THRIFT-5830: Add per-call transport support for netstd

.github/agents/openspec.agent.md

@@ -0,0 +1,65 @@
+---
+name: OpenSpec
+description: "Manages OpenSpec changes, specs, and workflows using the OpenSpec CLI. Use this agent for proposing changes, exploring ideas, validating artifacts, checking status, and archiving completed work."
+tools:
+  - "execute"
+  - "read"
+  - "search"
+  - "edit"
+---
+
+<!-- Generated by OpenSpec for GitHub Copilot coding agent support. -->
+
+# OpenSpec Agent
+
+You are a specialized agent for managing OpenSpec workflows. Before using the `openspec` CLI, run `openspec --version`. If it is unavailable, install it with `npm install -g @fission-ai/openspec@1.14.1`.
+
+## What is OpenSpec?
+
+OpenSpec is a structured change management system for codebases. It organizes work into **changes** with planning artifacts (proposals, specs, designs, tasks) that guide implementation.
+
+## Available Commands
+
+### Agent-Compatible CLI Commands (prefer `--json` for structured output)
+
+| Command | Purpose |
+|---------|---------|
+| `openspec list [--json]` | List all changes and specs |
+| `openspec show <item> [--json]` | View a specific change or spec |
+| `openspec validate [--all] [--json]` | Validate changes and specs for issues |
+| `openspec status [--change <name>] [--json]` | Show artifact progress for a change |
+| `openspec instructions [artifact] [--change <name>] [--json]` | Get next-step instructions for an artifact |
+| `openspec templates [--json]` | List available templates |
+| `openspec schemas [--json]` | List available workflow schemas |
+| `openspec archive <change> --json [--yes]` | Archive a completed change; use `--yes` only after confirming all tasks are complete |
+
+### Interactive CLI Commands (use when prompted by the user)
+
+| Command | Purpose |
+|---------|---------|
+| `openspec init` | Initialize OpenSpec in the project |
+| `openspec update` | Update OpenSpec configuration and artifacts |
+| `openspec view` | Interactive dashboard |
+| `openspec config` | View or modify settings |
+
+## Workflow
+
+Run OpenSpec commands from `lib/netstd`, where this change's OpenSpec project is located.
+
+1. Run `openspec list --json` to see active changes and specs.
+2. Run `openspec status --change <name> --json` for the selected change.
+3. Run `openspec instructions [artifact] --change <name> --json` for the next-step instructions.
+4. Run `openspec validate <name> --json` before completing.
+
+## Key Directories
+
+- `lib/netstd/openspec/` — OpenSpec directory
+- `lib/netstd/openspec/changes/` — Active changes and their artifacts
+- `lib/netstd/openspec/config.yaml` — Project configuration
+
+## Best Practices
+
+- Always use `--json` when parsing output programmatically
+- Run `openspec validate` after creating or modifying artifacts
+- Check `openspec status` before starting work to understand the current state
+- When archiving, ensure all tasks are completed and validated first

.github/workflows/copilot-setup-steps.yml

@@ -0,0 +1,26 @@
+# Generated by OpenSpec for GitHub Copilot coding agent support.
+
+name: "Copilot Setup Steps"
+
+# Run manually to verify the setup; Copilot runs these steps when starting agent work.
+on:
+  workflow_dispatch:
+
+jobs:
+  # The job MUST be called `copilot-setup-steps` for Copilot coding agent to pick it up.
+  copilot-setup-steps:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+
+    permissions:
+      contents: read
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Install OpenSpec CLI
+        run: npm install -g @fission-ai/openspec@1.14.1
+
+      - name: Verify OpenSpec CLI
+        run: openspec --version

compiler/cpp/src/thrift/generate/t_netstd_generator.cc

@@ -2432,13 +2432,30 @@ void t_netstd_generator::generate_service_client(ostream& out, t_service* tservi
         << indent() << "{" << '\n';
     indent_up();
 
