docs: call ASIO and COROSIO I/O runtimes, and show the shape of an ooperation

.github/instructions/anyhttp.instructions.md

@@ -10,14 +10,14 @@ Everything is asynchronous and written with C++20 coroutines. Every operation is
 (`read_some`, `write`, `write_eof`, `submit`, `get_response`, `connect`, `get`) that yields a
 tuple and reports errors instead of throwing.
 
-There are two API styles, chosen at configure time with `-DANYHTTP_API=`:
+There are two I/O runtimes, chosen at configure time with `-DANYHTTP_API=`:
 
 * **ASIO** (default), on Boost.Asio: every operation is also an ASIO asynchronous operation
   (`async_x(..., token)`) and takes any completion token, adhering to its [async model].
 * **COROSIO**, on [capy](https://github.com/cppalliance/capy) and
   [corosio](https://github.com/cppalliance/corosio): coroutines only.
 
-Both styles build the same library sources and tests; code that belongs to one of them is guarded
+Both runtimes build the same library sources and tests; code that belongs to one of them is guarded
 with `ANYHTTP_ASIO` / `ANYHTTP_COROSIO` (exactly one is 1, from the generated `anyhttp/config.hpp`)
 and lives in `asio/` or `corosio/`, under both `include/anyhttp/` and `src/`. Generic code reaches
 the runtime only through `anyhttp/runtime.hpp`, `anyhttp/net.hpp` and `anyhttp/formatter.hpp`, and
@@ -33,7 +33,7 @@ functions `snake_case`, members `foo_`. Commit subjects are prefixed `fix:`, `re
 `chore:`, `style:`, `test:`, or with the protocol (`h1:`, `h2:`, `h3:`).
 
 Builds are out-of-source, one tree per configuration (`build/` Debug, `build-release/` Release,
-`build-corosio/` the COROSIO style, plus sanitizer trees); `scripts/configure.sh` sets them up.
+`build-corosio/` the COROSIO runtime, plus sanitizer trees); `scripts/configure.sh` sets them up.
 Run the tests from the repository root -- they load TLS material through relative paths -- and
 prefer `gtest-parallel build/test/test_all`.
 

.github/workflows/release.yml

@@ -18,7 +18,7 @@ env:
 
 jobs:
   #
-  # Both API styles in all three flavors, on OpenSSL: COROSIO needs it (corosio's openssl_stream
+  # Both I/O runtimes in all three flavors, on OpenSSL: COROSIO needs it (corosio's openssl_stream
   # uses a zero-copy BIO API that AWS-LC lacks), so it is the baseline rather than the default
   # AWS-LC. The legs after the product each isolate one variable the product holds fixed -- the
   # standard library, the TLS library -- so that a red job names its own cause. COROSIO TSAN is
@@ -50,7 +50,7 @@ jobs:
         api: [ASIO, COROSIO]
         flavor: [RelWithDebInfo, ASAN, TSAN]
         include:
-          # Flags per flavor, merged into the leg of each API style.
+          # Flags per flavor, merged into the leg of each I/O runtime.
           - flavor: RelWithDebInfo
             cmake: -DCMAKE_BUILD_TYPE=RelWithDebInfo
           - flavor: ASAN
@@ -59,7 +59,7 @@ jobs:
           - flavor: TSAN
             cmake: -DCMAKE_BUILD_TYPE=Debug -DCMAKE_CXX_FLAGS="-fsanitize=thread -fno-omit-frame-pointer" -DCMAKE_C_FLAGS="-fsanitize=thread -fno-omit-frame-pointer"
             opts: TSAN_OPTIONS=report_signal_unsafe=0
-          # Legs of their own: GCC and libstdc++ for both styles, and the other TLS library.
+          # Legs of their own: GCC and libstdc++ for both runtimes, and the other TLS library.
           - api: ASIO
             flavor: GCC
             cc: gcc

CLAUDE.md

@@ -13,10 +13,10 @@ benchmark with the latter, never with `build/`. Both use clang, which builds aga
 the libraries in `/opt/libc++` (`USE_LIBCXX`, `LIBCXX_ROOT`); `build-gcc/` is GCC with libstdc++
 and `/usr/local`. Keep `build/` on clang: clangd reads its `compile_commands.json`.
 
