make it less pg dependant

[kyleconroy]

make it less pg dependant

[eullerpereira94]

With an update to Go 1.13, the ties to pg will be even smaller, but right now, pq.Array is too convenient so substitute.
But since this is a code generation tool, I think that creating array to slice scanner types is reasonable task.But I can't think of the right approach to do it.
One way would be identify every instance of where arrays are used in the creation scripts and generate package-wide scanner types to be used in every piece of code that an array would be needed. The good side is that the code would be clean and elegant. The downside is that I think it would greatly increases the complexity of the tool.
Another would be generate the scanners on demand. Whenever a array would appear in the script, a scanner type with its methods would be prepended to code that is currently generating. The good side is that I think it wouldn't add too much complexity to the tool. But the downside is that the code would be ugly (but this is a personal opinion).

[mh-cbon]

I did not realize pq.Array was doing so much... this needs some deeper inspection

two examples of differences i can think of
https://stackoverflow.com/a/45352902/4466350
https://stackoverflow.com/a/34627688/4466350

[eullerpereira94]

The solution for arrays indeed works, but it is for a specific case that was already covered by a previous issue (refer to #77). I think a more general and idiomatic approach is to make full use of sql.Scanner and the driver.Valuer interfaces. I think this piece of code could be something that would be useful:

package models

import (
	"database/sql"
	"database/sql/driver"
	"encoding/csv"
	"errors"
	"fmt"
	"regexp"
	"strings"
)

type StringSlice []string

// StringSlice is a slice of strings.
type StringSlice []string

// quoteEscapeRegex is the regex to match escaped characters in a string.
var quoteEscapeRegex = regexp.MustCompile(`([^\\]([\\]{2})*)\\"`)

// Scan satisfies the sql.Scanner interface for StringSlice.
func (ss *StringSlice) Scan(src interface{}) error {
	buf, ok := src.([]byte)
	if !ok {
		return errors.New("invalid StringSlice")
	}

	// change quote escapes for csv parser
	str := quoteEscapeRegex.ReplaceAllString(string(buf), `$1""`)
	str = strings.Replace(str, `\\`, `\`, -1)

	// remove braces
	str = str[1 : len(str)-1]

	// bail if only one
	if len(str) == 0 {
		*ss = StringSlice([]string{})
		return nil
	}

	// parse with csv reader
	cr := csv.NewReader(strings.NewReader(str))
	slice, err := cr.Read()
	if err != nil {
		fmt.Printf("exiting!: %v\n", err)
		return err
	}

	*ss = StringSlice(slice)

	return nil
}

// Value satisfies the driver.Valuer interface for StringSlice.
func (ss StringSlice) Value() (driver.Value, error) {
	v := make([]string, len(ss))
	for i, s := range ss {
		v[i] = `"` + strings.Replace(strings.Replace(s, `\`, `\\\`, -1), `"`, `\"`, -1) + `"`
	}
	return "{" + strings.Join(v, ",") + "}", nil
}

Note: this is a straight copy of some of the code generated by the xo package. But this code works and can be refactored for any of the basic types in Go. It can be used either as field in the generated structs and as a parameter in the db.Query* methods.

If we include scanner types for the ints, the floats, string, bool, Time and interface{} in the generated db.go or models.go file, and use them to substitute each instance of an Postgres array, then the dependence to the pq package would be eliminated in Go 1.13 onwards. Any version below that would be impossible to not have any dependence to the pq package.

[kyleconroy]

@mh-cohen I understand the appeal of having sqlc work with multiple databases. However, the current implementation is built on the top of the PostgreSQL query parser (https://github.com/lfittl/pg_query_go). If there is a Go package containing the SQL parser for your database of choice, I can add it to sqlc.