Address open issues on GitHub


Functionality and API

  • Implement spatial and temporal types.

  • Add timers to Neo4j::Driver::ResultSummary (see Neo4j::Bolt).

  • croak() error objects (e. g. Exception::Class) instead of strings. It seems there are about four types that would need to be distinguished: illegal usage errors, internal driver errors, Network errors, and Neo4j server errors. See also #7.

Experimental features

Tests, code quality, documentation

  • Test roundtrip of special numeric values (very large integers, -0.0, ±Inf, ±NaN).

  • Convert tests in xt/compatibility to use Module::Load, so that testing/dev dependencies are fewer.

  • Improve test coverage:

    • Many yet uncovered code paths are obviously fine, but difficult or impossible to cover. In some of these cases, it may be possible to refactor the code, such as by banking on autovivification (i.e. don't defend against undefined $a in expressions like $a->{b}; see "Using References" in perlref).

    • The deep_bless subs contain a lot of assertions and other checks that are not normally necessary. Since this logic seems to work fine, it should be simplified. (This may also make it easier to unroll the recursion later.)

    • The current policy of not documenting deprecated methods is informed by the principle to "design interfaces that are: consistent; easy to use correctly; hard to use incorrectly". Perhaps simply listing the deprecated method names with short note like "deprecated in 0.13" would be an acceptable addition that also fulfils the pod coverage requirements.

    • Documenting each attribute in Neo4j::Driver::SummaryCounters as individual methods might be a quick way to bring up the pod coverage stats a bit.

  • Neo4j::Test should auto-detect the Neo4j server version and set the cypher_filter config option accordingly.

  • Neo4j::Sim should implement the driver HTTP net module API instead of the REST::Client API, so that we can eventually replace REST::Client.

  • Write new unit tests for all modules.

  • Optimise the simulator for $hash = 0. Use of << causes statements to end with \n, which the simulator could filter out. The internals test "transaction: REST 404 error handling" should run a distinct statement.

  • Try to change "no connection" tests such that no NXDOMAIN responses are required by (ab)using a particular port number on localhost; the port could be checked beforehand and the test skipped if it happens to be open for some reason. Alternatively, the tests in question could be changed to a simulated connection failure.

  • Verify that "get" in Neo4j::Driver::Type::Node and "get" in Neo4j::Driver::Type::Relationship really do return undef (as documented), even when called in list context.

  • Add test for byte arrays (RETURN apoc.text.bytes("DATA") AS x).

  • Check that nodes without labels and nodes without properties are handled correctly in all aspects (there might be issues with perlbolt).

  • Check that after a server error, the next statement will succeed (there might be issues with perlbolt; see majensen/libneo4j-client#8).

  • List possible croak output in "DIAGNOSTICS" in Neo4j::Driver, allowing for indexing by search engines.

  • "notifications" in Neo4j::Driver::ResultSummary: Clarify docs that this method is to be called in list context.

Other ideas for specific modules


  • Make the URL a config option, so that it can be queried (and changed).

  • Make the auth data a config option, so that it can be queried. As alternative ways to set the auth data, basic_auth() should continue to be supported as an alias and the user info should be parsed from the URL if given (however, URLs without user info should not change the auth data stored in the driver). A possible implementation would be to create a new AuthToken module that would offer suitable methods, but this seems like overkill. That said, userinfo is actually forbidden now by RFC7230:2.7.1, so this should perhaps not be added after all?

  • Allow passing config options directly to the constructor, e. g. in place of the URL (Neo4j::Driver->new({ url=>"bolt:", timeout=>30 })).

  • Change the default URI scheme from HTTP to auto-detect, i. e. try Bolt first, then HTTP in case of failure. This could be explicitly specified as e. g. //localhost.

  • The neo4j scheme could perhaps be mapped onto Bolt or onto the default URL scheme, just so that neo4j://... URLs will kind of work.

  • Consider writing a concrete example that re-creates LOMS logic by re-blessing the structural types into custom types from the business logic using $cypher_types->{init}. (For example, check for Neo4j nodes that are labelled :Person and re-bless those as Local::Person or whatever.)


