Add make_timestamp Spark function #8812

marin-ma · 2024-02-21T05:10:41Z

Add sparksql function make_timestamp for non-ansi behavior, which creates a timestamp from year, month, day, hour, min, sec and timezone (optional) fields. The timezone field indicates the timezone of the input timestamp. If not specified, the input timestamp is treated as the time in the session timezone. The output datatype is timestamp type, which internally stores the number of microseconds from the epoch of 1970-01-01T00:00:00.000000Z (UTC+00:00)

In spark, the result is shown as the time in the session timezone(spark.sql.session.timeZone).

set spark.sql.session.timeZone=Asia/Shanghai;
SELECT make_timestamp(2014, 12, 28, 6, 30, 45.887); -- 2014-12-28 06:30:45.887
SELECT make_timestamp(2014, 12, 28, 6, 30, 45.887, 'CET'); -- 2014-12-28 13:30:45.887

In Velox, it should return the timestamp in UTC timezone, so that the result can be correctly converted by Spark for display.

The non-ansi behavior returns NULL for invalid inputs.

Spark documentation:

https://spark.apache.org/docs/latest/api/sql/#make_timestamp

Spark implementation:

https://github.com/apache/spark/blob/branch-3.3/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/datetimeExpressions.scala#L2512-L2712

netlify · 2024-02-21T05:10:59Z

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	`16124b3`
🔍 Latest deploy log	https://app.netlify.com/sites/meta-velox/deploys/65f1059baa5ac90008086be9

marin-ma · 2024-02-21T05:11:05Z

cc: @rui-mo @PHILO-HE

rui-mo · 2024-02-22T08:48:13Z

velox/functions/sparksql/DateTimeFunctions.cpp

+    DecodedVector* micros) {
+  auto totalMicros = micros->valueAt<int64_t>(row);
+  auto seconds = totalMicros / util::kMicrosPerSec;
+  auto nanos = totalMicros % util::kMicrosPerSec;


nit: maybe the variables seconds and nanos are not needed if they're only for a check. It seems the value of nanos is not accurate (to be multiplied by 10^3), but does not affect the nanos == 0 check here.

rui-mo · 2024-02-22T08:56:02Z

velox/functions/sparksql/DateTimeFunctions.cpp

+VELOX_DECLARE_VECTOR_FUNCTION(
+    udf_make_timestamp,
+    MakeTimestampFunction::signatures(),
+    std::make_unique<MakeTimestampFunction>());


Maybe better to declare as a stateful vector function, and check argument size and types in a function creation method instead of in apply.

rui-mo · 2024-02-22T08:57:11Z

velox/functions/sparksql/DateTimeFunctions.cpp

+    VELOX_USER_CHECK(
+        microsType.scale() == 6,
+        "Seconds fraction must have 6 digits for microseconds but got {}",
+        microsType.scale());


Could be moved to a method creation method.

rui-mo · 2024-02-22T08:59:36Z

velox/functions/sparksql/DateTimeFunctions.cpp

+      const auto sessionTzName = queryConfig.sessionTimezone();
+      if (!sessionTzName.empty()) {
+        sessionTzID = util::getTimeZoneID(sessionTzName);
+      }


int64_t sessionTzID = sessionTzName.empty() ? 0 : util::getTimeZoneID(sessionTzName);

rui-mo · 2024-02-22T09:01:30Z

velox/functions/sparksql/DateTimeFunctions.cpp

+      // use default value 0(UTC timezone).
+      int64_t sessionTzID = 0;
+      const auto& queryConfig = context.execCtx()->queryCtx()->queryConfig();
+      const auto sessionTzName = queryConfig.sessionTimezone();


By declaring this function as stateful, sessionTzName could become a construction parameter of MakeTimestampFunction, instead of being calculated for each apply.

rui-mo · 2024-02-22T09:04:38Z

velox/functions/sparksql/tests/DateTimeFunctionsTest.cpp

+    auto result = hasTimeZone
+        ? evaluate("make_timestamp(c0, c1, c2, c3, c4, c5, c6)", data)
+        : evaluate("make_timestamp(c0, c1, c2, c3, c4, c5)", data);
+    facebook::velox::test::assertEqualVectors(expected, result);


Maybe we need to test different encodings as a fast path for constant encoding is implemented. testEncodings could be used for that purpose.

