varchar Type segmentby Columns Not Used for Filtering Prior to Decompression

[tylerfontaine]

Relevant system information:

• OS: docker
• PostgreSQL version (output of postgres --version): 13.3
• TimescaleDB version (output of \dx in psql): 2.3.0
• Installation method: docker

Describe the bug
segmentby columns of type varchar are not evaluated for exclusion prior to decompression, while segmentby columns of type text are, regardless of casting at query time.

To Reproduce

create table testcasting (a timestamp, b varchar, c text);

select create_hypertable('testcasting', 'a');

insert into testcasting VALUES
(now(), 'a', 'b'),
(now() - interval '1 month', 'c', 'd');

alter table testcasting set (timescaledb.compress, timescaledb.compress_segmentby='b', timescaledb.compress_orderby = 'a');



select show_chunks('testcasting');

select compress_chunk('<some chunk from the above query>');

explain (analyze,buffers) select * from testcasting
where b = 'a'::varchar;

Expected behavior
In the explain, analyze, I would expect something like this, which shows the filter running before compression:

       ->  Custom Scan (DecompressChunk) on _hyper_31_9375_chunk  (cost=1.01..1.01 rows=1000 width=12) (actual time=0.008..0.009 rows=0 loops=1)
        Buffers: shared hit=1
        ->  Seq Scan on compress_hyper_33_9377_chunk  (cost=0.00..1.01 rows=1 width=116) (actual time=0.007..0.007 rows=0 loops=1)
              Filter: (b = 'a'::text)
              Rows Removed by Filter: 1
              Buffers: shared hit=1

Actual behavior
What I see instead is this, showing filtering only happens after decompression:

  ->  Custom Scan (DecompressChunk) on _hyper_31_9375_chunk  (cost=1.01..1.01 rows=1000 width=12) (actual time=0.027..0.028 rows=0 loops=1)
        Filter: ((b)::text = 'a'::text)
        Rows Removed by Filter: 1
        Buffers: shared hit=1
        ->  Seq Scan on compress_hyper_34_9378_chunk  (cost=0.00..1.01 rows=1 width=116) (actual time=0.007..0.008 rows=1 loops=1)
              Buffers: shared hit=1