Rework Config file structure... take 2

[nyurik]

It seems config is still fairly convoluted, and may need some cleanup... Another proposal:

server:
  keep-alive: 75
  listen-addresses: '0.0.0.0:3000'
  worker-processes: 8

# for now connections can only support postgres.
# File sources are either auto-published or listed directly in sources 
connections:
  pg1:
    # connection type is detected by the connection field
    connection: postgres://...
    default-srid: ...
    disable-bounds: ...
    pool-size: ...
    maxzoom: 10  # some values can set defaults for all sources that use this connection

# These define how sources will be auto-discovered from connections (PG) or from directories
auto-publish:
  postgres:
    - connection: pg1
      from-schemas:
        - public
        - my_schema
      tables:
        id-format: '{schema}.{table}.{column}'
      functions:
        id-format: '{schema}.{function}'
    - connection: pg2
      # ...
  mbtiles:
    - /dir/path
  pmtiles:
    - /dir/path
    - path: /dir/path
      minzoom: 0
      maxzoom: 10

# These are the "hard-coded" tile sources from different connections
sources:
  # source type is detected by filename extension
  src1: data.mbtiles
  src2:
    # publish pmtiles with extra config
    file: path/to/file.pmtiles
    maxzoom: 10
  src3:
    # publish a postgres table
    # source type is detected by the connection field referencing above connection object
    connection: pg1
    schema: foo
    table: bar
  src4:
    # Publish a postgres function
    connection: pg1
    schema: foo
    function: bar

Additionally, there are some settings to allow override generated tilejson. Any of these values can be added to any of the sources, connections, and auto-publish sections.

minzoom: 0
maxzoom: 10
name: "source name"
attribution: "source attribution"
description: "source description"
legend: "source legend"
bounds: [-10, -20, 30, 40]
center: [10.2, 34.5, 3]
version: "1.42"

[nyurik]

Related topic is how to parse this config with serde enum - it could either be untagged, in which case user may see unexpected errors if the config value cannot be matched perfectly, or we could do it "by hand":

• parse into objects like sources: HashMap<String, Value>
• manually test it, e.g. sources.get("connection") is a string and starts with "postgres://" => treat as postgres.

[nyurik]

I tried hacking on this implementation, and one already apparent annoyance is the auto-publish separation for pg:

connections:
  pg1:
    connection: postgres://...

auto-publish:
  postgres:
    - connection: pg1

The issues with the above:

• multiple auto-publish for one connection
• auto-publish is separate from the connection, which is kinda weird, but is inline with other sources that do not have connections (mbtiles/pmtiles)
• duplicate line connection: with different meaning, and kinda verbose

Possible but non-ideal solutions:

connections:
  pg1:
    connection: postgres://...

auto-publish:
  postgres:
    pg1: ...  # the key here must match the key in connections. This is bad because other auto-publish types are lists

connections:
  pg1:
    connection: postgres://...
    auto-publish: ...  # just like before -- autopublish is part of the connection. This is bad because other types like mbtiles/pmtiles do not have connections