Weird behaviour with attachment and multiple selectors

[yohanboniface]

@springmeyer here is what I think is the smallest use case of the behaviour exposed on IRC today.

project.mml:

{
  "bounds": [
    -72.2653,
    19.5326,
    -71.7586,
    19.8133
  ],
  "center": [
    -72.1198,
    19.6698,
    16
  ],
  "format": "png",
  "interactivity": false,
  "minzoom": 0,
  "maxzoom": 20,
  "srs": "+proj=merc +a=6378137 +b=6378137 +lat_ts=0.0 +lon_0=0.0 +x_0=0.0 +y_0=0 +k=1.0 +units=m +nadgrids=@null +wktext +no_defs +over",
  "Stylesheet": [
    "roads.mss"
  ],
  "Layer": [
    {
      "Datasource": {
        "dbname": "hdm",
        "geometry_field": "way",
        "host": "",
        "id": "roads_high",
        "key_field": "",
        "password": "",
        "port": "5432",
        "project": "hdm",
        "srs": null,
        "table": "( SELECT way, name, highway AS type, CASE WHEN highway='primary' THEN 'bar' ELSE 'baz' END AS foo FROM planet_osm_line WHERE highway IS NOT NULL ORDER BY layer, z_order) AS data",
        "type": "postgis",
        "extent": "-20037508,-19929239,20037508,19929239",
        "user": "ybon"
      },
      "class": "",
      "geometry": "linestring",
      "extent": [
        -179.99999692067183,
        -84.96651228427099,
        179.99999692067183,
        84.96651228427098
      ],
      "id": "roads_high",
      "name": "roads_high",
      "srs": "+proj=merc +a=6378137 +b=6378137 +lat_ts=0.0 +lon_0=0.0 +x_0=0.0 +y_0=0.0 +k=1.0 +units=m +nadgrids=@null +wktext +no_defs +over",
      "srs-name": "900913",
      "advanced": {
        "extent_from_subquery": "true"
      }
    }
  ],
  "scale": 1,
  "metatile": 2,
  "attribution": "Data © OpenStreetMap (and) contributors, CC-BY-SA",
  "description": "",
  "name": "hdm"
}

roads.mss:

#roads_high::outline {
  [type='primary'] {
    line-color: blue;
  }
    [foo='bar'] { line-width: 15; }
}

#roads_high {
  [type='primary'] {
    line-color: red;
  }
    [foo='bar'] { line-width: 10; }
}

Then if you have a psql database with the default osm2pgsql import you will be able to run it.

Here is what I have (only the highway=primary are rendered):

The casing is black, but I expect it to be blue.
If I move the line-width in the [type=primary] selector in any of the main selectors, the problem disappears.

Mapnik version: 2.2.0
Carto version: 0.9.4
Running head of TileMill: mapbox/tilemill@7874cce3d82e70239f44f09a0aab1f49f62bef8f

[yohanboniface]

Note: if I try running with carto 0.9.3, I fall into #283

[yohanboniface]

And as expected, if I install carto from cfc90da (before #247 fix) I have the expected behaviour. So I will run this for now, in order to generate the MBTiles.

[springmeyer]

thanks, nice testcase. I have replicated very different behavior between 6eda91a (the parent of the known breakage of #247 in 4d4abb2) and master. Will add a failing test shortly. Hoping this will help @tmcw fix.

[springmeyer]

okay, stepping through commits and running this new test reveals that this regressed in 4203578. So, this problem is new and only appeared when we attempted to fix #247 in master.

[yohanboniface]

@springmeyer @tmcw just thinking: what about changing the trunk version in something like "0.9.5-dev" ("0.9.5-alpha", or whatever is more semverish)? I've been first trapped here because the working instance and the bugged one were both saying to run the "0.9.4" carto version.

[tmcw]

I'd rather not: git tags and npm releases constitute tagged releases, everything else is a snapshot and you should ignore the version number until it's pushed/tagged.

[springmeyer]

I'm traveling/offline for the remainder of this week, but keen to work with @tmcw early next week to solve this bug.

[springmeyer]

fixed in bbeff81 thanks to @tmcw.

Now we need more tests to ensure we avoid any future regressions, please see: #245