Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add helper function to return the number of days since epoch for a week-of-month date #10604

Closed
Show file tree
Hide file tree
Changes from 3 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 40 additions & 0 deletions velox/type/TimestampConversion.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -144,6 +144,26 @@ bool isValidWeekDate(int32_t weekYear, int32_t weekOfYear, int32_t dayOfWeek) {
return true;
}

bool isValidWeekOfMonthDate(
Copy link
Collaborator

@rui-mo rui-mo Jul 30, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

According to the discussion at https://github.com/facebookincubator/velox/pull/10511/files#r1686566980, true result does not always indicate a valid date.

If the date falls outside the valid range, the week or day from the preceding or following month's will be used.

With current implementation, for some invalid date, we return user error status while for the others we return a int64_t result in the daysSinceEpochFromWeekOfMonthDate function, which seems to be confusing.

I notice the method isValidDate provides an accurate evaluation for valid date. Does it make sense to use that?
https://github.com/facebookincubator/velox/blob/main/velox/type/TimestampConversion.cpp#L518-L529

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You are right. We should remove the validation for week of month. According to SimpleDateFormat's doc the range of day is [1,7] so isValidDate isn't suitable for this.

u | Day number of week (1 = Monday, ..., 7 = Sunday) | Number | 1

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done. See e718074

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

According to SimpleDateFormat's doc the range of day is [1,7] so isValidDate isn't suitable for this.

Could you clarify the purpose of isValidWeekOfMonthDate? Is it to follow certain rule defined in SimpleDateFormat to check if a date is valid, and what are the rules? Perhaps add some comments for this function.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I want to do validation follows SimpleDateFormat. However, SimpleDateFormat does not provide clear rules or relevant documentation. For example, document states WEEK_OF_MONTH field range is from 0 to 6, , but in my tests, a value of 99 is considered valid.(https://github.com/facebookincubator/velox/pull/10511/files#r1686566980)

Values calculated for the WEEK_OF_MONTH field range from 0 to 6

Another example is the valid range for the year field, which is not documented.

I tried to determine the valid range through testing, but it is too costly and not reliable enough. I wonder how to properly validate dates in Velox. Could you give me some advice?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@rui-mo I don't believe this functionality is needed for Presto functions. Hence, we are free to define semantics in any "sensible" way. Assuming we need these semantics to match Spark, we can use Spark semantics.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the pointer. Understood.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@mbasmanova @rui-mo According to this discussion, Spark date validation rules in document don't match with the test results. Which one should I follow? Thanks.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@NEUpanning I assume we need to follow the actual behavior of Spark. How do you think?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@rui-mo I agree with you. I will change the part of date validation for following the actual behavior of Spark.

int32_t year,
int32_t month,
int32_t weekOfMonth,
int32_t dayOfWeek) {
if (dayOfWeek < 1 || dayOfWeek > 7) {
return false;
}
if (weekOfMonth < 1 || weekOfMonth > 5) {
return false;
}
if (month < 1 || month > 12) {
return false;
}
if (year < kMinYear || year > kMaxYear) {
return false;
}
return true;
}

inline bool validDate(int64_t daysSinceEpoch) {
return daysSinceEpoch >= std::numeric_limits<int32_t>::min() &&
daysSinceEpoch <= std::numeric_limits<int32_t>::max();
Expand Down Expand Up @@ -593,6 +613,26 @@ Status daysSinceEpochFromWeekDate(
return Status::OK();
}

Status daysSinceEpochFromWeekOfMonthDate(
int32_t year,
int32_t month,
int32_t weekOfMonth,
int32_t dayOfWeek,
int64_t& out) {
if (!isValidWeekOfMonthDate(year, month, weekOfMonth, dayOfWeek)) {
return Status::UserError(
"Date out of range: {}-{}-{}-{}", year, month, weekOfMonth, dayOfWeek);
}
int64_t daysSinceEpochOfFirstDayOfMonth;
VELOX_RETURN_NOT_OK(
daysSinceEpochFromDate(year, month, 1, daysSinceEpochOfFirstDayOfMonth));
int32_t firstDayOfWeek =
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I believe I asked this earlier, but I don't think I saw an answer, hence, asking again. What happens if weekOfMonth is 5, but month has only 4 weeks (Feb)? Similarly, what happens if weekOfMonth is 5 and dayOfWeek is 7 (no month has 35 days)?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In Java's SimpleDateFormat, "invalid" weekOfMonth and dayOfWeek etc. are allowed. If the date falls outside the valid range, the week or day from the preceding or following month's will be used. For example, if weekOfMonth is 5 but the current month only has 4 weeks (such as February), the first week of March will be considered as the 5th week of February. Similarly, if weekOfMonth is 5 and dayOfWeek is 7, the corresponding day from the next month will be used.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you for clarifying. Let's update method comments to explain this as it is not obvious. Do we already have tests for these cases? If not, let's make sure to add.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I added this comments for clarifying and two tests for cases that days in previous/next month.

