encode-db writes a Version-1 header the library itself rejects (invalid version)

[arnoldwolfe]

Version: ltx CLI v0.5.2 release binary (darwin-arm64), library at the same tag.

ltx encode-db sets Version: 1 in the header it passes to the encoder (cmd/ltx/encode_db.go), but Header.Validate requires Version == 3, so the command always fails:

$ sqlite3 t.db "PRAGMA page_size=4096; CREATE TABLE x(a); INSERT INTO x VALUES (1);"
$ ltx encode-db -o out.ltx t.db
encode ltx header: invalid version

Expected: encode-db writes a version-3 header (the only version the library accepts).
Actual: the command cannot produce any output the library itself would read.

Found while validating an independent implementation of the format against the reference CLI. Happy to provide more detail.

[corylanou]

Confirmed and reproduced. Some extra context for whoever picks this up.

Version: 1 at cmd/ltx/encode_db.go:80 has been there since the command was added in 5fb227c (July 2023). It only became fatal at 2e6df57 (Sept 2025), which bumped the format to version 3. That commit is in v0.5.0, v0.5.1 and v0.5.2, so encode-db has been completely non-functional in every release for the last eleven months.

One correction to the title: no version-1 header is ever actually written. Header.Validate runs before the write (encoder.go:159), and Version is not serialized at all — UnmarshalBinary derives it from the magic and sets it to the current constant unconditionally (ltx.go:323). The magic LTX1 is the only on-disk version discriminator. So the field is purely in-memory and the fix is Version: ltx.Version, not a hardcoded 3.

Scope note: #87 covers two further bugs in the same function (missing O_TRUNC on the output, and no rejection of input==output) plus a dead flags variable. Both are currently masked because the command never reaches its write path. They should be fixed in the same PR as this issue, since #81 is what makes that path reachable.

Worth adding the first test under cmd/ltx while here — there are currently zero test files for seven command files, which is the direct reason this shipped broken for three releases. An encode-db then Decoder.Verify round-trip would have caught it.