-`build-corosio/` is the COROSIO API style (`-DANYHTTP_API=COROSIO -DTLS_LIBRARY=OpenSSL`, see
+`build-corosio/` is the COROSIO I/O runtime (`-DANYHTTP_API=COROSIO -DTLS_LIBRARY=OpenSSL`, see
 [docs/corosio-port-plan.md](docs/corosio-port-plan.md)); capy and corosio are FetchContent'd at
-pinned SHAs. Both styles build the same library sources and test files; what belongs to one style
-only is guarded with `ANYHTTP_ASIO` or `ANYHTTP_COROSIO` (exactly one of them is 1). So is the
+pinned SHAs. Both runtimes build the same library sources and test files; what belongs to one
+runtime only is guarded with `ANYHTTP_ASIO` or `ANYHTTP_COROSIO` (exactly one of them is 1). So is the
 `server` program; the `client` program is ASIO-only so far. `build-corosio-asan/` is its ASAN
 tree, `build-corosio-release/` its Release tree (benchmark COROSIO with it) and
 `build-corosio-gcc/` its GCC one -- capy and corosio do build against libstdc++, but GCC is
@@ -137,10 +137,9 @@ buffer (see `WriteMode`), and `build-h2v2-asan/` is the version 2 ASAN tree for
 Each protocol lives in a directory of its own, `h1/` (Beast/HTTP/1.1), `h2/` (nghttp2), `quic/`
 (ngtcp2) and `h3/` (nghttp3, on top of `quic/`), in both `include/anyhttp/` and `src/`. **Only
 files in `h2/` may include `<nghttp2/*>`, only files in `h2v2/` `<nghttp2v2/*>`, only files in
-`quic/` and `h3/` `<ngtcp2/*>`, only files in `h3/` `<nghttp3/*>`.** What belongs to one API style
-only
-is in `asio/` or `corosio/`, again in both; the library compiles the `src/` directory of its style
-and not the other. `src/apps/` has the `server` and `client` programs, `src/issues/` standalone
+`quic/` and `h3/` `<ngtcp2/*>`, only files in `h3/` `<nghttp3/*>`.** What belongs to one async
+runtime only is in `asio/` or `corosio/`, again in both; the library compiles the `src/` directory
+of its runtime and not the other. `src/apps/` has the `server` and `client` programs, `src/issues/` standalone
 reproducers of upstream bugs. Everything else at the top level is generic.
 
