capnproto

FORK: Cap'n Proto serialization/RPC system - core tools and C++ library
git clone https://git.neptards.moe/neptards/capnproto.git
Log | Files | Refs | README | LICENSE

roadmap.md (7394B)


      1 ---
      2 layout: page
      3 title: Road Map
      4 ---
      5 
      6 # Road Map
      7 
      8 This is a list of big ideas we'd like to implement in Cap'n Proto. We don't know in what order
      9 these will actually happen; as always, real work is driven by real-world needs.
     10 
     11 ## Language Features
     12 
     13 * **Inline lists:**  Lets you define a field composed of a fixed number of elements of the same
     14   type, and have those elements be stored directly within the struct rather than as a separate
     15   object.  Useful mainly to avoid the need to validate list lengths when the length should always
     16   be the same.  Also saves a pointer's worth of space.
     17 * **Type aliases:**  Ability to define a type which is just an alias of some other type, and
     18   have it show up as e.g. a `typedef` in languages that support that.  (The current `using`
     19   keyword is intended only for local use and does not affect code generation.)
     20 * **Doc comments:**  Harvest doc comments from schema files and use them to generate doc comments
     21   on generated code.  Also make them available in the compiled schema so that a documentation
     22   generator could use them.
     23 * **Encapsulated types:**  This will allow you to create a hand-written wrapper around a
     24   type which will be automatically injected into the generated code, so that you can provide a
     25   nicer interface which encapsulates the type's inner state.
     26 * **Maps:**  Based on encapsulated and parameterized types.
     27 
     28 ## RPC Protocol Features
     29 
     30 * **Dynamic schema transmission:**  Allow e.g. Python applications to obtain schemas directly from
     31   the RPC server so that they need not have a local copy.  Great for interactive debugging.
     32 * **Three-way introductions (level 3 RPC):**  Allow RPC interactions between more than two parties,
     33   with new connections formed automatically as needed.
     34 * **Bulk and Realtime**: Add features that make it easier to design Cap'n Proto APIs for bulk
     35   data transfers (with flow control) and realtime communications (where it's better to drop
     36   messages than to deliver them late).
     37 * **UDP transport**: Cap'n Proto RPC could benefit from implementing a UDP transport, in order
     38   to achieve zero-round-trip three-party introductions and to implement "realtime" APIs (see
     39   "bulk and realtime", above).
     40 * **Encrypted transport**: Cap'n Proto RPC should support an encrypted transport which uses
     41   capability-based authorization (not PKI), can accomplish zero-round-trip three-party
     42   introductions (via a pre-shared key from the introducer) and based on modern crypto. TLS is
     43   not designed for this, but we don't want to invent new crypto; we intend to build on
     44   [libsodium](https://github.com/jedisct1/libsodium) and the
     45   [Noise Protocol Framework](http://noiseprotocol.org/) as much as possible.
     46 
     47 ## C++ Cap'n Proto API Features
     48 
     49 * **Plain Old C Structs:** The code generator should also generate a POCS type corresponding
     50   to each struct type. The POCS type would use traditional memory allocation, thus would not
     51   support zero-copy, but would support a more traditional and easy-to-use C++ API, including
     52   the ability to mutate the object over time without convoluted memory management. POCS types
     53   could be extracted from and inserted into messages with a single copy, allowing them to be
     54   used easily in non-performance-critical code.
     55 * **Multi-threading:**  It should be made easy to assign different Cap'n Proto RPC objects
     56   to different threads and have them be able to safely call each other. Each thread would still
     57   have an anyschronous event loop serving the objects that belong to it.
     58 * **Shared memory RPC:**  Zero-copy inter-process communication.
     59 * **JSON codec customization:**  Extend the JSON library to support customizing the JSON
     60   representation using annotations. For example, a field could be given a different name in
     61   JSON than it is in Cap'n Proto. The goal of these features would be to allow any reasonable
     62   pre-existing JSON schema to be representable as a Cap'n Proto type definition, so that
     63   servers implementing JSON APIs can use Cap'n Proto exclusively on the server side.
     64 * **LZ4 integration:**  Integrate LZ4 compression into the API to further reduce bandwidth needs
     65   with minimal CPU overhead.
     66 * **Annotations API:**  For each annotation definition, generate code which assists in extracting
     67   that annotation from schema objects in a type-safe way.
     68 
     69 ## C++ KJ API Features
     70 
     71 KJ is a framework library that is bundled with Cap'n Proto, but is broadly applicable to C++
     72 applications even if they don't use Cap'n Proto serialization.
     73 
     74 * **Fiber-based concurrency:**  The C++ runtime's event loop concurrency model will be augmented
     75   with support for fibers, which are like threads except that context switches happen only at
     76   well-defined points (thus avoiding the need for mutex locking).  Fibers essentially provide
     77   syntax sugar on top of the event loop model.
     78 * **TLS bindings:** Write bindings for e.g. OpenSSL to make it easy to integrate with the KJ
     79   I/O framework, Cap'n Proto RPC, and the KJ HTTP library.
     80 * **Modern crypto bindings:** A thin wrapper around
     81   [libsodium](https://github.com/jedisct1/libsodium) with a nice C++ API, e.g. representing
     82   keys using fixed-size, trivially-copyable classes.
     83 * **Event loop integrations:** We should provide off-the-shelf integrations with popular event
     84   loop libraries, such as libuv, libev, libevent, boost::asio, and others, so that it's easier
     85   to use Cap'n Proto RPC in applications that already use another event framework.
     86 
     87 ## Storage
     88 
     89 * **ORM interface:**  Define a standard interface for capabilities that represent remotely-stored
     90   objects, with get, put, publish, and subscribe methods.  Ideally, parameterize this interface
     91   on the stored type.
     92 * **mmap-friendly mutable storage format:**  Define a standard storage format that is friendly
     93   to mmap-based use while allowing modification.  (With the current serialization format, mmap
     94   is only useful for read-only structures.)  Possibly based on the ORM interface, updates only
     95   possible at the granularity of a whole ORM entry.
     96 
     97 ## Tools
     98 
     99 * **Schema compatibility checker:**  Add a `capnp` command which, given two schemas, verifies
    100   that the latter is a compatible upgrade from the former.  This could be used as a git hook
    101   to prevent submission of schema changes that would break wire compatibility.
    102 * **RPC debugger:**  Add a `capnp` command which sends an RPC from the command line and prints
    103   the result.  Useful for debugging RPC servers.
    104 
    105 ## Quality Assurance
    106 
    107 These things absolutely must happen before any 1.0 release.  Note that it's not yet decided when
    108 a 1.0 release would happen nor how many 0.x releases might precede it.
    109 
    110 * **Expand test coverage:**  There are lots of tests now, but some important scenarios, such as
    111   handling invalid of invalid input, need better testing.
    112 * **Performance review:**  Performance is already very good compared to competitors, but at some
    113   point we need to break out the profiler and really hone down on the details.
    114 * **Security review:**  We need a careful security review to make sure malicious input cannot
    115   crash an application or corrupt memory.
    116 
    117 ## Infrastructure
    118 
    119 Note:  These are very large projects.
    120 
    121 * **JSON-HTTP proxy:**  Develop a web server which can expose a Cap'n Proto RPC backend as a
    122   JSON-over-HTTP protocol.
    123 * **Database:**  A fast storage database based on Cap'n Proto which implements the ORM interface
    124   on top of the mmap storage format.