Repository navigation
Run the append aggregates of the catalog on both typed SQL surfaces - #172
Merged
estebanzimanyi merged 1 commit intoOct 11, 2026
Merged
Conversation
appendInstantAgg and appendSequenceAgg state no combine: each is the constructor of a temporal value fed one row at a time, appendInstantAgg building a sequence from instants as tsequence_make builds one from an array, and appendSequenceAgg a sequence set from sequences as tsequenceset_make does. Their transitions accept a value only after those already folded in time, as the constructors accept an array only in that order, so time is the one order that answers. Neither Spark nor Flink hands the rows of a group in an order, and each joins partial aggregates, so for an aggregate stating no combine both surfaces gather the rows of a group, a partial joining another by taking its rows, and fold them at the end ordered by the start timestamp of their first argument, which temporal_start_timestamptz answers: the answer PostgreSQL gives over the same rows ordered by time, and the error it raises over rows no order accepts. The transition of an aggregate takes the further arguments as the signature of the transition the aggregate names reads them, its boundArgs and sqlArgParams, so appendInstantAgg(tint) passes the interpolation of its type, a distance of -1.0 and no interval, and appendInstantAgg(tbool, text, interval) a distance of -1.0. A text and a float are arguments both surfaces encode, beside the interval of the window aggregates, and Spark resolves an aggregate as it resolves a function, an exact fit first and then one widening a number, so the distance 1.5, a decimal, reaches the float. A final taking its state const keeps it, as temporal_append_finalfn answers a compact copy, and the state is released after its final; every other final consumes its state. The Flink MeosValue is Serializable, as the Spark one is, so the accumulator of an aggregate gathering rows writes them. Witness: with the generator of JMEOS main ba1e61c neither surface registers appendInstantAgg or appendSequenceAgg, the 77 overloads of their catalog names left out as stating no combine. Measured over MobilityDB 3a8e6dba09 and the catalog of MEOS-API 4baf11705c: each surface registers 22 aggregates of 248 overloads, against 20 of 171 with the generator of main, the 77 of appendInstantAgg and appendSequenceAgg among them, and leaves out no aggregate as stating no combine. Every class of the other 20 aggregates differs only by the two arguments the roles take, false and null. Why: a trajectory built from its instants or its pieces answers on Spark and Flink as it does in PostgreSQL, whatever order the engine hands the rows in.
estebanzimanyi
added a commit
to estebanzimanyi/JMEOS
that referenced
this pull request
Oct 11, 2026
The tests of jmeos-core run concurrently, classes and methods alike (junit-platform.properties), and the binding holds only the latest error handler MEOS receives, in the static field the wrapper of meos_initialize_error_handler fills. ErrorHandlerRegistrationTest registers a handler it keeps no reference to and checks that a collection leaves it reachable, so a handler registered meanwhile, by the other test of the class or by a class such as STBoxTest, replaces the one it checks and leaves that one to the collector. The class is @isolated, as MeosInitializeHandlerTest is, so nothing registers a handler while it runs. Witness: the fork run 38103656194 of JMEOS MobilityDB#172 at 89fd5ee fails ErrorHandlerRegistrationTest.unreferencedHandlerStaysReachable with "the registered handler was collected while MEOS holds it", while the upstream run 38103672322 of the same commit passes it. Measured over MobilityDB 3a8e6dba09 and the catalog of MEOS-API 4baf11705c: three runs of the jmeos-core suite pass 1804 tests each, the two of ErrorHandlerRegistrationTest among them. Why: a test of the binding's handler answers for the binding, not for the order in which the concurrent tests register theirs.
estebanzimanyi
added a commit
that referenced
this pull request
Oct 11, 2026
The tests of jmeos-core run concurrently, classes and methods alike (junit-platform.properties), and the binding holds only the latest error handler MEOS receives, in the static field the wrapper of meos_initialize_error_handler fills. ErrorHandlerRegistrationTest registers a handler it keeps no reference to and checks that a collection leaves it reachable, so a handler registered meanwhile, by the other test of the class or by a class such as STBoxTest, replaces the one it checks and leaves that one to the collector. The class is @isolated, as MeosInitializeHandlerTest is, so nothing registers a handler while it runs. Witness: the fork run 38103656194 of JMEOS #172 at 89fd5ee fails ErrorHandlerRegistrationTest.unreferencedHandlerStaysReachable with "the registered handler was collected while MEOS holds it", while the upstream run 38103672322 of the same commit passes it. Measured over MobilityDB 3a8e6dba09 and the catalog of MEOS-API 4baf11705c: three runs of the jmeos-core suite pass 1804 tests each, the two of ErrorHandlerRegistrationTest among them. Why: a test of the binding's handler answers for the binding, not for the order in which the concurrent tests register theirs.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
appendInstantAgg and appendSequenceAgg state no combine: each is the constructor of a temporal
value fed one row at a time, appendInstantAgg building a sequence from instants as
tsequence_make builds one from an array, and appendSequenceAgg a sequence set from sequences as
tsequenceset_make does. Their transitions accept a value only after those already folded in
time, as the constructors accept an array only in that order, so time is the one order that
answers. Neither Spark nor Flink hands the rows of a group in an order, and each joins partial
aggregates, so for an aggregate stating no combine both surfaces gather the rows of a group, a
partial joining another by taking its rows, and fold them at the end ordered by the start
timestamp of their first argument, which temporal_start_timestamptz answers: the answer
PostgreSQL gives over the same rows ordered by time, and the error it raises over rows no order
accepts.
The transition of an aggregate takes the further arguments as the signature of the transition
the aggregate names reads them, its boundArgs and sqlArgParams, so appendInstantAgg(tint) passes
the interpolation of its type, a distance of -1.0 and no interval, and appendInstantAgg(tbool,
text, interval) a distance of -1.0. A text and a float are arguments both surfaces encode, beside
the interval of the window aggregates, and Spark resolves an aggregate as it resolves a function,
an exact fit first and then one widening a number, so the distance 1.5, a decimal, reaches the
float. A final taking its state const keeps it, as temporal_append_finalfn answers a compact
copy, and the state is released after its final; every other final consumes its state. The Flink
MeosValue is Serializable, as the Spark one is, so the accumulator of an aggregate gathering rows
writes them.
Witness: with the generator of JMEOS main ba1e61c neither surface registers appendInstantAgg or
appendSequenceAgg, the 77 overloads of their catalog names left out as stating no combine.
Measured over MobilityDB 3a8e6dba09 and the catalog of MEOS-API 4baf11705c: each surface
registers 22 aggregates of 248 overloads, against 20 of 171 with the generator of main, the 77
of appendInstantAgg and appendSequenceAgg among them, and leaves out no aggregate as stating no
combine. Every class of the other 20 aggregates differs only by the two arguments the roles take,
false and null.
Why: a trajectory built from its instants or its pieces answers on Spark and Flink as it does in
PostgreSQL, whatever order the engine hands the rows in.