[C++] Unable to parse strings into timestamps

[asfimport]

Hi,

I'm working with parquet files generated by a AWS RDS Postgres snapshot export. 

I'm trying to parse a date column stored as a string into a timestamp, but it fails.

I've managed to parse the same date format (as in the first example below) when reading from a csv, so I tried to investigate it as far as I could on my own, and here's my results:

import pyarrow as pa
import pytz

#################################################################################
## the format I get from the database
us_tz_arr = pa.array([
  "2014-12-07 07:48:59.285332+00",
  "2014-12-07 08:01:49.758975+00",
  "2014-12-07 10:11:35.884304+00"])

us_tz_arr.cast(pa.timestamp('us', tz=pytz.UTC))
-> ArrowInvalid: Failed to parse string: 2014-12-07 10:11:35.884304+00

#################################################################################
## tried removing the timezone
us_arr = pa.array([
  "2014-12-07 07:48:59.285332",
  "2014-12-07 08:01:49.758975",
  "2014-12-07 10:11:35.884304"])

us_arr.cast(pa.timestamp('us'))
-> ArrowInvalid: Failed to parse string: 2014-12-07 10:11:35.884304

#################################################################################
## tried removing the microseconds but keeping the timezone
second_tz_arr = pa.array([
  "2014-12-07 07:48:59+00",
  "2014-12-07 08:01:49+00",
  "2014-12-07 10:11:35+00"])

second_tz_arr.cast(pa.timestamp('s', tz=pytz.UTC))
-> ArrowInvalid: Failed to parse string: 2014-12-07 10:11:35+00

#################################################################################
## removing microseconds and timezone, makes it work!
s_arr = pa.array([
  "2014-12-07 07:48:59",
  "2014-12-07 08:01:49",
  "2014-12-07 10:11:35"])

s_arr.cast(pa.timestamp('s'))
-> <pyarrow.lib.TimestampArray object at 0x7fbdf81ae460>
[
  2014-12-07 07:48:59,
  2014-12-07 08:01:49,
  2014-12-07 10:11:35
]

 PS. This is my first bug report, so apologies if important things are missing.

Environment: macOS 10.15.7, Python 3.8.2
Reporter: Niclas Roos
Watchers: Rok Mihevc / @rok

Related issues:

• [C++] Strptime ignores timezone information (is fixed by)
• [C++][CSV] Timestamp parsing should accept any valid ISO 8601 without requiring custom parse strings (is related to)

_{Note: This issue was originally created as ARROW-10343. Please see the migration documentation for further details.}

[asfimport]

Antoine Pitrou / @pitrou:
Thanks for the report. This appears to work on git master:

>>> us_arr = pa.array([
...:   "2014-12-07 07:48:59.285332",
...:   "2014-12-07 08:01:49.758975",
...:   "2014-12-07 10:11:35.884304"])
>>> us_arr.cast(pa.timestamp('us'))
<pyarrow.lib.TimestampArray object at 0x7f6f11c5ea60>
[
  2014-12-07 07:48:59.285332,
  2014-12-07 08:01:49.758975,
  2014-12-07 10:11:35.884304
]

Timezone indicators are not supported currently, except for UTC where you have to use "Z", not "+00":

>>> us_arr = pa.array([
...:   "2014-12-07 07:48:59.285332Z",
...:   "2014-12-07 08:01:49.758975Z",
...:   "2014-12-07 10:11:35.884304Z"])
>>> us_arr.cast(pa.timestamp('us'))
<pyarrow.lib.TimestampArray object at 0x7f6f1199d8a0>
[
  2014-12-07 07:48:59.285332,
  2014-12-07 08:01:49.758975,
  2014-12-07 10:11:35.884304
]

[asfimport]

Antoine Pitrou / @pitrou:
Timezone indicators are supported now as of 6.0.0:

>>> us_tz_arr
<pyarrow.lib.StringArray object at 0x7f7ad653b820>
[
  "2014-12-07 07:48:59.285332+00",
  "2014-12-07 08:01:49.758975+00",
  "2014-12-07 10:11:35.884304+00"
]
>>> us_tz_arr.cast(pa.timestamp('us', tz=pytz.UTC))
<pyarrow.lib.TimestampArray object at 0x7f7ad54b6e20>
[
  2014-12-07 07:48:59.285332,
  2014-12-07 08:01:49.758975,
  2014-12-07 10:11:35.884304
]