Using it

Shell and checks

The shell is how you load a file and run a query. The check scripts are how you know a feature is done.

Commands that start with \ are one line. A SPARQL query may span lines. It runs when a { } group has opened and closed and strings are closed, so a PREFIX line stays pending until the query that uses it is finished. \quit leaves even in the middle of a query. A # starts a comment, except inside a string or an IRI. That exception matters: ub:Student is <http://example.edu/univ#Student>, and treating the # as a comment swallows the rest of the query.

\load data/tiny.ttl
PREFIX ub: <http://example.edu/univ#>
SELECT ?name WHERE {
  ?s a ub:Student ; ub:name ?name .
}

That load reports 53 triples. The rows are Alice, Bob, and Cara.

\quit, \help, and \session run on the main thread. Everything else runs on a session worker. The worker catches a not-implemented plan id, a parse or bind error, a transaction abort, and any other exception, and prints it.

The prompt is printed only when stdin is a terminal.

Commands

Command	What it does now
`\load path`	Loads Turtle, N-Triples, or RDF/XML
a SPARQL query or update	Runs the memory engine
`\explain`	Prints the plan, does not execute
`\explain analyze`	Not built (plan 5.5)
`\stats`	Prints I/O counters, which stay zero until you increment them
`\timing on` or `off`	Stores the knob
`\set backend mem` or `indexed`	`indexed` is not built (plan 3.4) and does not stick
`\set pool_size N`, `lru_k`, `join`, `isolation`	Stored. `\set` with no arguments prints them
`\bpm`, `\page`, `\trace`	Not built (plan 1.6)
`\tree spo`, `pos`, or `osp`	Not built (plan 2.2)
`\begin`, `\commit`, `\abort`, `\txns`, `\locks`	Not built (concurrency)
`\log`, `\checkpoint`, `\crash`	Not built (recovery)
`\session N`	Sends later commands to worker N
`\help`, `\quit`	Work

\set pool_size 8 prints ok. A following \set prints pool_size 8.

Data

data/tiny.ttl is a small university graph, 53 triples. data/pizza.ttl is a small original pizza subset used by the query tests. scripts/fetch-data pizza downloads the full ontology into data/downloads/, which is not in git. scripts/fetch-data lubm needs Java, runs the official generator, and writes University0_0.owl. The 14 LUBM queries live in data/lubm/queries/. Query 7 in the official text is not a triple; the file here is the repaired form. All 14 parse.

Checks

scripts/check key
scripts/check concurrency
scripts/status
scripts/crashtest --seeds 4

scripts/check configures the preset, builds, runs that feature or that slice, and prints pass or fail plus the next id. Concurrency uses TSan. scripts/status prints the feature groups. Progress starts at 0%. A finished comparison is written up from benchmarks and writeups.

Presets are dev (Debug, ASan, UBSan), tsan, and release.

Harnesses

The query harness is a text file of SPARQL plus expected rows in canonical spelling. Comparison is a multiset unless the file says the order matters. ORDER BY queries stay out of the enabled suite until sort exists.

The isolation harness runs a scripted schedule against two sessions and checks blocked, unblocks, abort, contains, equals, and absent. Its self-test does not need a lock manager.

The crash harness forks a child, kills it, and compares the reopened store to a memory store that replayed only acknowledged commits. Its self-test uses fake stores that lie in known ways.