modernize, cleanup, prevent unnessecary valCopies.

[the-hotmann]

@jackc Could you please rerun all tests and benchmarks to confirm there are no flaky results?
I’ve applied a lot of optimizations, so the package should now have significantly fewer allocations.

Additionally, I included a number of other improvements.

I skipped many that would degrade DX and have not touched the structure of the structs, even though I think there is a lot to improve:

./pgx-new/internal/sanitize/sanitize.go:204:15: 24 bytes saved: struct with 56 pointer bytes could be 32
./pgx-new/pgconn/internal/bgreader/bgreader.go:18:15: 8 bytes saved: struct with 40 pointer bytes could be 32
./pgx-new/pgproto3/backend.go:11:14: 32 bytes saved: struct with 448 pointer bytes could be 416
./pgx-new/pgproto3/close.go:9:12: 8 bytes saved: struct with 16 pointer bytes could be 8
./pgx-new/pgproto3/copy_both_response.go:13:23: 8 bytes saved: struct with 16 pointer bytes could be 8
./pgx-new/pgproto3/copy_in_response.go:13:21: 8 bytes saved: struct with 16 pointer bytes could be 8
./pgx-new/pgproto3/copy_out_response.go:13:22: 8 bytes saved: struct with 16 pointer bytes could be 8
./pgx-new/pgproto3/describe.go:9:15: 8 bytes saved: struct with 16 pointer bytes could be 8
./pgx-new/pgproto3/error_response.go:9:20: 24 bytes saved: struct with 264 pointer bytes could be 240
./pgx-new/pgproto3/error_response.go:224:22: 24 bytes saved: struct with 280 pointer bytes could be 256
./pgx-new/pgproto3/error_response.go:277:10: 24 bytes saved: struct with 280 pointer bytes could be 256
./pgx-new/pgproto3/frontend.go:12:15: 8 bytes saved: struct of size 1056 could be 1048
./pgx-new/pgproto3/function_call.go:11:19: 8 bytes saved: struct of size 64 could be 56
./pgx-new/pgproto3/function_call_response.go:74:22: 8 bytes saved: struct with 24 pointer bytes could be 16
./pgx-new/pgproto3/gss_enc_request.go:41:22: 16 bytes saved: struct with 32 pointer bytes could be 16
./pgx-new/pgproto3/notification_response.go:11:27: 8 bytes saved: struct with 32 pointer bytes could be 24
./pgx-new/pgproto3/notification_response.go:60:22: 8 bytes saved: struct with 48 pointer bytes could be 40
./pgx-new/pgproto3/ssl_request.go:41:22: 16 bytes saved: struct with 32 pointer bytes could be 16
./pgx-new/pgproto3/startup_message.go:15:21: 8 bytes saved: struct with 16 pointer bytes could be 8
./pgx-new/pgproto3/startup_message.go:85:22: 16 bytes saved: struct with 32 pointer bytes could be 16
./pgx-new/pgproto3/trace.go:15:13: 16 bytes saved: struct with 40 pointer bytes could be 24
./pgx-new/pgconn/auth_scram.go:109:18: 8 bytes saved: struct with 200 pointer bytes could be 192
./pgx-new/pgconn/config.go:34:13: 8 bytes saved: struct of size 256 could be 248
./pgx-new/pgconn/config.go:124:21: 16 bytes saved: struct with 32 pointer bytes could be 16
./pgx-new/pgconn/config.go:131:23: 8 bytes saved: struct with 56 pointer bytes could be 48
./pgx-new/pgconn/errors.go:32:14: 16 bytes saved: struct with 248 pointer bytes could be 232
./pgx-new/pgconn/errors.go:82:26: 8 bytes saved: struct with 48 pointer bytes could be 40
./pgx-new/pgconn/errors.go:109:23: 8 bytes saved: struct with 48 pointer bytes could be 40
./pgx-new/pgconn/errors.go:157:18: 8 bytes saved: struct with 32 pointer bytes could be 24
./pgx-new/pgconn/pgconn.go:41:19: 8 bytes saved: struct with 32 pointer bytes could be 24
./pgx-new/pgconn/pgconn.go:76:13: 8 bytes saved: struct of size 1112 could be 1104
./pgx-new/pgconn/pgconn.go:1449:24: 8 bytes saved: struct with 56 pointer bytes could be 48
./pgx-new/pgconn/pgconn.go:1551:19: 24 bytes saved: struct with 128 pointer bytes could be 104
./pgx-new/pgconn/pgconn.go:1566:13: 16 bytes saved: struct with 80 pointer bytes could be 64
./pgx-new/pgconn/pgconn.go:1741:12: 16 bytes saved: struct with 40 pointer bytes could be 24
./pgx-new/pgconn/pgconn.go:1959:19: 16 bytes saved: struct with 64 pointer bytes could be 48
./pgx-new/pgconn/pgconn.go:2043:15: 8 bytes saved: struct with 112 pointer bytes could be 104
./pgx-new/pgconn/pgconn.go:2076:20: 16 bytes saved: struct with 64 pointer bytes could be 48
./pgx-new/pgconn/pgconn.go:2480:41: 16 bytes saved: struct with 40 pointer bytes could be 24
./pgx-new/internal/stmtcache/lru_cache.go:10:15: 24 bytes saved: struct with 56 pointer bytes could be 32
./pgx-new/pgtype/array.go:19:18: 8 bytes saved: struct with 16 pointer bytes could be 8
./pgx-new/pgtype/array_codec.go:327:25: 8 bytes saved: struct with 40 pointer bytes could be 32
./pgx-new/pgtype/composite.go:31:26: 8 bytes saved: struct with 24 pointer bytes could be 16
./pgx-new/pgtype/composite.go:281:29: 8 bytes saved: struct of size 96 could be 88
./pgx-new/pgtype/composite.go:364:27: 24 bytes saved: struct with 80 pointer bytes could be 56
./pgx-new/pgtype/composite.go:460:29: 32 bytes saved: struct with 64 pointer bytes could be 32
./pgx-new/pgtype/composite.go:517:27: 32 bytes saved: struct with 64 pointer bytes could be 32
./pgx-new/pgtype/multirange.go:276:30: 8 bytes saved: struct with 40 pointer bytes could be 32
./pgx-new/batch.go:12:18: 16 bytes saved: struct with 56 pointer bytes could be 40
./pgx-new/conn.go:119:17: 8 bytes saved: struct with 32 pointer bytes could be 24
./pgx-new/copy_from.go:43:20: 16 bytes saved: struct with 40 pointer bytes could be 24
./pgx-new/copy_from.go:74:19: 16 bytes saved: struct with 48 pointer bytes could be 32
./pgx-new/copy_from.go:109:15: 16 bytes saved: struct with 80 pointer bytes could be 64
./pgx-new/derived_types.go:151:22: 24 bytes saved: struct with 104 pointer bytes could be 80
./pgx-new/named_args.go:39:15: 40 bytes saved: struct with 80 pointer bytes could be 40
./pgx-new/rows.go:122:15: 8 bytes saved: struct with 240 pointer bytes could be 232
./pgx-new/rows.go:343:19: 16 bytes saved: struct with 40 pointer bytes could be 24
./pgx-new/rows.go:707:24: 8 bytes saved: struct with 32 pointer bytes could be 24
./pgx-new/tracer.go:23:24: 8 bytes saved: struct with 32 pointer bytes could be 24
./pgx-new/tracer.go:42:26: 16 bytes saved: struct with 72 pointer bytes could be 56
./pgx-new/tracer.go:67:27: 8 bytes saved: struct with 32 pointer bytes could be 24
./pgx-new/tracer.go:86:26: 8 bytes saved: struct with 24 pointer bytes could be 16
./pgx-new/tx.go:152:11: 16 bytes saved: struct with 32 pointer bytes could be 16
./pgx-new/pgxpool/pool.go:27:19: 16 bytes saved: struct with 104 pointer bytes could be 88
./pgx-new/pgxpool/pool.go:78:11: 8 bytes saved: struct of size 224 could be 216
./pgx-new/pgxpool/pool.go:122:13: 8 bytes saved: struct of size 128 could be 120
./pgx-new/tracelog/tracelog.go:137:15: 8 bytes saved: struct with 32 pointer bytes could be 24
./pgx-new/stdlib/sql.go:256:16: 32 bytes saved: struct with 360 pointer bytes could be 328
./pgx-new/stdlib/sql.go:330:13: 8 bytes saved: struct with 16 pointer bytes could be 8
./pgx-new/stdlib/sql.go:417:11: 32 bytes saved: struct with 384 pointer bytes could be 352
./pgx-new/stdlib/sql.go:635:11: 8 bytes saved: struct with 64 pointer bytes could be 56
exit status 1

