles, c(1L, 1L, 2L, 2L, 3L, 4L, 4L))
  expect_identical(res$haystack, c(2L, 4L, 1L, 3L, 5L, 1L, 3L))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max", incomplete = "compare", multiple = "first")
  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, c(2L, 1L, 5L, 1L))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max", incomplete = "compare", multiple = "any")
  expect_identical(res$needles, 1:4)
  expect_identical(res$haystack, c(2L, 1L, 5L, 1L))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "max", incomplete = NA)
  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 4L))
  expect_identical(res$haystack, c(2L, 4L, NA, 5L, NA))
})

test_that("`filter` works with mixed NA and NaN", {
  needles <- c(1, NA, 4, NaN)
  haystack <- c(NA, 1, NaN, 1, 3)

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "min", incomplete = "compare", nan_distinct = FALSE)
  expect_identical(res$needles, c(1L, 1L, 2L, 2L, 3L, 3L, 4L, 4L))
  expect_identical(res$haystack, c(2L, 4L, 1L, 3L, 2L, 4L, 1L, 3L))

  res <- vec_locate_matches(needles, haystack, condition = ">=", filter = "min", incomplete = "compare", nan_distinct = TRUE)
  expect_identical(res$needles, c(1L, 1L, 2L, 3L, 3L, 4L))
  expect_identical(res$haystack, c(2L, 4L, 1L, 2L, 4L, 3L))
})

test_that("`filter` is validated", {
  expect_error(vec_locate_matches(1, 2, filter = 1.5), "character vector")
  expect_error(vec_locate_matches(1, 2, filter = "x"), 'one of "none", "min", or "max"')
  expect_error(vec_locate_matches(1, 2, filter = c("min", "max")), "length 1, or the same length as")
})

# ------------------------------------------------------------------------------
# vec_locate_matches() - edge cases

test_that("zero row `needles` results in zero row data frame output", {
  res <- vec_locate_matches(integer(), 1:3)

  expect_identical(res$needles, integer())
  expect_identical(res$haystack, integer())

  res <- vec_locate_matches(integer(), 1:3, condition = "<")

  expect_identical(res$needles, integer())
  expect_identical(res$haystack, integer())
})

test_that("zero row `haystack` results in no-matches for all needles", {
  res <- vec_locate_matches(1:3, integer())

  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, rep(NA_integer_, 3))

  res <- vec_locate_matches(1:3, integer(), condition = "<")

  expect_identical(res$needles, 1:3)
  expect_identical(res$haystack, rep(NA_integer_, 3))
})

test_that("zero row `haystack` still allows needle incomplete handling", {
  res <- vec_locate_matches(c(1, NA), integer(), incomplete = NA, no_match = 0L)

  expect_identical(res$needles, 1:2)
  expect_identical(res$haystack, c(0L, NA))

  res <- vec_locate_matches(c(1, NA), integer(), incomplete = NA, no_match = 0L, condition = "<")

  expect_identical(res$needles, 1:2)
  expect_identical(res$haystack, c(0L, NA))
})

test_that("zero column data frames are not allowed", {
  expect_error(
    vec_locate_matches(data_frame(.size = 2L), data_frame(.size = 2L)),
    "at least 1 column"
  )
})

test_that("zero column input still checks `condition` correctness", {
  x <- data_frame(.size = 2)
  y <- data_frame(.size = 3)

  expect_error(
    vec_locate_matches(x, y, condition = c("==", "<=")),
    "length 1, or the same length as the number of columns"
  )
})

test_that("`multiple = 'first'/'last'` returns the first/last by appearance", {
  x <- c(1, 2, 3)
  y <- c(2, 1, 0)

  res <- vec_locate_matches(x, y, condition = ">=", multiple = "first")
  expect_identical(res$haystack, c(2L, 1L, 1L))

  res <- vec_locate_matches(x, y, condition = ">=", multiple = "last")
  expect_identical(res$haystack, c(3L, 3L, 3L))
})

test_that("NA adjustment of `>` and `>=` conditions is protected from empty haystack", {
  res <- vec_locate_matches(1L, integer(), condition = ">")
  expect_identical(res$needles, 1L)
  expect_identical(res$haystack, NA_integer_)
})

test_that("potential overflow on large output size is caught informatively", {
  # Windows 32-bit doesn't support long vectors of this size, and the
  # intermediate `r_ssize` will be too large
  skip_if(.Machine$sizeof.pointer < 8L, message = "No long vector support")

  expect_snapshot({
    (expect_error(vec_locate_matches(1:1e7, 1:1e7, condition = ">=")))
  })
})