Instead of making a blocking foreign call without a timeout, we set a
timeout of one second. This way, the thread returns in Haskell code (and
thus is interruptible) at least every second. This is useful in order to
be able to kill `Server`s, see
https://github.com/awakesecurity/gRPC-haskell/issues/22.
* remove parent ptr from unregistered calls -- unneeded
* begin unregistered high level server loop
* undo changes to highlevel server, add mkConfig for unregistered server
* move call CQ create/destroy into call create/destroy
* async normal call function
* preliminary unregistered server loop for non-streaming methods
* working unregistered highlevel example
* loop counters for benchmarking
* changes for benchmarking, add ruby example server for benchmarking
* async version of withCall, refactor unregistered server loop to handle all method types
* unregistered client streaming
* add remaining streaming modes
* unregistered server streaming test
* unregistered streaming tests
* add error logging
* fix bug in add example
* remove old TODOs
* fix bug: don't assume slices are null-terminated
* add TODO re: unregistered client streaming functions
* Tweak runOps param order, inline common op sequences, clean up serverHandleNormalCall
* More ops sequence inlining for clarity, experimenting with Managed
* Checkpoint: preliminary support for all streaming modes; much cleanup/refactoring and api design still needed
* Use mempty for default StatusDetails; tweak bad status matching mechanism
* Preliminary user-facing, server-streaming, low-level api and test
* renaming wibbles
* Preliminary user-facing, client-streaming, low-level api and test
* Move sendMsgs comb to Network.GRPC.LowLevel.Op; misc cleanup/DCR
* Modify bidi streaming to omit request payload
* Add transformers dep
* Preliminary user-facing low-level bidirectional streaming api and test
* Fix missing peek import
* Remove TimeoutSeconds params on streaming mode functions
* Fix serverHandleNormalCall rebase wart
* Fix rebase warts; minor hlint fixes and wibbles
* Post-rebase tweaks to optional payload use in serverRequestCall (i.e., now respects payloadHandling again)
* Cleanup/refactor serverRequestCall
* Fix comment
* Change ServerRWHandler type so that handler does not have to invoke a finalizer
* Change ServerReaderHandler type so that handler does not have to invoke a finalizer
* Simplify serverWriter interface and ServerWriterHandler structure
* Simplify serverRW (get rid of exec param), improve bidi streaming tests
* Use ExceptT in serverRW impl
* Change ServerRWHandler type to pass recv/send operations.
* Renaming
* Define ClientRWHandler, pass recv/send ops
* wibbles
* Use ExceptT in clientRW impl
* Add DataKinded phantom typing to RegisteredMethod; misc cleanup
* Simplify sendMsgs interface; add SingleSend type and related helpers
* Rename SingleSend to SendSingle, use ExceptT to clean up {client,server}Writer and sendMsgs
* More ExceptT cleanup in clientWriter
* Factor out reusable bits of clientWriter
* Shrink ServerReaderHandler
* Delete stale comments
* begin high-level server interface
* update to datakind representation
* clean up
* move method type info to type level, parametrize ServerCall by payload
* convert for writer handler
* start switching over to Message-based handlers
* begin work on highlevel example
* comment out old code
* parametrize StreamSend
* parametrize StreamRecv
* conversion for ServerReaderHandler
* finish handler conversions
* Add high level version and payload checking to echo-client
* Decouple server CQs from call-bound CQs (registered methods); use more consistent naming conventions
* Decouple server/call-bound CQs for unregistered methods; refactor U.serverRequestCall; misc cleanup
* Make convertRecv total; formatting wibbles
pluckers) would register a CQ notification via
grpc_server_request_registered_call prior to obtaining pluck permission, block
on acquiring such permission, and thus miss notification of the server call.
The bug can could be witnessed with particular interleavings by ensuring that
GRPC_MAX_COMPLETION_QUEUE_PLUCKERS server threads are blocked on their own
server call notifications, and then using the
(GRPC_MAX_COMPLETION_QUEUE_PLUCKERS+1)th server thread to attempt to block on a
notification for an additional server call.
In order to fix this, we simply extended the pluck permission held window for
serverRequestCall, and created a variant of pluck (called pluck') which assumes
permission to pluck is already held by the calling thread.
