fix: auto-detect 2-digit-year (M/D/YY) dates and datetimes (#36088)

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: vroom
 Title: Read and Write Rectangular Text Data Quickly
-Version: 1.7.0.2
+Version: 1.7.0.3
 Authors@R: c(
     person("Jim", "Hester", role = "aut",
            comment = c(ORCID = "0000-0002-2739-7082")),

src/DateTimeParser.h

@@ -205,7 +205,7 @@ class DateTimeParser {
 
       switch (datePart[i]) {
         case 'y':
-          if (!consumeInteger(4, &year_)) return false;
+          if (!consumeYearFlexible()) return false;
           break;
         case 'm':
           if (!consumeInteger(2, &mon_, false)) return false;
@@ -252,43 +252,87 @@ class DateTimeParser {
     return isComplete();
   }
 
-  // Heuristic for year-last date patterns: D/M/YYYY or M/D/YYYY
-  // Matches: \d{1,2}[sep]\d{1,2}[sep]\d{4}
-  // Disambiguation: if part1 > 12 → DMY; if part2 > 12 → MDY; else → MDY (default)
-  bool parseYearLastHeuristic() {
-    int part1, part2;
-
-    if (!consumeInteger(2, &part1, false)) return false;
-    if (!consumeDateSeparator()) return false;
-    if (!consumeInteger(2, &part2, false)) return false;
-    if (!consumeDateSeparator()) return false;
-    if (!consumeInteger(4, &year_)) return false;
-    if (!isComplete()) return false;
-
-    // Validate year is plausible
-    if (year_ < 1000) return false;
+  // Consume a year that may be 2 or 4 digits. 2-digit years use the same pivot
+  // as the %y format specifier (00-68 -> 2000s, 69-99 -> 1900s). 3-digit values
+  // (100-999) are implausible and rejected. (Issue #36088)
+  bool consumeYearFlexible() {
+    if (!consumeInteger(4, &year_, false)) return false;
+    if (year_ < 100) {
+      year_ += (year_ < 69) ? 2000 : 1900;
+    } else if (year_ < 1000) {
+      return false;
+    }
+    return true;
+  }
 
+  // Disambiguate a year-last date's first two components into month and day.
+  // part1 > 12 -> DMY; part2 > 12 -> MDY; otherwise default to MDY (US).
+  // Returns false if the resulting month/day are out of range.
+  bool disambiguateDayMonth(int part1, int part2) {
     if (part1 > 12) {
-      // Must be DMY
       day_ = part1;
       mon_ = part2;
     } else if (part2 > 12) {
-      // Must be MDY
       mon_ = part1;
       day_ = part2;
     } else {
-      // Ambiguous: default to MDY (US convention)
       mon_ = part1;
       day_ = part2;
     }
-
-    // Validate month and day are in plausible range
     if (mon_ < 1 || mon_ > 12) return false;
     if (day_ < 1 || day_ > 31) return false;
-
     return true;
   }
 
+  // Heuristic for year-last date patterns: D/M/Y or M/D/Y (Y = 2 or 4 digits)
+  // Matches: \d{1,2}[sep]\d{1,2}[sep]\d{2,4}
+  // Disambiguation: if part1 > 12 → DMY; if part2 > 12 → MDY; else → MDY (default)
+  bool parseYearLastHeuristic() {
+    int part1, part2;
+
+    if (!consumeInteger(2, &part1, false)) return false;
+    if (!consumeDateSeparator()) return false;
+    if (!consumeInteger(2, &part2, false)) return false;
+    if (!consumeDateSeparator()) return false;
+    if (!consumeYearFlexible()) return false;
+    if (!isComplete()) return false;
+
+    return disambiguateDayMonth(part1, part2);
+  }
+
+  // Year-last datetime heuristic: a year-last date (M/D/Y or D/M/Y, 2 or 4 digit
+  // year) followed by a T/space separator and a HH[:MM[:SS]] time with optional
+  // timezone. Mirrors the time tail of parseISO8601. (Issue #36088)
+  bool parseYearLastHeuristicDateTime() {
+    int part1, part2;
+
+    if (!consumeInteger(2, &part1, false)) return false;
+    if (!consumeDateSeparator()) return false;
+    if (!consumeInteger(2, &part2, false)) return false;
+    if (!consumeDateSeparator()) return false;
+    if (!consumeYearFlexible()) return false;
+    if (!disambiguateDayMonth(part1, part2)) return false;
+
+    // Time portion is required (date-only is handled by parseYearLastHeuristic).
+    char next;
+    if (!consumeChar(&next)) return false;
+    if (next != 'T' && next != ' ') return false;
+
+    if (!consumeInteger(2, &hour_)) return false;
+    consumeThisChar(':');
+    consumeInteger(2, &min_);
+    consumeThisChar(':');
+    consumeSeconds(&sec_, &psec_);
+
+    if (isComplete()) return true;
+
+    // Optional timezone
+    tz_ = "UTC";
+    if (!consumeTzOffset(&tzOffsetHours_, &tzOffsetMinutes_)) return false;
+
+    return isComplete();
+  }
+
   bool isComplete() { return dateItr_ == dateEnd_; }
 
   void setDate(const char* start, const char* end) {

src/guess_type.cc

@@ -120,11 +120,17 @@ static bool isDateTime(const std::string& x, LocaleInfo* pLocale) {
     return false;
   }
 
