Fix test

[sharkAndshark]

Try to fix #1630

• Use ogrmerge of gdal to ignore features ordering(geozero do no support mvt and ogrmerge works well)
• Do not diff *.pbf
• Add test-latest start-latest to justfile
• fix martin-cp test on latest Postgres/PostGIS

[nyurik]

I am seeing a very strange issue - the martin_cp result differs on the GitHub runners vs my local runs. This is specifically apparent in the cp_flat.mbtiles:

❯ diff failed/tests/output/martin-cp/flat_summary.txt expected/martin-cp/flat_summary.txt 
7,9c7,9
<     1 |         4 |      338B |      705B |      439B | -180,-85,180,85
<     2 |         5 |      123B |      690B |      356B | -90,-67,180,67
<     3 |         8 |       75B |      727B |      245B | -45,-41,180,67
---
>     1 |         2 |      150B |      172B |      161B | -180,-85,0,85
>     2 |         4 |      291B |      690B |      414B | -90,-67,90,67
>     3 |         7 |       75B |      727B |      263B | -45,-41,90,67
13c13
<   all |       127 |       75B |      727B |      197B | -180,-85,180,85
---
>   all |       123 |       75B |      727B |      190B | -180,-85,180,85

P.S. what the above means is that the failed (CI) runs generate fewer tiles than my local installation.

[sharkAndshark]

Yes I noticed and trying to figure out

[nyurik]

how can it be that martin-cp sees fewer tiles in CI? Could it be a postgres issue (some magical SQL returns a different result?)... or something else? This is probably the biggest concern tbh - as it shows that in CI we have different content somehow

[sharkAndshark]

as_mvtgeom filter more than ever may be? Need to dig into

[sharkAndshark]

debugging

[sharkAndshark]

The failed test is running on PostgreSQL16.6.0 / PostGIS 3.5.1.
But you could easily duplicate it on PostgreSQL 17.2.0 / PostGIS 3.5.1 (no docker images whose tag is exactly equal to 16.6.0+3.5.1 so I go to 17-3.5 )

  db-latest:
    # This should match the version of postgres used in the CI workflow
    image: postgis/postgis:17-3.5
    restart: unless-stopped
    ports:
      - '${PGPORT:-5411}:5432'
    command:
      - '-c'
      - 'max_connections=200'
    environment:
      # POSTGRES_* variables are used by the postgis/postgres image
      # PG_* variables are used by psql
      - POSTGRES_DB=db
      - POSTGRES_USER=postgres
      - POSTGRES_PASSWORD=postgres
      - PGDATABASE=db
      - PGUSER=postgres
      - PGPASSWORD=postgres
    volumes:
      - ./tests/fixtures:/fixtures
      - ./tests/fixtures/initdb-dc.sh:/docker-entrypoint-initdb.d/20_martin.sh

# a command added in jsutfile in this PR
just test-latest

So It seems that the new version PG/PostGIS return more than ever.

[sharkAndshark]

Add a temp test here(would be removed before merge). We could see the tile range is same with our local machine.

[nyurik]

I think we need to use the same version of postgres in both CI and in our local development - for consistency. I also think we should (for now) stay at an older version, e.g. 15, just so that we don't analyze too many things at the same time - and rebless the tests and release. Once done, we should figure out why v16+ (?) generates significantly different results - as this might be a major bug in our ST_AsMVT tile generation

[sharkAndshark]

Totally agree.