@rui-mo Rui has a good point. Would you address her comment?

@mbasmanova Method testConstantTimezone below was used to address this comment.

marin-ma · 2024-02-23T06:34:21Z

@rui-mo Could you help to review again? Thanks!

PHILO-HE

Let's keep consistency for PR title by using "Add xxx Spark function". Thanks!
See community convention: https://github.com/facebookincubator/velox/blob/main/CONTRIBUTING.md#adding-scalar-functions

PHILO-HE · 2024-02-23T07:41:57Z

velox/docs/functions/spark/datetime.rst

@@ -201,3 +201,21 @@ These functions support TIMESTAMP and DATE input types.
 .. spark:function:: year(x) -> integer

    Returns the year from ``x``.
+
+.. spark:function:: make_timestamp(year, month, day, hour, min, sec[, timezone]) -> timestamp


Please keep function name in alphabetical order.

PHILO-HE · 2024-02-23T08:03:19Z

velox/functions/sparksql/DateTimeFunctions.cpp

+    auto localMicros =
+        hour * util::kMicrosPerHour + minute * util::kMicrosPerMinute + micros;
+    return util::fromDatetime(daysSinceEpoch, localMicros);
+  } catch (const VeloxException& e) {


If user error can only be VeloxUserError here, maybe we can just catch such exception.

PHILO-HE · 2024-02-23T08:03:38Z

velox/functions/sparksql/DateTimeFunctions.cpp

+      throw;
+    }
+    return std::nullopt;
+  } catch (const std::exception&) {


Can be removed?

PHILO-HE · 2024-02-23T08:31:57Z

velox/functions/sparksql/DateTimeFunctions.cpp

+        monthVector->valueAt<int32_t>(row),
+        dayVector->valueAt<int32_t>(row));
+    auto localMicros =
+        hour * util::kMicrosPerHour + minute * util::kMicrosPerMinute + micros;


Can we simply call util::fromTime()?

PHILO-HE · 2024-02-23T08:40:05Z

velox/functions/sparksql/DateTimeFunctions.cpp

+      }
+    } else {
+      // Otherwise use session timezone. If session timezone is not specified,
+      // use default value 0(UTC timezone).


I once noted Spark always has a session timezone in its config. By default, it's the one detected from OS. And we always let Gluten pass session timezone to Velox. So maybe, if session timezone is not found from config, we can simply throw an exception.

rui-mo · 2024-02-26T07:07:29Z

velox/functions/sparksql/DateTimeFunctions.cpp

+  VELOX_USER_CHECK(
+      inputArgs[5].type->isShortDecimal(),
+      "Seconds must be short decimal type but got {}",
+      inputArgs[5].type->toString());


Maybe add type checks for all input args.

I wonder why it would be necessary to check all input args. This decimal check is aligned with Spark's https://github.com/apache/spark/blob/e98872fb5d07d570e6d0516b49a5d2e58876d1a6/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/datetimeExpressions.scala#L2601-L2602

rui-mo · 2024-02-26T07:10:57Z

velox/functions/sparksql/DateTimeFunctions.cpp

+    auto localMicros = util::fromTime(hour, minute, 0, (int32_t)micros);
+    return util::fromDatetime(daysSinceEpoch, localMicros);
+  } catch (const VeloxUserError& e) {
+    return std::nullopt;


To use try-catch for each row may have poor performace. Maybe we can take use of Status to represent the computing outcome of daysSinceEpochFromDate method.

Maybe we can optimize that in a separate PR. @mbasmanova Do you have any suggestion?

@rui-mo Rui has a good point. We should not throw and catch exceptions per row. See https://velox-lib.io/blog/optimize-try_cast.

marin-ma · 2024-02-27T06:39:28Z

@mbasmanova Could you help to review? Thanks!

mbasmanova

Some comments.

velox/functions/sparksql/DateTimeFunctions.h

mbasmanova · 2024-02-27T11:34:58Z

velox/functions/sparksql/DateTimeFunctions.cpp

+  }
+}
+
+class MakeTimestampFunction : public exec::VectorFunction {


Can this function be implemented as a simple function?