-  // Auto-detection: ISO8601 only (existing behavior — no change)
-  bool ok = parser.parseISO8601();
-  if (!ok) return false;
-  DateTime dt = parser.makeDateTime();
-  return dt.validDateTime();
+  // Auto-detection: ISO8601 first, then year-last (M/D/Y or D/M/Y) heuristic
+  // so MDY/DMY datetimes (including 2-digit years) are recognized. (Issue #36088)
+  if (parser.parseISO8601()) {
+    DateTime dt = parser.makeDateTime();
+    if (dt.validDateTime()) return true;
+  }
+
+  parser.setDate(x.c_str(), x.c_str() + x.size());
+  if (!parser.parseYearLastHeuristicDateTime()) return false;
+  DateTime dt2 = parser.makeDateTime();
+  return dt2.validDateTime();
 }
 
 std::string guess_type__(

src/vroom_dttm.cc

@@ -21,6 +21,12 @@ double parse_dttm(
     res = parser.parseDateOrder(locale->dateOrder_);
   } else if (format.empty()) {
     res = parser.parseISO8601();
+    if (!res) {
+      // Fall back to the year-last (M/D/Y or D/M/Y) heuristic so MDY/DMY
+      // datetimes (including 2-digit years) materialize. (Issue #36088)
+      parser.setDate(begin, end);
+      res = parser.parseYearLastHeuristicDateTime();
+    }
   } else {
     res = parser.parse(format);
   }

tests/testthat/test-datetime.R

@@ -596,3 +596,71 @@ test_that("vroom() reads dot-separated MDY dates", {
   expect_s3_class(result$date, "Date")
   expect_equal(result$date, as.Date(c("2024-10-02", "2024-03-15")))
 })
+
+# --- 2-digit-year (M/D/YY) auto-detection (Issue exploratory-io/tam#36088) ---
+
+test_that("vroom() auto-detects 2-digit-year MDY dates (M/D/YY)", {
+  csv <- "id,date\n1,5/29/26\n2,5/31/26\n3,12/25/26"
+  result <- vroom::vroom(I(csv), delim = ",", show_col_types = FALSE)
+  expect_s3_class(result$date, "Date")
+  expect_equal(result$date, as.Date(c("2026-05-29", "2026-05-31", "2026-12-25")))
+})
+
+test_that("vroom() auto-detects 2-digit-year DMY dates (D/M/YY)", {
+  # 29 > 12 in first part: unambiguously DMY
+  csv <- "id,date\n1,29/5/26\n2,20/1/26"
+  result <- vroom::vroom(I(csv), delim = ",", show_col_types = FALSE)
+  expect_s3_class(result$date, "Date")
+  expect_equal(result$date, as.Date(c("2026-05-29", "2026-01-20")))
+})
+
+test_that("vroom() applies the %y pivot to 2-digit years (00-68 -> 2000s, 69-99 -> 1900s)", {
+  csv <- "id,date\n1,5/29/68\n2,5/29/69"
+  result <- vroom::vroom(I(csv), delim = ",", show_col_types = FALSE)
+  expect_s3_class(result$date, "Date")
+  expect_equal(result$date, as.Date(c("2068-05-29", "1969-05-29")))
+})
+
+test_that("vroom guess_type detects 2-digit-year year-last dates", {
+  expect_true(inherits(vroom::guess_type(c("5/29/26", "5/31/26")), "collector_date"))
+})
+
+test_that("vroom() does not treat invalid or 3-digit-year values as dates", {
+  # 13/25/26: invalid as both MDY and DMY; 100/200/300: 3-digit year rejected
+  for (v in c("13/25/26", "100/200/300")) {
+    result <- vroom::vroom(I(paste0("x\n", v, "\n")), delim = ",", show_col_types = FALSE)
+    expect_type(result$x, "character")
+  }
+})
+
+test_that("vroom() auto-detects 2-digit-year MDY datetimes (M/D/YY HH:MM:SS)", {
+  csv <- "id,dt\n1,5/29/26 14:30:00\n2,12/25/26 23:59:59"
+  result <- vroom::vroom(I(csv), delim = ",", show_col_types = FALSE)
+  expect_s3_class(result$dt, "POSIXct")
+  expect_equal(
+    result$dt,
+    as.POSIXct(c("2026-05-29 14:30:00", "2026-12-25 23:59:59"), tz = "UTC")
+  )
+})
+
+test_that("vroom() auto-detects 4-digit-year MDY datetimes (M/D/YYYY HH:MM:SS)", {
+  csv <- "id,dt\n1,5/29/2026 14:30:00\n2,10/15/2024 09:00:00"
+  result <- vroom::vroom(I(csv), delim = ",", show_col_types = FALSE)
+  expect_s3_class(result$dt, "POSIXct")
+  expect_equal(
+    result$dt,
+    as.POSIXct(c("2026-05-29 14:30:00", "2024-10-15 09:00:00"), tz = "UTC")
+  )
+})
+
+test_that("vroom() reads 2-digit-year dates with explicit date_order", {
+  csv_mdy <- "id,date\n1,5/29/26\n2,3/15/26"
+  res_mdy <- vroom::vroom(I(csv_mdy), locale = locale(date_order = "mdy"), show_col_types = FALSE)
+  expect_s3_class(res_mdy$date, "Date")
+  expect_equal(res_mdy$date, as.Date(c("2026-05-29", "2026-03-15")))
+
+  csv_dmy <- "id,date\n1,29/5/26\n2,15/3/26"
+  res_dmy <- vroom::vroom(I(csv_dmy), locale = locale(date_order = "dmy"), show_col_types = FALSE)
+  expect_s3_class(res_dmy$date, "Date")
+  expect_equal(res_dmy$date, as.Date(c("2026-05-29", "2026-03-15")))
+})