st_generatepoints declared IMMUTABLE

[pnorman]

On my 2.3 system st_generatepoints is declared as IMMUTABLE. This is only valid for functions that always generate the same output given the same parameters. Random functions should be VOLATILE because they can give different results each time they are called.

On a closely related matter, it is also marked as PARALLEL SAFE, which might not be true - ​https://www.postgresql.org/docs/11/parallel-safety.html says that random is PARALLEL RESTRICTED because it accesses "miscellaneous backend-local state which the system cannot synchronize across workers" and the same might be true for st_generatepoints

This originates from ​https://lists.osgeo.org/pipermail/postgis-devel/attachments/20190115/292e0a64/attachment.html

[komzpa]

​https://postgis.net/docs/ST_GeneratePoints.html says "ST_GeneratePoints generates pseudo-random points until the requested number are found within the input area". There is no guarantee that sequence will be the same or not the same each time. IMMUTABLE only affects caching in Postgres as for now, and per current documentation it's okay to cache ST_GeneratePoints output for the same polygon. Will a note about this behavior be sufficient?

[mwtoews]

Another reason for the behavior in the email is because the RNG is initialized with srand(time(NULL));, which means if you re-run the query multiple times within one clock second, you will get the same results. (Try it, but you need to be quick!)

I'll try to address this ticket with #4304

[mwtoews]

See ​https://github.com/postgis/postgis/pull/365 which attempts to implement these changes.

[komzpa]

In 17248:

Add seed parameter for ST_GeneratePoints.

Patch by Mike Taves.

Closes #4299
Closes #4304
Closes ​https://github.com/postgis/postgis/pull/365