npgsql: Resultset with many columns in sequential mode triggers exception

The issue

When executing select * from nfl_game_team_stats ts where (("ts"."team_id" = 11939) OR ("ts"."team_id" = 13994)) AND ("ts"."season_id" = 2017); Npgsql seems to throw a DataException which rolls up to Dapper. This previously/currently works in 3.2.7.

This is selecting from a view which is defined as:

CREATE OR REPLACE VIEW nfl_game_team_stats AS SELECT ps.*
    FROM (nfl_game g
     JOIN nfl_game_team_play_stats ps ON (((ps.game_id = g.game_id) AND (ps.play_id = g.last_play_id))));

In our logs, with Npgsql logging set to trace, we’re seeing the following when executing the above statement. Ultimately Dapper throws an exception as well.

[2018-06-20 11:58:40.211] [NLogLogger.Log] [8] [Debug]: Executing statement(s):
	select * from nfl_game_team_stats ts where ((("ts"."team_id" = $1) OR ("ts"."team_id" = $2)) AND ("ts"."season_id" = $3)) 
Exception thrown: 'System.Data.DataException' in System.Private.CoreLib.dll
[2018-06-20 11:58:40.288] [NLogLogger.Log] [7] [Error]: Breaking connector 
[2018-06-20 11:58:40.290] [NLogLogger.Log] [7] [Trace]: Cleaning up connector 
[2018-06-20 11:58:40.291] [NLogLogger.Log] [7] [Trace]: Closing connection... 
[2018-06-20 11:58:40.291] [NLogLogger.Log] [7] [Trace]: Closing connector 
[2018-06-20 11:58:40.291] [NLogLogger.Log] [7] [Debug]: Connection closed 
[2018-06-20 11:58:40.568] [StatMilk.Lib.UniversalDataService.UdsConnection+<GetListAsync>d__20`1.MoveNext] [7] [Error]: Method [QueryGameTeamStatsAsync] Error [Error parsing column 1 (game_id=562971428257794 - Int64)] SQL = [select * from nfl_game_team_stats ts where ((("ts"."team_id" = @TeamId_0) OR ("ts"."team_id" = @TeamId_1)) AND ("ts"."season_id" = @SeasonId_2))] Params = @TeamId_0=11939, @TeamId_1=13994, @SeasonId_2=2017] 

When running the query above manually via command line, 46 rows are returned (none of them having a game_id of that).

What’s odd is that our sports have views defined exactly the same and they are not throwing this exception. We have a store procedure generate the view so there is zero difference in syntax between the generation of the offending view. All columns on other sports have fields that match the same type on the joins: bigint on game_id, play_id, and last_play_id.

The only distinction I am able to tell between the different sports/tables is that this specific table has 234 columns while the next largest has 141. Not sure if it’s a data reader issue with the column layout or some change in the mapping which seems fairly substantial in 4.0.0.

Further technical details

Npgsql version: 4.0.0 (also tried with 4.1.0-ci.1065+sha.bb4a1ef4d) PostgreSQL version: 9.6.9 Operating system: OSX High Sierra (also Windows Server 2012 in AWS) Dapper: 1.50.5