  • Once a session is created, the driver object becomes immutable. It should therefore be possible to store the ServerInfo in the driver object once it is obtained. If the default database is added as well, the Discovery API doesn't need to be used again for a new session. This change would keep down network utilisation in scenarios where many sessions are created (such as running the driver's test suite).

  • Consider whether to offer transaction functions. If available, these should consist of subrefs passed to methods called write_transaction and read_transaction. These access modes are only an optimisation for Enterprise features. We don't target those at present, but read_transaction could then eventually be routed to a high-performance read-only server once clusters are supported. It would make sense to offer both methods right away even though initially they'd work exactly the same.


  • Consider supporting re-using Record objects for query parameters in run. The Java and C# drivers do this.

  • Run statements lazily: Just like with the official drivers, statements passed to run should be gathered until their results are actually accessed. Then, and only then, all statements gathered so far should be sent to the server using a single request. Challenges of this approach include that notifications are not associated with a single statement, so there must be an option to disable this behaviour; indeed, disabled should probably be the default when stats are requested. Additionally, there are some bugs with multiple statements (see tests non-arrayref individual statement and include empty statement). Since stats are now requested by default, this item might mean investing time in developing an optimisation feature that is almost never used. Since the server is often run on localhost anyway where latency is very close to zero, this item should not have high priority.

  • The current API to "Execute multiple statements at once" in Neo4j::Driver::Transaction seems a bit too complicated, especially for something that is probably hardly ever actually used in practice. It might make sense to provide a dedicated method (e. g. _run_multiple()) for just that purpose. This would eventually free up the run() implementation from unneeded baggage. The dedicated method may be private as long as the plan still is to provide this functionality by running statements lazily.



  • Perhaps fetch() should always buffer two records instead of just one. With the current implementation, the bolt connection might remain attached longer than desirable in cases where the client knows in advance how many records there will be and calls fetch() exactly that number of times. (In theory, such a change might even slightly improve performance if the driver uses Perl threads to fill the buffer in the background.)


  • The entire package can probably be removed now.


  • Profile the server-side performance penalty of requesting stats for various kinds of queries. If the penalty turns out to be high, stats should perhaps have to be requested explicitly by clients (rather than being obtained by default, as with 0.13 and higher). However, using Bolt always provides stats, and different APIs for HTTP and Bolt seem like a bad idea.


  • use Class::Accessor::Fast 0.34;

  • It seems Neo4j 4 added new counters for system updates.


  • Rollback behaviour on errors needs further study. Neo4j Status Codes says that all errors have a rollback effect, but in at least some cases, the effect seems to be merely to mark the tx as failed and uncommittable, which isn't quite the same thing. This may or may not vary across error types, Neo4j versions, or Bolt versions. OTOH, some errors are internal client errors that shouldn't rollback the tx (majensen/libneo4j-client#7). Not sure if these occur in practice, but we should probably be able to handle them correctly anyway.


  • Consider unrolling deep_bless recursion. Based on initial profiling, this may save up to about 5% CPU time (for a specific HTTP test query cached in RAM, performance went from about 2700/s to 2850/s when skipping the call to deep_bless entirely). However, when accessing the database, the bottleneck is typically I/O (querying Neo4j itself instead of the RAM-cached response let the performance for the very same query drop down to 650/s when executed over HTTP). So this optimisation may not be worth it (OTOH, Bolt performance was something like 7000/s, so optimising deep_bless may be more useful there).

  • If a 201 is received without a Location header, it is currently simply ignored by _parse_tx_status(). (The simulator requires this.) According to RFC 7231, such a response means the location hasn't changed, i. e. the resource has been created at the default transaction endpoint. That should never happen; in fact, it should only ever happen for a PUT request, but we don't use those here. So ignoring this is probably the right choice. But it may still be useful to revisit this logic later on.

  • Parse the expires date for transactions, sync client and server system dates using the Date header and track transaction expiration.


  • Profile whether X-Stream has any performance impact. Remove if in doubt, as we don't implement streaming (and have no plans to in the future, considering that Bolt should be used when speed is important).

  • Remove the REST::Client dependency and use <LWP::UserAgent> directly.