The 6th parameter is of decimal type. I'm not sure it it's possible to implement it as a simple function.

The 6th parameter is of decimal type.

Wow... I didn't realize that. Would you update the documentation to state that clearly? Are there restrictions on precision and scale? I assume scale must be <= 3.

@marin-ma Can you remind me why can't we implement functions with decimal inputs as simple functions? I wonder if it is worth looking into extending the framework to support such functions. Does Fuzzer support such functions? If not, we need to extend it.

CC: @rui-mo @PHILO-HE @FelixYBW

@mbasmanova Looks like the resolver used by simple fucntions doesn't support resolving decimal type https://github.com/facebookincubator/velox/blob/main/velox/expression/UdfTypeResolver.h
The expression fuzzer test doesn't use above framework so it can create decimal vectors.

The expression fuzzer test doesn't use above framework so it can create decimal vectors.

@marin-ma Are you saying that Fuzzer does cover this function? Would you run the Fuzzer with --only make_timestamp to make sure there are no failures?

@mbasmanova I get exception with this command ./velox/expression/tests/spark_expression_fuzzer_test --only make_timestamp

E0311 22:59:16.962633 2410940 Exceptions.h:69] Line: /home/sparkuser/github/oap-project/velox/velox/expression/tests/ExpressionFuzzer.cpp:1415, Function:fuzzReturnType, Expression: !signatures_.empty() No function signature available., Source: RUNTIME, ErrorCode: INVALID_STATE terminate called after throwing an instance of 'facebook::velox::VeloxRuntimeError' what(): Exception: VeloxRuntimeError Error Source: RUNTIME Error Code: INVALID_STATE Reason: No function signature available. Retriable: False Expression: !signatures_.empty() Function: fuzzReturnType File: /home/sparkuser/github/oap-project/velox/velox/expression/tests/ExpressionFuzzer.cpp

Does it mean this function is not covered by the fuzzer test? Or did I use the wrong command?

@marin-ma I believe Fuzzer doesn't support DECIMAL types yet. It would be nice to add this support, otherwise, test coverage of VectorFunctions that use DECIMAL types is limited.

CC: @rui-mo @PHILO-HE @majetideepak

@mbasmanova Yes, fuzzer test for decimal type is not supported. #5791 (comment) is about my previous finding , and I will remove the two limitations to see where the gap is, thanks.

mbasmanova · 2024-02-27T11:35:45Z

velox/functions/sparksql/DateTimeFunctions.cpp

+    auto localMicros = util::fromTime(hour, minute, 0, (int32_t)micros);
+    return util::fromDatetime(daysSinceEpoch, localMicros);
+  } catch (const VeloxUserError& e) {
+    return std::nullopt;


@rui-mo Rui has a good point. We should not throw and catch exceptions per row. See https://velox-lib.io/blog/optimize-try_cast.

marin-ma · 2024-03-07T00:48:30Z

@mbasmanova Could you help to review again? Thanks!

marin-ma · 2024-03-08T00:12:30Z

@mbasmanova Could you please help to review again? Thanks!

mbasmanova

@marin-ma Overall looks good. Some comments below.

mbasmanova · 2024-03-08T12:42:43Z

velox/functions/sparksql/DateTimeFunctions.cpp

+    auto* resultFlatVector = result->as<FlatVector<Timestamp>>();
+
+    exec::DecodedArgs decodedArgs(rows, args, context);
+    auto year = decodedArgs.at(0);


auto*

here and in the next few lines

mbasmanova · 2024-03-08T12:44:25Z

velox/functions/sparksql/DateTimeFunctions.cpp

+            util::getTimeZoneID(args[6]
+                                    ->asUnchecked<ConstantVector<StringView>>()
+                                    ->valueAt(0)
+                                    .str());


do we need to copy this value to std::string via .str()? I see that util::getTimeZoneID accepts an std::string_view.

mbasmanova · 2024-03-08T12:45:24Z

velox/functions/sparksql/DateTimeFunctions.cpp

+        rows.applyToSelected([&](vector_size_t row) {
+          auto timestamp = makeTimeStampFromDecodedArgs(
+              row, year, month, day, hour, minute, micros);
+          setTimestampOrNull(row, timestamp, constantTzID, resultFlatVector);


In what cases the result is NULL while no input is NULL? Please, update documentation to describe these and add test cases. Please, double check that this behavior matches Spark.