We treat days of the previous or next months as a part of the specified WEEK_OF_MONTH. For example, if
weekOfMonth is 5 but the current month only has 4 weeks (such as February),the first week of March will be considered as the 5th week of February.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm seeing that DATE type is based on INTEGER type, hence, allows only int32_t number of days since epoch. If that's the case, then, perhaps, daysSinceEpochFromDate should return int32_t, not int64_t.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This PR is intended for date parsing, and the DateTimeFormatter::parse method returns a Timestamp. A Timestamp is based on int64_t seconds and can represent up to 106,751,991,167 days, which is calculated as (2^63 - 1) / 1000 / 86400. This value is greater than the maximum value of an int32_t. Therefore, I think that daysSinceEpochFromDate should return an int64_t.

extractISODayOfTheWeek(daysSinceEpochOfFirstDayOfMonth);
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

extractISODayOfTheWeek accepts an int32_t value while the input is of int64_t type. Shall we check for overflow?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I notice extractISODayOfTheWeek casts input to int64_t. How about changing parameter to int64_t type instead of int32_t type

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like the cast is to avoid overflow in addition and negating. If we change it as int64_t, overflow needs to considered as well by casting to a larger type.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for your explanation. I see daysSinceEpochFromWeekDate function also calls this funtion with a int64_t argument. I think we change extractISODayOfTheWeek parameter to int64_t and cast it to int128_t when do calculate would be better. How do you think?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wonder if there is any limitation for year? If the days cannot exceed INT32_MAX, it would be safe to call it.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The max year is 292278994 so the days can be (292278994-1970)*365=106,681,113,760 that is bigger than 2,147,483,647(INT32_MAX).

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we change extractISODayOfTheWeek parameter to int64_t and cast it to int128_t when do calculate

Looks fine to me. @mbasmanova What are your thoughts? Thanks.

out = daysSinceEpochOfFirstDayOfMonth - (firstDayOfWeek - 1) +
7 * (weekOfMonth - 1) + dayOfWeek - 1;
return Status::OK();
}

Status
daysSinceEpochFromDayOfYear(int32_t year, int32_t dayOfYear, int64_t& out) {
if (!isValidDayOfYear(year, dayOfYear)) {
Expand Down
9 changes: 9 additions & 0 deletions velox/type/TimestampConversion.h
Original file line number Diff line number Diff line change
Expand Up @@ -107,6 +107,15 @@ Status daysSinceEpochFromWeekDate(
int32_t dayOfWeek,
int64_t& out);

/// Computes the (signed) number of days since unix epoch (1970-01-01).
/// Returns UserError status if the date is invalid.
Status daysSinceEpochFromWeekOfMonthDate(
int32_t year,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The implementation imposes limits on the 'year' value. Would you document these?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks. Done.

int32_t month,
int32_t weekOfMonth,
int32_t dayOfWeek,
int64_t& out);

/// Computes the (signed) number of days since unix epoch (1970-01-01).
/// Returns UserError status if the date is invalid.
Status
Expand Down
29 changes: 29 additions & 0 deletions velox/type/tests/TimestampConversionTest.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -104,6 +104,35 @@ TEST(DateTimeUtilTest, fromDateInvalid) {
1970, 6, 31, "Date out of range: 1970-6-31"));
}

TEST(DateTimeUtilTest, fromWeekOfMonthDate) {
auto testDaysSinceEpochFromWeekOfMonthDate =
[](int32_t year, int32_t month, int32_t weekOfMonth, int32_t dayOfWeek) {
int64_t daysSinceEpoch;
auto status = util::daysSinceEpochFromWeekOfMonthDate(
year, month, weekOfMonth, dayOfWeek, daysSinceEpoch);
EXPECT_TRUE(status.ok());
return daysSinceEpoch;
};
EXPECT_EQ(4, testDaysSinceEpochFromWeekOfMonthDate(1970, 1, 2, 1));
EXPECT_EQ(361, testDaysSinceEpochFromWeekOfMonthDate(1971, 1, 1, 1));
EXPECT_EQ(396, testDaysSinceEpochFromWeekOfMonthDate(1971, 2, 1, 1));

EXPECT_EQ(10952, testDaysSinceEpochFromWeekOfMonthDate(2000, 1, 1, 1));
EXPECT_EQ(19905, testDaysSinceEpochFromWeekOfMonthDate(2024, 7, 1, 1));

// Before unix epoch.
EXPECT_EQ(-3, testDaysSinceEpochFromWeekOfMonthDate(1970, 1, 1, 1));
EXPECT_EQ(-2, testDaysSinceEpochFromWeekOfMonthDate(1970, 1, 1, 2));
EXPECT_EQ(-31, testDaysSinceEpochFromWeekOfMonthDate(1969, 12, 1, 1));
EXPECT_EQ(-367, testDaysSinceEpochFromWeekOfMonthDate(1969, 1, 1, 1));
EXPECT_EQ(-724, testDaysSinceEpochFromWeekOfMonthDate(1968, 1, 2, 1));
EXPECT_EQ(-719533, testDaysSinceEpochFromWeekOfMonthDate(0, 1, 1, 1));

// Negative year - BC.
EXPECT_EQ(-719561, testDaysSinceEpochFromWeekOfMonthDate(-1, 12, 1, 1));
EXPECT_EQ(-719897, testDaysSinceEpochFromWeekOfMonthDate(-1, 1, 1, 1));
}

TEST(DateTimeUtilTest, fromDateString) {
for (ParseMode mode : {ParseMode::kPrestoCast, ParseMode::kSparkCast}) {
EXPECT_EQ(0, parseDate("1970-01-01", mode));
Expand Down
Loading