In strptime as compiled for Emscripten, when given the US locale string format ("%m/%d/%Y, %I:%M:%S %p"), it incorrectly parses "12:00:00 AM" as "12:00:00 PM". In order to fix this, we will need a custom-written string parser similar to the CustomISO8601Parser from arrow_csv.cpp. This parser uses the arrow::TimestampParser interface from Apache Arrow, and should be relatively easy to write and test out. A custom parser is required because the current issue manifests itself inside a call to strptime (especially strange, I know), and not due to any custom code.

One can reproduce the issue by adding this block of code to make_table inside emscripten.cpp (where it is guaranteed to run as part of Perspective's Table constructor):

        std::tm res;
        char* ret = strptime("3/1/2020, 12:30:55 AM", "%m/%d/%Y, %I:%M:%S %p", &res);
        if (ret != NULLPTR) {
            // parses to 2020/3/1 12:30:55, which is incorrect
            // should parse to 2020/3/1 00:30:55
            std::cout << "parsed using strptime: " << res.tm_year + 1900 << "/" << res.tm_mon + 1 << "/" << res.tm_mday << " " << res.tm_hour << ":" << res.tm_min << ":" << res.tm_sec << std::endl;
        }

Adding this same block of code in make_table in python/table.cpp generates the correct output. Tests to reproduce this issue are in #1310 - although more tests should be added to make sure no other edge cases are created.

For users, the best way to pass in datetime values is through new Date() - date strings will always be parsed as UTC.