Thanks for the insight. Indeed, I could not replicate Adrian's error message by substituting his date example in my code. It just gives me a generic 'cannot cast type date to integer' with no mention of a column name. I think I better understand how the context affects the ability to provide certain information in error messages.
I'll attempt to solve my problem by querying pg_attribute for the columns of the table I'm dealing with and then I'll just loop over them all until I get a hit on the value returned in my error message. It won't be pretty, but it's better than browsing thousands of columns in CSVs trying to find these pitfalls.
Glad to hear it's here just in time. I am using 9.4, though, so I wish I could figure out why it's returning NULL when I use it. And the error message string doesn't contain any column name to parse in my output.
As far as I recall, COLUMN_NAME is new in 9.4. If you're trying with an earlier version, you can't get that info other than by parsing the error message string.
Adrian provided an example of trying to place a valid date into a non-date column. The data itself was correct but the place it is being stored to is invalid and can be reported explicitly.
Shawn provided an example of trying to create an integer using an invalid value. Type input errors are not column specific and so the error - which is basically a parse error - does not provide column information.
There is likely some more experimenting here, and maybe room to attached optional contextual markers to make better error messages, but fundamentally these are two different kinds of errors which cannot be generalized over to the extent of saying "errors should provide column name information"...because what would you do for { SELECT 'a'::int }?