 What gets installed is the public headers, the ones reachable from `client.hpp`, `server.hpp`,
@@ -194,7 +193,7 @@ may: the h2c upgrade and formatter tests drive nghttp2 by hand.)
 **Generic and protocol code include nothing from Boost.Asio but `<boost/asio/buffer.hpp>`**, which
 Beast's parser, serializer and `Fields` bring anyway, and nothing from capy or corosio. They reach
 the runtime through `anyhttp/runtime.hpp`, `anyhttp/net.hpp` and `anyhttp/formatter.hpp`, which
-include the `asio/` or `corosio/` header of the build's style. Addresses and endpoints are the
+include the `asio/` or `corosio/` header of the build's runtime. Addresses and endpoints are the
 runtime's own (`IpAddress`, `TcpEndpoint`, `UdpEndpoint`); where the two differ, `net.hpp` declares
 a function each runtime defines (`io::make_address()`, `io::to_sockaddr()`, ...). Verify in a
 COROSIO tree that no library object includes more of Asio:
@@ -204,7 +203,7 @@ ninja -C build-corosio -t deps | awk '/^[^ ].*: #deps/{o=$1} /boost\/asio\// &&
 ```
 
 It must print nothing. (Test objects may: `test_external.cpp` runs its child processes on an ASIO
-context in both styles.)
+context on both runtimes.)
 
 QUIC and HTTP/3 are two layers, and each shares one implementation between server and client.
 `quic/session.*` has all ngtcp2 callbacks, packet writing, timers and connection flow control,
@@ -222,16 +221,16 @@ shared files.
 ## Conventions
 
 - clang-format is authoritative: 3-space indent, 100 columns.
-- The two API styles are ASIO and COROSIO: `ANYHTTP_ASIO`/`ANYHTTP_COROSIO`,
+- The two I/O runtimes are ASIO and COROSIO: `ANYHTTP_ASIO`/`ANYHTTP_COROSIO`,
   `asio/`/`corosio/`, `build-corosio/`. "capy" names only the library itself (`capy::task`,
   `<boost/capy/...>`), the non-I/O base that corosio adds I/O to.
-- Where the two styles differ by more than a few lines, each gets a file of its own, of the same
+- Where the two runtimes differ by more than a few lines, each gets a file of its own, of the same
   name in `asio/` and `corosio/`, with the same definitions in the same order, so that the pair can
   be read side by side: `runtime.hpp`, `net.hpp`, `src/*/net.cpp`. What both share stays in the
   generic file of that name (`anyhttp/runtime.hpp`, which selects one of the two, `anyhttp/net.hpp`,
   `src/net.cpp`). The test fixtures follow the old pattern, `test/test_fixtures_*.hpp`. Keep the
   order when adding to either.
-- The style comes from the generated `anyhttp/config.hpp` (`cmake/config.hpp.in`), not from a
+- The runtime comes from the generated `anyhttp/config.hpp` (`cmake/config.hpp.in`), not from a
   compile definition: a file that tests `ANYHTTP_ASIO`/`ANYHTTP_COROSIO` must include it (or
   `anyhttp/runtime.hpp`) first, or the test is silently false.
 - Test `.cpp` files put `using namespace testing;` after the includes and use `HasSubstr`,
@@ -271,7 +270,7 @@ After adding a dependency, check `ldd` of an AWS-LC build shows no `libssl.so.3`
 `libcrypto.so.3` -- a shared OpenSSL would interpose the executable's AWS-LC symbols. `curl`,
 `osslclient` and `osslserver` are OpenSSL builds and are useful for interop testing.
 
-**A context has to run dry, in both styles.** This is anyhttp's doing, not capy's any more. Since
+**A context has to run dry, on both runtimes.** This is anyhttp's doing, not capy's any more. Since
 capy `8e1244f` and corosio `232c79f`, destroying an `io_context` unwinds the coroutines suspended on
 it, provided the continuation they are parked with carries `env->owner`: whatever in
 `corosio/runtime.hpp` parks or queues a coroutine has to set it. What that cannot reach is a cycle

CMakeLists.txt

@@ -133,11 +133,11 @@ endif()
 # --------------------------------------------------------------------------------------------------
 
 #
-# API style and I/O runtime, one per build tree: ASIO (completion tokens, on Boost.Asio) or COROSIO
+# The I/O runtime, one per build tree: ASIO (completion tokens, on Boost.Asio) or COROSIO
 # (coroutines only, on capy/corosio). The code tests ANYHTTP_ASIO or ANYHTTP_COROSIO, exactly one of
 # which is 1, from the generated anyhttp/config.hpp. See docs/corosio-port-plan.md.
 #
-set(ANYHTTP_API ASIO CACHE STRING "API style and I/O runtime: ASIO or COROSIO")
+set(ANYHTTP_API ASIO CACHE STRING "I/O runtime: ASIO or COROSIO")
 set_property(CACHE ANYHTTP_API PROPERTY STRINGS ASIO COROSIO)
 if (ANYHTTP_API STREQUAL "ASIO")
     set(ANYHTTP_ASIO 1)
@@ -170,7 +170,7 @@ elseif (ANYHTTP_API STREQUAL "COROSIO")
         GIT_TAG 41d0de18b6e4fa1b397c38a866f2daf52c6c36b5
         EXCLUDE_FROM_ALL)
     FetchContent_MakeAvailable(capy corosio)
-    message(STATUS "Using API style: COROSIO (capy/corosio)")
+    message(STATUS "Using I/O runtime: COROSIO (capy/corosio)")
 else()
     message(FATAL_ERROR "ANYHTTP_API must be ASIO or COROSIO, not '${ANYHTTP_API}'")
 endif()

CMakePresets.json

@@ -42,7 +42,7 @@
         {
             "name": "corosio",
             "hidden": true,
-            "description": "The COROSIO API style, on capy/corosio, which needs OpenSSL",
+            "description": "The COROSIO I/O runtime, on capy/corosio, which needs OpenSSL",
             "inherits": "openssl",
             "cacheVariables": {
                 "ANYHTTP_API": "COROSIO"
@@ -177,7 +177,7 @@
         {
             "name": "build-corosio",
             "displayName": "COROSIO",
-            "description": "The COROSIO API style, see docs/corosio-port-plan.md; capy and corosio are FetchContent'd at pinned SHAs",
+            "description": "The COROSIO I/O runtime, see docs/corosio-port-plan.md; capy and corosio are FetchContent'd at pinned SHAs",
             "inherits": [
                 "clang",
                 "corosio",

README.md

@@ -4,7 +4,7 @@
 [![Documentation](https://img.shields.io/badge/documentation-read-blue)](https://pgit.github.io/anyhttp/)
 [![Coverage](https://img.shields.io/badge/coverage-report-blue)](https://pgit.github.io/anyhttp/coverage/)
 
-Low-level C++23 HTTP client and server library, in one of two API styles, chosen at configure time:
+Low-level C++23 HTTP client and server library, on one of two I/O runtimes, chosen at configure time:
 
 * **ASIO**, on [Asio](https://www.boost.org/doc/libs/release/doc/html/boost_asio.html): every
   operation is an asynchronous operation of Asio and takes any completion token.
@@ -28,7 +28,7 @@ This project started as a hand-written replacement for [nghttp2-asio](https://gi
 
 **<https://pgit.github.io/anyhttp/>** has the [programmer's guide](docs/programmers-guide.rst), which
 is the place to start, the API reference and how it all works:
-[API styles](docs/api-styles.rst), [the end of a body](docs/end-of-body.rst),
+[I/O runtimes](docs/io-runtimes.rst), [the end of a body](docs/end-of-body.rst),
 [concurrent requests](docs/concurrent-requests.rst), [moving to HTTP/3](docs/alt-svc.rst) and
 [the HTTP/2 library](docs/http2-library.rst). [docs/README.md](docs/README.md) says how to make it.
 
@@ -78,7 +78,7 @@ Task<void> do_session(client::Client& client, boost::urls::url url)
 }
 ```
 
