build: simplify change of C++ version in CMakeLists.txt#1000
Open
JochenHiller wants to merge 28 commits intoCOVESA:masterfrom
Open
build: simplify change of C++ version in CMakeLists.txt#1000JochenHiller wants to merge 28 commits intoCOVESA:masterfrom
JochenHiller wants to merge 28 commits intoCOVESA:masterfrom
Conversation
Preparation for easy compiling with latest C++ standards. This change does work with C++17, C++20 and C++23 with current toolchain from Dockerfile. E.g. `cmake -DVSOMEIP_CXX_STANDARD=23 .` will compile against C++23 standard
…EIP_CXX_STANDARD * QNX build: hard reference to C++17, use `VSOMEIP_CXX_STANDARD` instead * tests will be compiled with C++20 since this [commit](COVESA@9ff1977#diff-33394812ba204689144fd2f80832db83853ba1cb32403edb4e15fe4893e675fd). Change will enforce to use C++20 when C++17 is specified in `VSOMEIP_CXX_STANDARD` * internal_routing_disabled_acceptance_test: hard reference to C++17, use `VSOMEIP_CXX_STANDARD` instead
Author
|
Updated PR: Change more build related references to C++ standard and changed them to use build variable
I did tests of compile and execute unit tests of these variants (using current toolchain in Dockerfile, gcc 11.4.0):
Question: Is it intended that tests are now compiled with C++20? If so, maybe documentation should be updated. |
The scenario is: there is a client, a server, and vsomeipd client connects to server, does a SUBSCRIBE server receives SUBSCRIBE, tries to connect to do CONFIG_ID+SUB_ACK+initial value client is unreachable for a bit (suppose: is in a guest) server has to reconnect a few times client should receive all of the above
Add application names Cleanup .sh starter files Add test documentation
This reverts commit 71cd2d5.
Fix routing tests return codes and add documentation Updated sleep times for external_local_routing_test_starter.sh and local_routing_test_starter.sh. Fixed return codes. Fixed memory leaks in usr/bin/* that are used in test scripts, but it should be fine to ignore as it does not impact the functioning of the test. Added documentation.
Consistency when existing test scripts Consistency across test scripts Most of tests are running with interpret as bash. There is no plausible reason for this tests to run w/ sh.
Add include of utility. Drop unneeded include of asio/io_context.hpp
Document and fix test scripts for offer_test Remove unnecessary sleeps, fixed some typos
In local communication, if a subscribe is sent to the service-provider, but right before the subscription ack/nack is sent, the connection service-provider -> service-consumer is removed, service-provider detects and executes handle_client_error, however service-consumer will not detect any connection problem and still be waiting for subscribe-ack. What was happening is that right after replacing the old connection, remove_connection was executed from ltsei::receive_cbk when detecting an issue. This would then make the disconnect_from impossible to be executed on the endpoint to the client, as no connection was present. The client would be unaffected and not detect any problem keeping it in a "limbo" state. By specifying the connection to remove, we can avoid removing a just-updated connection and prevent this behavior. Also made the connection's stop and destructor be executed outside of connections_mutex_.
The local receiver is being created on demand, but this causes issues with TCP communication, as it is the local receiver that defines the port used for vsomeip_sec Therefore it should be created unconditionally, all the time, at start
Introduce a thread-safe timer utility This timer utility can be used to either run a job periodically, or as a one shot timer. This utility is going to be used in upcoming work. Meanwhile introduce the capability to tinker with these timers, e.g. in fake_socket tests. Note that this would first require the actual use of this utility.
It is a simple scenario, routingd-as-server. It makes sense, because there is a different implementation for routing client vs routing host
If a user forgets to set the message type (so neither request, nor response, nor notification...), the message will fail to arrive anywhere, but often with a very cryptic error, such as: rmi::send: Routing error. Endpoint for (...) could not be found! This is too difficult to debug. Made it obvious!
It is important for local TCP communication that libvsomeip polls for data acknowledgement before closing a connection with TCP RST, which is why 1f18549 makes sense However, it does not make sense to wait neither for data to be sent, nor for data ACKs if a connection is closed due to an error Therefore, change endpoints such that the "are we stopping due to an error?" information is propagated up to the endpoint This requires a change on all endpoints, but note that it only really matters for local client endpoints - local server endpoints currently do not send data (excluding ASSIGN_CLIENT), and boardnet endpoints do not care about graceful shutdown
SonarQube IDE (along a few other things) rely on compile_commands.json, so set the variable to create it. For SonarQube: Build project with "CMAKE_EXPORT_COMPILE_COMMANDS ON" Install SonarQube for IDE extension SonarQube should detect the new "compile_commands.json" present in the build directory Get SonarQube complaints and suggestions live on your IDE
Most of the network tests run shell scripts that run many different executables, so need to instruct valgrind to produce a different log file for each The valgrind documentation recommends the same when using --log-file and --trace-children=yes
Fix broken link from markdown to png asset. subscribe_notify_test.png is subscribe_notify_test_one_tests.png for subscribe_notify_test_one_tests.md
Remove unused state (use_count_, opened_, local_mutex_); it is used nowhere and makes endpoints considerably bigger Furthermore, move magic cookie state to tcei/tsei, as this is exclusive to TCP communication, and anyhow the endpoint manager knows it is dealing with TCP endpoints due to the configuration lookups
This test covers two issues found in the start-stop-start process of an application. Start-stop-start of a single application: Prior to e8b6519, with local TCP communication, the following sequence would break a vsomeip application: Scenario: app->start(); app->stop(); app->start(); // application never registers to the daemon, nothing works e8b6519 solved this problem. Start-stop-start with two applications: If a new application is started between the stop and the second start of the first application, the start of the first application will fail. This happens because, since 'create_local_server' is not called in routing_manager_client::start(), the server endpoint will be assigned, in 'create_local', to a memory port that is already being used by application 2. Scenario: app1->start() - port 31493 app1->stop() - endpoints destroyed and ports released app2->start() - port 31493 app1->start() - port 31493 because create_local_server is not called -> registration fails because the address is already assigned b9f8c84 solves the problem.