-    out << indent() << "public Client(TProtocol protocol) : this(protocol, protocol)" << '\n'
+    out << indent() << "/// <summary>Initializes the client with a shared protocol.</summary>" << '\n'
+        << indent() << "/// <remarks>Use this constructor when the application already owns and configures the protocol. It preserves shared-protocol behavior; use the transport and protocol-factory constructor to opt into per-call transports.</remarks>" << '\n'
+        << indent() << "/// <param name=\"protocol\">The shared protocol used for input and output.</param>" << '\n'
+        << indent() << "public Client(TProtocol protocol) : this(protocol, protocol)" << '\n'
         << indent() << "{" << '\n'
         << indent() << "}" << '\n'
         << '\n'
+        << indent() << "/// <summary>Initializes the client with shared input and output protocols.</summary>" << '\n'
+        << indent() << "/// <remarks>This constructor remains available for callers that manage protocol instances directly and preserves their shared-transport semantics. For concurrent per-call operations, use a supporting transport with the protocol-factory constructor.</remarks>" << '\n'
+        << indent() << "/// <param name=\"inputProtocol\">The shared input protocol.</param>" << '\n'
+        << indent() << "/// <param name=\"outputProtocol\">The shared output protocol.</param>" << '\n'
         << indent() << "public Client(TProtocol inputProtocol, TProtocol outputProtocol) : base(inputProtocol, outputProtocol)" << '\n'
         << indent() << "{" << '\n'
         << indent() << "}" << '\n'
+        << '\n'
+        << indent() << "/// <summary>Initializes the client with protocols created for a supplied transport.</summary>" << '\n'
+        << indent() << "/// <remarks>When <paramref name=\"transport\"/> supports <see cref=\"ITPerCallTransportProvider\"/>, generated asynchronous calls use a separate transport and protocol state per call. Existing protocol constructors retain shared-protocol behavior.</remarks>" << '\n'
+        << indent() << "/// <param name=\"transport\">The shared transport or per-call transport provider.</param>" << '\n'
+        << indent() << "/// <param name=\"inputProtocolFactory\">The factory used to create input protocols.</param>" << '\n'
+        << indent() << "/// <param name=\"outputProtocolFactory\">The factory used to create output protocols.</param>" << '\n'
+        << indent() << "public Client(TTransport transport, TProtocolFactory inputProtocolFactory, TProtocolFactory outputProtocolFactory)" << '\n'
+        << indent() << "    : base(transport, inputProtocolFactory, outputProtocolFactory)" << '\n'
+        << indent() << "{" << '\n'
+        << indent() << "}" << '\n'
         << '\n';
 
     vector<t_function*> functions = tservice->get_functions();
@@ -2448,23 +2465,31 @@ void t_netstd_generator::generate_service_client(ostream& out, t_service* tservi
     {
         string raw_func_name = (*functions_iterator)->get_name();
         string function_name = raw_func_name + (add_async_postfix ? "Async" : "");
+        string call_cancellation_token = tmp("callCancellationToken");
 
         // async
         generate_deprecation_attribute(out, (*functions_iterator)->annotations_);
         out << indent() << "public async " << function_signature_async(*functions_iterator, "") << '\n'
             << indent() << "{" << '\n';
         indent_up();
+        bool returns_value = !(*functions_iterator)->is_oneway() && !(*functions_iterator)->get_returntype()->is_void();
+        out << indent() << (returns_value ? "return await " : "await ")
+            << "ExecutePerCallAsync(async (" << call_cancellation_token << ") =>" << '\n'
+            << indent() << "{" << '\n';
+        indent_up();
         out << indent() << "await send_" << function_name << "(";
         string call_args = argument_list((*functions_iterator)->get_arglist(),false);
         if(! call_args.empty()) {
             out << call_args << ", ";
         }
-        out << CANCELLATION_TOKEN_NAME << ");" << '\n';
+        out << call_cancellation_token << ");" << '\n';
         if(! (*functions_iterator)->is_oneway()) {
-            out << indent() << ((*functions_iterator)->get_returntype()->is_void() ? "" : "return ")
-                            << "await recv_" << function_name << "(" << CANCELLATION_TOKEN_NAME << ");" << '\n';
+            out << indent() << (returns_value ? "return " : "")
+            << "await recv_" << function_name << "(" << call_cancellation_token << ");" << '\n';
         }
         indent_down();
+        out << indent() << "}, " << CANCELLATION_TOKEN_NAME << ");" << '\n';
+        indent_down();
         out << indent() << "}" << '\n' << '\n';
 
         // async send

lib/netstd/Makefile.am

@@ -74,6 +74,8 @@ dist-hook:
 
 EXTRA_DIST = \
 	README.md \
+	per-call-transport.md \
+	openspec \
 	Directory.Build.props \
 	Benchmarks/Thrift.Benchmarks \
 	Tests/codegen/run-NetStd-Codegen-Tests.ps1 \

lib/netstd/README.md

@@ -14,6 +14,30 @@ The library ships as two packages so that non-web projects no longer pull in the
   *in addition to* `ApacheThrift` only when you host Thrift over ASP.NET Core. Existing code
   keeps compiling unchanged once the package reference is added.
 
+# Per-call HTTP client calls
+
+For concurrent asynchronous calls, construct the generated client with a transport that
+implements `ITPerCallTransportProvider` and protocol factories. `THttpTransport`,
+`TBufferedTransport`, and `TFramedTransport` over a supporting transport provide this capability. Each generated
+high-level call then gets independent transport and protocol state while the HTTP connection
+pool is shared:
+
+```csharp
+var transport = new THttpTransport(new Uri("http://localhost:9090"), new TConfiguration());
+var protocolFactory = new TBinaryProtocol.Factory();
+using var client = new Calculator.Client(transport, protocolFactory, protocolFactory);
+
+var results = await Task.WhenAll(Enumerable.Range(1, 8)
+    .Select(value => client.add(value, value, cancellationToken)));
+```
+
+The existing constructors that accept `TProtocol` instances remain the shared-protocol path
+for applications that manage protocol instances directly. Per-call behavior is opt-in; a
+transport without the provider capability continues to use shared behavior. The per-call scope
+covers generated high-level asynchronous methods that perform a complete RPC. Calling generated
+`send_*` and `recv_*` methods separately continues to use the shared protocols and must not be
+interleaved on the same client.
+
 # Build the library
 
 ## How to build on Windows