-Both API styles have every operation as such a coroutine, that yields a tuple and reports errors
+Both I/O runtimes have every operation as such a coroutine, that yields a tuple and reports errors
 instead of throwing. On ASIO, each is also an asynchronous operation that takes any completion token.
 The whole programs are [hello_server.cpp](docs/examples/hello_server.cpp) and
 [hello_client.cpp](docs/examples/hello_client.cpp).

cmake/config.hpp.in

@@ -1,7 +1,7 @@
 #pragma once
 
 //
-// Generated by CMake from cmake/config.hpp.in: the API style this library was built for (see
+// Generated by CMake from cmake/config.hpp.in: the I/O runtime this library was built for (see
 // anyhttp/runtime.hpp). Code using the library has to see the same one, so it comes with the
 // headers instead of a compile definition.
 //

docs/CMakeLists.txt

@@ -19,7 +19,7 @@ target_link_libraries(hello_client PRIVATE anyhttp)
 #    cmake --build build --target docs
 #
 # Doxygen does not compile the headers, so unlike the snippets the documentation depends on neither
-# the API style of the tree nor on what that style needs. See Doxyfile.in.
+# the I/O runtime of the tree nor on what that runtime needs. See Doxyfile.in.
 #
 find_program(DOXYGEN_EXECUTABLE doxygen DOC "Doxygen, which reads the comments in the headers")
 find_program(SPHINX_EXECUTABLE sphinx-build DOC "Sphinx, which makes the documentation")

docs/Doxyfile.in

@@ -8,7 +8,7 @@ STRIP_FROM_PATH        = @PROJECT_SOURCE_DIR@/include
 
 #
 # The public interface, and nothing else. The comments on the vocabulary of the two runtimes are in
-# asio/runtime.hpp, and corosio/runtime.hpp has the same names in the same order -- api-styles.rst
+# asio/runtime.hpp, and corosio/runtime.hpp has the same names in the same order -- io-runtimes.rst
 # has the table of what they map to. What these headers have for the library's own use is between
 # \cond and \endcond.
 #
@@ -33,7 +33,7 @@ JAVADOC_AUTOBRIEF      = YES
 
 # Doxygen does not compile anything, so what a compiler would get from config.hpp and from Asio's
 # headers is said here -- which is also why the reference takes neither Boost nor a configured
-# runtime, and comes out the same from a tree of either style
+# runtime, and comes out the same from a tree of either runtime
 MACRO_EXPANSION        = YES
 EXPAND_ONLY_PREDEF     = YES
 PREDEFINED             = ANYHTTP_ASIO=1 \

