Looking great, thanks for doing this! 🚮 Just had a few small questions

Reviewed 71 of 80 files at r1, 9 of 9 files at r2, all commit messages.
Reviewable status: all files reviewed, 4 unresolved discussions (waiting on @RaduBerinde and @sumeerbhola)

format_major_version.go line 139 at r2 (raw file):

		FormatSSTableValueBlocks
*/
const numDeprecatedVersions = 12

nit: maybe we should keep lines within the const stanza to increment the iota automatically? Could do something like:

const (
    // FormatMostCompatible (deprecated) ...
    _ = iota

    // FormatVersioned (deprecated) ...
   _

open.go line 148 at r2 (raw file):

		//
		// To error in the first case, we set ErrorIfPristine.
		opts.ErrorIfNotPristine = true

I'm forgetting, will this error if there's a CURRENT file?

compaction_iter_test.go line 97 at r2 (raw file):

		}
		if formatVersion < FormatDeleteSizedAndObsolete {
			return "testdata/compaction_iter_set_with_del"

should we keep this one for format major versions earlier than FormatDeleteSizedAndObsolete?

db_test.go line 79 at r2 (raw file):

func TestBasicReads(t *testing.T) {
	t.Skip("TODO")

still intended to be skipped?

TFTR!

Reviewable status: 77 of 81 files reviewed, 4 unresolved discussions (waiting on @jbowens and @sumeerbhola)

format_major_version.go line 139 at r2 (raw file):

Good idea, done.

open.go line 148 at r2 (raw file):

No, it will error if any data operations have been performed.

compaction_iter_test.go line 97 at r2 (raw file):

Good catch! Fixed.

db_test.go line 79 at r2 (raw file):

Done.

Couple of nits only. Thanks for doing this!

Reviewed 11 of 80 files at r1, 4 of 9 files at r2, 2 of 4 files at r3, all commit messages.
Reviewable status: 79 of 81 files reviewed, 6 unresolved discussions (waiting on @jbowens and @RaduBerinde)

format_major_version.go line 139 at r2 (raw file):

can we add a trivial unit test to assert the actual numbers to ensure we don't renumber something by mistake?

open.go line 147 at r3 (raw file):

		//  - we are retrying a failed creation.
		//
		// To error in the first case, we set ErrorIfPristine.

nit: should this say "... set ErrorIfNotPristine"?

open.go line 1022 at r3 (raw file):

	// mem is nil here.
	if !d.opts.ReadOnly && batchesReplayed > 0 {

curious about the addition of batchesReplayed > 0 here -- was this an existing bug?

TFTR!

Reviewable status: 76 of 81 files reviewed, 4 unresolved discussions (waiting on @jbowens and @sumeerbhola)

open.go line 1022 at r3 (raw file):

I don't think it was a bug, just an unnecessary operation. I was hitting some panics in this code path while I was developing the change and adding the check helped. If I remove it now, the tests pass, but it makes sense to keep it.