Allowing omitting the "T" separator from a datetime string

Our customers have issue using spark to create a dataframe on a
datetime column, because spark interpolates the provided datetime
string, but drops the "T" separator when filling the range.

It seems that omitting the "T" character is fairly acceptable, and
ISO 8601 does allow it to be omitted in a few cases. This is a
very simple patch to allow it.

Signed-off-by: Rivers Zhang <[email protected]>

Author: riverszhang89

db/types.c

@@ -7262,7 +7262,7 @@ static int _isISO8601(const char *str, int len)
     /* (month<1 || month>12)*/                                                 \
                                                                                \
     /*get day*/                                                                \
-    day = getInt(in, len, &offset, &ltoken, 1, 2, "-T", &skipped);             \
+    day = getInt(in, len, &offset, &ltoken, 1, 2, "-T ", &skipped);            \
     if (!ltoken)                                                               \
         return CONV_WRONG_MDAY;                                                \
     /*(day <1 || !is_valid_days(year, month, day))*/                           \
@@ -7307,8 +7307,14 @@ static int _isISO8601(const char *str, int len)
                 ss = 0;                                                        \
         } else                                                                 \
             mm = 0;                                                            \
-    } else                                                                     \
+    } else {                                                                   \
         hh = 0;                                                                \
+        /* Need to compensate if the last separator is a space */              \
+        /* Because we expect the timezone string to have a leading space */    \
+        /* See a few lines above for the same logic in parsing the fraction */ \
+        if (in[offset - 1] == ' ')                                             \
+            offset--; /*incremented afterwards*/                               \
+    }                                                                          \
                                                                                \
     /* fill in the blanks*/                                                    \
     out->tm.tm_year = year - 1900;                                             \

tests/datetime.test/t25.expected

@@ -0,0 +1,6 @@
+(rows inserted=1)
+(rows inserted=1)
+(d="1970-01-01T000000.000 UTC")
+(d="1970-08-17T001501.000 UTC")
+(epoch=0)
+(epoch=19700101)

tests/datetime.test/t25.sql

@@ -0,0 +1,8 @@
+SET TIMEZONE UTC
+DROP TABLE IF EXISTS t25
+CREATE TABLE t25(d datetime)$$
+INSERT INTO t25 VALUES("1970-01-01 00:00:00 UTC")
+INSERT INTO t25 VALUES("19700101")
+SELECT d FROM t25
+SELECT CAST(d AS INT) AS epoch FROM t25
+DROP TABLE t25

tests/udf.test/nextval.req renamed to tests/udf.test/nextval.0.req

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 set -x
-for t in *.req; do
+for t in `ls *.req | sort`; do
     cdb2sql --tabs --script --cdb2cfg ${CDB2_CONFIG} ${1} default - 2> ${t}.output 1>&2 < ${t}
     set -e
     diff ${t}.expected ${t}.output