This behavior aligns with spark's output under non-ansi mode. Spark returns NULL if for invalid inputs, such as month > 12, seconds > 60, etc. Added examples with invalid inputs in the document.

mbasmanova · 2024-03-08T12:45:33Z

velox/functions/sparksql/DateTimeFunctions.cpp

+          setTimestampOrNull(row, timestamp, constantTzID, resultFlatVector);
+        });
+      } else {
+        auto timeZone = decodedArgs.at(6);


mbasmanova · 2024-03-08T12:46:13Z

velox/functions/sparksql/DateTimeFunctions.cpp

+          auto timestamp = makeTimeStampFromDecodedArgs(
+              row, year, month, day, hour, minute, micros);
+          auto tzID =
+              util::getTimeZoneID(timeZone->valueAt<StringView>(row).str());


same comment re: .str()

mbasmanova · 2024-03-08T12:53:40Z

velox/docs/functions/spark/datetime.rst

@@ -139,6 +139,35 @@ These functions support TIMESTAMP and DATE input types.

        SELECT quarter('2009-07-30'); -- 3

+.. spark:function:: make_timestamp(year, month, day, hour, min, sec[, timezone]) -> timestamp


do not abbreviate: min -> minute, sec -> second

mbasmanova · 2024-03-08T12:53:55Z

velox/docs/functions/spark/datetime.rst

+    Otherwise the function assumes the inputs are in the session's configured time zone.
+    Requires ``session_timezone`` to be set, or an exceptions will be thrown.
+
+    Arguments:


Would you generate the docs and verify that they render nicely?

mbasmanova · 2024-03-08T12:54:22Z

velox/functions/sparksql/tests/DateTimeFunctionsTest.cpp

@@ -17,6 +17,7 @@
 #include "velox/common/base/tests/GTestUtils.h"
 #include "velox/functions/sparksql/tests/SparkFunctionBaseTest.h"
 #include "velox/type/tz/TimeZoneMap.h"
+#include "velox/vector/tests/utils/VectorTestBase.h"


Is this include needed?

mbasmanova · 2024-03-08T12:55:22Z

velox/functions/sparksql/tests/DateTimeFunctionsTest.cpp

+    auto result = hasTimeZone
+        ? evaluate("make_timestamp(c0, c1, c2, c3, c4, c5, c6)", data)
+        : evaluate("make_timestamp(c0, c1, c2, c3, c4, c5)", data);
+    facebook::velox::test::assertEqualVectors(expected, result);


@rui-mo Rui has a good point. Would you address her comment?

mbasmanova · 2024-03-08T12:56:53Z

velox/functions/sparksql/tests/DateTimeFunctionsTest.cpp

+    testMakeTimestamp(data, expected, true);
+  }
+
+  // Invalid cases.


Would you split this test method into 2: valid and invalid cases.

marin-ma · 2024-03-11T08:11:25Z

@mbasmanova Here's the generated doc:

facebook-github-bot · 2024-03-12T03:10:17Z

@mbasmanova has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

facebook-github-bot · 2024-03-12T13:15:13Z

@mbasmanova has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

mbasmanova · 2024-03-12T22:56:06Z

@marin-ma #9044 landed. Would you rebase this PR so it can specify fixed scale in the function signature?

marin-ma · 2024-03-13T01:26:05Z

@mbasmanova I've rebased my branch and addressed it in the latest commit https://github.com/marin-ma/velox-oap/tree/make-timestamp However, it appears that these changes have not been synchronized with the PR. Do you have any idea on how to resolve it?

facebook-github-bot · 2024-03-13T05:46:54Z

@mbasmanova has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

mbasmanova · 2024-03-13T05:47:44Z

I've rebased my branch and addressed it in the latest commit

Thanks.

However, it appears that these changes have not been synchronized with the PR. Do you have any idea on how to resolve it?

What do you mean by "have not been synchronized with the PR"? Would you elaborate a bit?

marin-ma · 2024-03-13T05:51:49Z

What do you mean by "have not been synchronized with the PR"? Would you elaborate a bit?