docs/README.md

@@ -31,7 +31,7 @@ apt-get install doxygen python3-sphinx python3-breathe python3-sphinx-rtd-theme
 ```
 
 The target is not part of a regular build, and not there at all where CMake finds no `doxygen` or no
-`sphinx-build`, or was told `ANYHTTP_BUILD_DOCS=OFF`. It can be made from a tree of either API style
+`sphinx-build`, or was told `ANYHTTP_BUILD_DOCS=OFF`. It can be made from a tree of either I/O runtime
 and gives the same result: the comments on the types of the two runtimes are in `asio/runtime.hpp`.
 
 ## Writing it
@@ -43,12 +43,12 @@ Warnings are errors in both tools, so a public symbol without a comment, a param
   What a public header has for the library's own use goes between `\cond internal` and `\endcond`.
 * In a comment, Doxygen takes `http::...` and `https://` for a link, and `\c` of either, or of a word
   that starts with `[`, comes out empty, which neither tool warns about. So write `errors::...`, as
-  that is the spelling that holds in both API styles anyway, and `<tt>[...]</tt>`.
+  that is the spelling that holds on both I/O runtimes anyway, and `<tt>[...]</tt>`.
 * Code in the guide comes from `snippets.cpp`, between a `// [name]` and a `// [name-end]` comment,
   and is shown with `literalinclude`. Only pseudo code is written in place.
 * In a page, names link to the reference with `` :cpp:func:`Session::submit` `` and the like, and
   files of the repository with `` :src:`src/server.cpp` ``.
-* In prose, the libraries are Asio, Corosio and Capy. `ASIO` and `COROSIO` are the API styles, the
+* In prose, the libraries are Asio, Corosio and Capy. `ASIO` and `COROSIO` are the I/O runtimes, the
   values of `ANYHTTP_API`: "an ASIO build". Capy is named only where Capy itself is meant.
 
 [`Doxyfile.in`](Doxyfile.in) configures Doxygen and [`conf.py`](conf.py) configures Sphinx.

docs/benchmark.md

@@ -107,7 +107,7 @@ The tables below use `-c 32` unless they say otherwise.
 ## HTTP/2: nghttpd vs anyhttp
 
 *2026-10-07, measured at `d445bdf`.* nghttpd, which `bench.sh` runs as the HTTP/2 reference,
-serves twice as many requests as either API style: `bench.sh -D 5 -c 32` gave 555k req/s against
+serves twice as many requests as either I/O runtime: `bench.sh -D 5 -c 32` gave 555k req/s against
 284k (ASIO) and 273k (COROSIO), on one thread each, with the same nghttp2 and OpenSSL. This is
 where the time goes; [Done since](#done-since) lists what has been fixed.
 
@@ -144,7 +144,7 @@ instructions may sit in a neighbouring row:
 
 COROSIO, by `perf` share of user time: malloc/free 21%, `std::format` 13%, capy/corosio 12%,
 nghttp2 15%, Boost.URL 5%, Beast 3.5%. The same shape, so the cost lies in the generic code and
-the HTTP/2 backend, not in the API style.
+the HTTP/2 backend, not in the I/O runtime.
 
 ### Where it goes
 
@@ -304,7 +304,7 @@ Scaling across independent contexts, h2 (`h2load -D 6 -c 32 -m 10 -t 4`, req/s):
 - **COROSIO's threaded scheduler is far better than ASIO's shared one,** but a context per thread
   takes most of its lead away. In the independent model COROSIO stays ahead on h1 only and is
   7-22% behind on h2 and h3.
-- **h1 and h3 gain from independent contexts in both styles.** COROSIO h1 drops from 26 to 15.5 µs
+- **h1 and h3 gain from independent contexts on both runtimes.** COROSIO h1 drops from 26 to 15.5 µs
   per request, 4.7 µs of the saving in system time.
 - **COROSIO h2 gets slower** with independent contexts: 6.2 instead of 5.0 µs per request, with
   system time up by half. Load is spread evenly (all threads equally busy), and
@@ -353,7 +353,7 @@ causes, both fixed:
    completed inside the initiating call, which `Response::submit()` always does. HTTP/3's write
    pass, posted by `wake_write()`, then ran first and sent each response head in a packet of its
    own, before the handler could add the body: twice the flushes, packets and ACKs of ASIO. Found
-   by counting events per request in both styles, not by profiling. Fixed by resuming at once,
+   by counting events per request on both runtimes, not by profiling. Fixed by resuming at once,
    after the initiating call has returned (`6c3c50f`): h3 97k → 143k (ASIO 124k), h2 197k → 247k
    (ASIO 207k), h2 cleartext 229k → 276k (ASIO 191k), with `bench.sh -D 3`.