Skip to content

vippsas/go-querysql

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

40 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

querysql - Syntax candy for querying SQL from Go

The querysql package provides several layers; all of them provide safety against common mistakes and avoids a lot of iterator boilerplate. Also included is a convenient logging mechanism.

All of the methods allows conveniently fetching results into your own structs, using reflection.

  • To execute a single select statement, use:
    • querysql.Slice returns a slice
    • querysql.Single validates that the result is exactly one row, and returns that single row
    • querysql.Iter calls a callback function for each row
  • To execute multiple select statements in the same database roundtrip, querysql.Query2, querysql.Query3, ... is available
  • querysql.New offers a lower-level API with more options, used to build the primitives above

Below the variables ctx and db are used for context.Context and *sql.DB.

Single select statement

The most common mode is to process a single select statement per query:

// Read a single integer; errors if not exactly 1 row
n := querysql.MustSingle[int](ctx, db, `select 1`)
n, err := querysql.Single[int](ctx, db, `select 1`)

// Read a slice of integers; can have any number of elements
slice := querysql.MustSlice[int](ctx, db, `select @p1`, 3)
slice, err := querysql.Slice[int](ctx, db, `select @p1`, 3)

// Support for reading into structs, or slices of structs
type row {
	X int
	Y string
}
singleRow, err := querysql.QuerySingle[row](ctx, db, `select 1 as X, "hello" as Y`)
sliceOfRows, err := querysql.QuerySlice[row](ctx, db, `select 1 as X, "hello" as Y`)

// Avoid allocating slices by passing a callback 
err := querysql.QueryIter(ctx, db, func(row int) error {
	fmt.Println("Row %d", row)
}, `select 1 union all select 2`)

This mode still supports logging and will return ErrNotDone if there are several non-logging select statements.

Multiple select statements

SQL supports doing several selects on the same round-trip, which can be very useful to get information from several tables in the same query. In this case a somewhat special syntax is used to pass the specifications for how to read each type:

qry := `
    select 1
    union all select 2;

    select 1 as X, "one" as Y 
` 
singleInteger, sliceOfStruct, err := querysql.Query2(
	querysql.SingleOf[int], 
	querysql.SliceOf[MyStruct],
	ctx, db, qry, arg1, arg2)

We have defined Query2, Query3 and Query4 for this use up to 4 select statements.

If you prefer, you can instead scan into pointers; this also allows using a single function for any number of results or dynamic number of results:

var singleInteger int
var sliceOfStruct []MyStruct
singleInteger, sliceOfStruct, err := querysql.Query(
	[]querysql.Target{
		querysql.SingleInto(&singleInteger),
		querysql.SliceInto(&sliceOfStruct),
	},
	ctx, db, qry, arg1, arg2)

Logging from SQL

When writing longer multi-statement SQL queries the lack of debugging between statements can be a real problem. A work-around is provided in this library. Any target-less select statements where the name of the first column is _log will be re-directed to a logger (if one is configured; and otherwise the data will be ignored). Example:

qry := `
    declare @a = 'world';

    select A='one';

    -- logging
    select _log='info', hello=@a;
    
    select B='two';

    -- log one entry per row, at a non-standard level
    select _log='info', hello=@a from SomeTable;

` 

// configure a logger on ctx
ctx := querysql.WithLogger(context.Background(), LogrusMSSQLLogger(logger, logrus.InfoLevel))

// do the query like normal; the middle select will be directed to the logger
firstResult, secondResult, err := querysql.Query2(ctx, ...)

The LogrusMSSQLLogger above is, as given by the name, specific to one combination of tools, and you may need to write your own implementation of RowsLogger based on the one provided in this library. The *sql.Rows is passed straight through to the RowsLogger, but by convention the first column in the result will be the special column _log, which may either contain a log-level (info, debug, warning, error).

Advanced use

For more advanced usecase you may use querysql.New. In the following example we first do a select to get a "command" back, then scan into different types depending on the value of the first select:

rs := querysql.New(ctx, dbi, `
    select 'ints';

    select 1 union all select 2;
`)
// rs is automatically closed after processing the last select statement,
// but it is still a good idea to defer a Close in case you do not process
// all the select statements
defer rs.Close()

mode, err := querysql.NextResult(rs, querysql.SingleOf[string])
if mode == "ints" {
	data, err := querysql.NextResult(rs, querysql.SliceOf[int])
	...
} else if mode == "structs" {
	type SomeStruct struct {
		X int
		Y string
	}
	data, err := querysql.Next(rs, querysql.Call(func (row MyStruct)) {
		fmt.Println(row)
		return nil
    })
}

You may also process a dynamic number of results; rs.Done() will be be true when there are no more result sets available. At this point, rs has also been automatically closed; although it is still a good idea to defer rs.Close() in case you do not reach the point where rs.Done() == true.

Future plans

  • Allow querying directly into map types, using the first columns as the key
  • Automatically deserialize XML or JSON results from SQL to structs