Anyway feel free to criticize :)

[the-hotmann]

I also fixed all shadow-imports like "sql" (named it sqlVar - sql variable).

Here: I would like to ask for your attention:

https://github.com/jackc/pgx/blob/master/pgtype/int.go#L1173
https://github.com/jackc/pgx/blob/master/pgtype/int.go#L1176

As for me this seems to be a bug:

	if n.Int64 < math.MinInt64 {
		return fmt.Errorf("%d is less than minimum value for Int8", n.Int64)
	}
	if n.Int64 > math.MaxInt64 {
		return fmt.Errorf("%d is greater than maximum value for Int8", n.Int64)
	}

It checks against math.MinInt64 and math.MaxInt64 which does not make sense, as n.Int64 can never be bigger than math.MaxInt64 and the error text hints it should check for math.MaxInt8.
I have fixed this in my PR - not knowing (but assuming) what actually is right.

[the-hotmann]

@jackc if you like I can break this PR down into multiple PRs based on the optimizations:

• Shadow-Import
• modernization
• val-copy optimization
• using pointers on methods

...

if that would help

[jackc]

This would definitely need to be broken up into multiple, maybe even many, separate PR's. I did look at it but +1,780 −1,638 is a lot to review at once.

However, I did notice several types of changes that I'm not optimistic about:

---

Changing loops to use indexes instead of the value. e.g.

for _, bi := range b.QueuedQueries {
	sc.Invalidate(bi.SQL)

to

for i := range b.QueuedQueries {
	sc.Invalidate(b.QueuedQueries[i].SQL)

I find the former easier to read.

---

Changing benchmarks to use b.N instead of b.Loop(). b.Loop() automatically resets the benchmark timer. So this change makes the benchmark less accurate. I believe b.Loop() is also idiomatic now.

---

Renaming variables like sql to sqlVar to avoid shadowing. If there's no ambiguity in the function, I don't see the harm in shadowing an import. It reads better.

---

Changing the function arguments or return values from values to pointers is a breaking change. We can't do that. e.g. GetConnector and OpenDB.