@mbasmanova Earlier, the commit history in this PR wasn't updated, and it didn't trigger CI after the rebase. But it looks normal now.

Summary: Add sparksql function `make_timestamp` for non-ansi behavior, which creates a timestamp from year, month, day, hour, min, sec and timezone (optional) fields. The timezone field indicates the timezone of the input timestamp. If not specified, the input timestamp is treated as the time in the session timezone. The output datatype is timestamp type, which internally stores the number of microseconds from the epoch of `1970-01-01T00:00:00.000000Z (UTC+00:00)` In spark, the result is shown as the time in the session timezone(`spark.sql.session.timeZone`). ``` set spark.sql.session.timeZone=Asia/Shanghai; SELECT make_timestamp(2014, 12, 28, 6, 30, 45.887); -- 2014-12-28 06:30:45.887 SELECT make_timestamp(2014, 12, 28, 6, 30, 45.887, 'CET'); -- 2014-12-28 13:30:45.887 ``` In Velox, it should return the timestamp in UTC timezone, so that the result can be correctly converted by Spark for display. The non-ansi behavior returns NULL for invalid inputs. Spark documentation: https://spark.apache.org/docs/latest/api/sql/#make_timestamp Spark implementation: https://github.com/apache/spark/blob/branch-3.3/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/datetimeExpressions.scala#L2512-L2712 Pull Request resolved: facebookincubator#8812 Reviewed By: amitkdutta Differential Revision: D54788353 Pulled By: mbasmanova

mbasmanova · 2024-03-13T06:26:19Z

@marin-ma I missed it during review, but the new function should go into a separate MakeTimestamp.cpp file and the test should go into MakeTimestampTest.cpp. I'll make these changes before landing. You can see a preview in #9060 . I created this PR to get CI signals before landing.

Summary: Add sparksql function `make_timestamp` for non-ansi behavior, which creates a timestamp from year, month, day, hour, min, sec and timezone (optional) fields. The timezone field indicates the timezone of the input timestamp. If not specified, the input timestamp is treated as the time in the session timezone. The output datatype is timestamp type, which internally stores the number of microseconds from the epoch of `1970-01-01T00:00:00.000000Z (UTC+00:00)` In spark, the result is shown as the time in the session timezone(`spark.sql.session.timeZone`). ``` set spark.sql.session.timeZone=Asia/Shanghai; SELECT make_timestamp(2014, 12, 28, 6, 30, 45.887); -- 2014-12-28 06:30:45.887 SELECT make_timestamp(2014, 12, 28, 6, 30, 45.887, 'CET'); -- 2014-12-28 13:30:45.887 ``` In Velox, it should return the timestamp in UTC timezone, so that the result can be correctly converted by Spark for display. The non-ansi behavior returns NULL for invalid inputs. Spark documentation: https://spark.apache.org/docs/latest/api/sql/#make_timestamp Spark implementation: https://github.com/apache/spark/blob/branch-3.3/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/datetimeExpressions.scala#L2512-L2712 Pull Request resolved: facebookincubator#8812 Reviewed By: amitkdutta Differential Revision: D54788353 Pulled By: mbasmanova

marin-ma · 2024-03-13T07:51:07Z

@marin-ma I missed it during review, but the new function should go into a separate MakeTimestamp.cpp file and the test should go into MakeTimestampTest.cpp. I'll make these changes before landing. You can see a preview in #9060 . I created this PR to get CI signals before landing.

@mbasmanova Thanks. Do I need to port these changes to this PR?

mbasmanova · 2024-03-13T07:53:36Z

Do I need to port these changes to this PR?

@marin-ma No. I just let you know so you are not surprised to see these changes when this PR lands.

facebook-github-bot · 2024-03-13T09:33:24Z

@mbasmanova merged this pull request in ee50d7e.

