zeekay
4897baee50
The README states the shape a caller actually gets: 18 packages, 99 source files, 410 passing tests, keys and posting lists in zapdb opened in managed mode with the caller supplying every timestamp. It also states the limit plainly rather than leaving it to be discovered. DQL parses to a syntax tree and stops — no package imports the parser, and the executor that walked that tree belonged to the cluster code. A caller reads by key or by index token and composes UID lists with `algo`; it does not traverse, filter or bind variables. An engine that advertises a query language it cannot execute costs the reader an afternoon. The proto Makefile drops the badger include path along with the import it served.
1.0 KiB
Hanzo Graph
Hanzo Graph forks Dgraph (github.com/dgraph-io/dgraph, Apache License 2.0) at commit 64804bd2 and
cuts it down to what runs inside a single process. It keeps the posting-list engine, the schema and
index layers, the tokenizers, the type system, the DQL parser, the UID set algebra, and the wire
types — 18 packages, 99 source files, 44 test files, 410 passing tests. Nothing here serves or
clusters: no Raft, no gRPC service, no query executor, no integration harness, and the only two
main packages are a build-variable emitter and a codec benchmark. Keys and posting lists live in
github.com/luxfi/zapdb v1.10.6, which the caller opens in managed mode and for which the caller
supplies every start and commit timestamp. DQL parses to a syntax tree and stops there — no package
imports the parser, and the executor that walked that tree belonged to the cluster code — so a
caller reads by key or by index token and composes UID lists with algo instead of traversing,
filtering, or binding variables.