This change fixes an edge-case in the subscription mechanism where the routing host was able to send multiple unsubscribe requests to the service for the same subscription. Additionally, it also improves the logging of subscription requests between applications. Fix This edge-case is visible when a remote client is using inconsistent Session IDs to subscribe to the service, triggering multiple reboot detections on the host. The host then sends multiple unsubscribe requests while the service is still busy processing the first one. The fix involves simply checking if a subscription has already been expired before sending another unsubscribe requests to the service. Log The change to the logs adds the subscription id to the subscription related traces, such SUBSCRIBE, UNSUBSCRIBE, SUBSCRIBE_ACK etc. This makes it easier to track the subscription as it bounces between service, client and host. For example: SUBSCRIBE(0000): [0001.0001.0001:ffff:0] accepted. id=0002 subscribers=2 SUBSCRIBE ACK(0101): [0001.0001.0001.ffff] id=0002
Set the TCP_QUICKACK socket option Since in our sockets the linger is set to 0, upon socket close the tcp connections are immediately destroyed This is a problem in the case that there are pending TCP ACKs to be sent for that connection, where with the connection being destroyed if any ACKs are not sent until then, the ACK will never be sent and the other side never receives it To prevent this situation, set the option TCP_QUICKACK in order for the ACKs to be sent before the 100ms timer between calling the shutdown_and_close to the socket and effectively closing it
A possible race condition between the transition of a binded socket towards a connection oriented UDP socket enables a scenario where packets can be dropped and a ICMP port unreachable sent back. When the routing host processes an offer that is to be offered/subscribed exclusively on an UDP port, the client endpoint is created, socket binded to local address and then it asynchronously starts the connection oriented process (define remote address for endpoint), while doing it asynchronously, the routing host sends the service availability to clients and might even receive their subscriptions and send it to the remote partner via service discovery. This asynchronous behavior is considered to be a possible root cause for the ICMP port unreachable message seen for the initial event being dropped in Other possible root causes for port unreachable are: Port unavailable, socket not binded, do not believe is the scenario as the endpoint creation is synchronous and previous to the availability. Sender is not the remote address defined for the connection, also not the scenario here. A fix is presented in this PR where the service availability is only sent to the clients when the UDP sockets is considered "connected". Climate test is updated as availability handler was called more than the expected number of times as the offer was processed with the service not being requested but since an availability handler was still registered and the client is the host, the handler was invoked with the service state: AS_OFFERED.
Deliver service availability only after endpoint creation set tcp quick ack Fix remote subscription expiration new network-test for start-stop-start application endpoint: drop unused state Update subscribe_notify_test_one_tests.md zuul: catch all valgrind failures build: enable CMAKE_EXPORT_COMPILE_COMMANDS endpoint: no waiting when stopping due to error app: fail early, fail fast for unknown messages tests: add another fake_socket test Introduce a thread-safe timer utility rmi: create local receiver on start fix race condition in connections_ map changes Fix payload tests return codes Fix/doc offer_test Support compilation with gcc15 tests: updates shebang line to interpret as bash Fix network_tests/routing_tests Revert "IO threads exit mitigation " Fix and document second_address tests tests: add regression test
Preparation for easy compiling with latest C++ standards. This change does work with C++17, C++20 and C++23 with current toolchain from Dockerfile. E.g. `cmake -DVSOMEIP_CXX_STANDARD=23 .` will compile against C++23 standard
…EIP_CXX_STANDARD * QNX build: hard reference to C++17, use `VSOMEIP_CXX_STANDARD` instead * tests will be compiled with C++20 since this [commit](COVESA@9ff1977#diff-33394812ba204689144fd2f80832db83853ba1cb32403edb4e15fe4893e675fd). Change will enforce to use C++20 when C++17 is specified in `VSOMEIP_CXX_STANDARD` * internal_routing_disabled_acceptance_test: hard reference to C++17, use `VSOMEIP_CXX_STANDARD` instead
…hanceChangeCppVersion
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Preparation for easy compiling with latest C++ standards.
This change does work with C++17, C++20 and C++23 with current toolchain from Dockerfile.
E.g.
cmake -DVSOMEIP_CXX_STANDARD=23 .will compile against C++23 standard