Summary: Add sparksql function `make_timestamp` for non-ansi behavior, which creates a timestamp from year, month, day, hour, min, sec and timezone (optional) fields. The timezone field indicates the timezone of the input timestamp. If not specified, the input timestamp is treated as the time in the session timezone. The output datatype is timestamp type, which internally stores the number of microseconds from the epoch of `1970-01-01T00:00:00.000000Z (UTC+00:00)` In spark, the result is shown as the time in the session timezone(`spark.sql.session.timeZone`). ``` set spark.sql.session.timeZone=Asia/Shanghai; SELECT make_timestamp(2014, 12, 28, 6, 30, 45.887); -- 2014-12-28 06:30:45.887 SELECT make_timestamp(2014, 12, 28, 6, 30, 45.887, 'CET'); -- 2014-12-28 13:30:45.887 ``` In Velox, it should return the timestamp in UTC timezone, so that the result can be correctly converted by Spark for display. The non-ansi behavior returns NULL for invalid inputs. Spark documentation: https://spark.apache.org/docs/latest/api/sql/#make_timestamp Spark implementation: https://github.com/apache/spark/blob/branch-3.3/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/datetimeExpressions.scala#L2512-L2712 Pull Request resolved: facebookincubator#8812 Reviewed By: amitkdutta Differential Revision: D54788353 Pulled By: mbasmanova fbshipit-source-id: bf28991c4373345876459ab4781eecb90ba30519

facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Feb 21, 2024

rui-mo reviewed Feb 22, 2024

View reviewed changes

PHILO-HE reviewed Feb 23, 2024

View reviewed changes

marin-ma changed the title ~~Add make_timestamp sparksql function~~ Add make_timestamp Spark function Feb 23, 2024

rui-mo reviewed Feb 26, 2024

View reviewed changes

marin-ma force-pushed the make-timestamp branch from 30baa47 to 0902503 Compare February 27, 2024 09:49

mbasmanova reviewed Feb 27, 2024

View reviewed changes

marin-ma mentioned this pull request Feb 29, 2024

Return Status for several utility datetime functions #8918

Closed

marin-ma force-pushed the make-timestamp branch from 0902503 to 16886e7 Compare March 6, 2024 15:12

mbasmanova reviewed Mar 8, 2024

View reviewed changes

mbasmanova approved these changes Mar 12, 2024

View reviewed changes

marin-ma force-pushed the make-timestamp branch from 960e419 to 8daf20e Compare March 12, 2024 06:54

marin-ma added 5 commits March 13, 2024 08:59

support make_timestamp sparksql function

64b989a

update doc

54cfd34

fix include

36a945b

address comments

108239a

use fixed scale

16124b3

marin-ma force-pushed the make-timestamp branch 2 times, most recently from 9e0f82d to 16124b3 Compare March 13, 2024 01:52

facebook-github-bot closed this in ee50d7e Mar 13, 2024

facebook-github-bot added the Merged label Mar 13, 2024

		@@ -139,6 +139,35 @@ These functions support TIMESTAMP and DATE input types.

		SELECT quarter('2009-07-30'); -- 3

		.. spark:function:: make_timestamp(year, month, day, hour, min, sec[, timezone]) -> timestamp

Add make_timestamp Spark function #8812

Add make_timestamp Spark function #8812

Conversation

marin-ma commented Feb 21, 2024 • edited Loading

netlify bot commented Feb 21, 2024 • edited Loading

✅ Deploy Preview for meta-velox canceled.

marin-ma commented Feb 21, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marin-ma commented Feb 23, 2024

PHILO-HE left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marin-ma commented Feb 27, 2024

mbasmanova left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rui-mo Mar 12, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marin-ma commented Mar 7, 2024

marin-ma commented Mar 8, 2024

mbasmanova left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

marin-ma commented Mar 11, 2024

facebook-github-bot commented Mar 12, 2024

facebook-github-bot commented Mar 12, 2024

mbasmanova commented Mar 12, 2024

marin-ma commented Mar 13, 2024

facebook-github-bot commented Mar 13, 2024

mbasmanova commented Mar 13, 2024

marin-ma commented Mar 13, 2024 • edited Loading

mbasmanova commented Mar 13, 2024

marin-ma commented Mar 13, 2024

mbasmanova commented Mar 13, 2024

facebook-github-bot commented Mar 13, 2024

marin-ma commented Feb 21, 2024 •

edited

Loading

netlify bot commented Feb 21, 2024 •

edited

Loading

rui-mo Mar 12, 2024 •

edited

Loading

marin-ma commented Mar 13, 2024 •

edited

Loading