Alexander J. Semenuk
5d5e35fb9b
## Problem Windows builds break with a current local toolchain (Scoop LLVM 22.1.8, CMake 4.4.0, VS 2026), in two independent ways: 1. The build stops at curl's deliberate guard: `#error "no non-blocking method was found/used/set"` in `third-party/curl/lib/nonblock.c`. 2. From the second configure onward, `cmake --build` re-runs CMake in an endless loop (observed 42 consecutive reconfigure cycles in a single build). Likely the same mechanism behind the "endlessly building" VS 2026 note in `docs/setup/dev/vs.md`. ## Root cause 1. `third-party/curl/CMake/CurlTests.c` passes `int *` to `ioctlsocket()`, whose third parameter is `u_long *`. Clang 22 promotes `-Wincompatible-pointer-types` to a hard error in C, so the `HAVE_IOCTLSOCKET_FIONBIO` try_compile silently fails and `curl_config.h` never defines it. Upstream CI does not see this because the windows-2022 runner image ships an older LLVM. GCC 14 promotes the same warning to a hard error, which is very likely the `CurlTests.c.obj` failure reported from MSYS2 in open-goal/jak-project#3551. Upstream curl hit the identical problem with GCC 14 and fixed the probe in curl 8.8.0 (curl/curl#13578). 2. The root CMakeLists copies the build tree's `compile_commands.json` into `<src>/build/` for clangd using `configure_file()`, which registers its input as a configure dependency. CMake rewrites `compile_commands.json` late in every generation, after `CTestTestfile.cmake` and `cmake_install.cmake` (outputs of the same Ninja regen rule), so once the dependency is registered the rule is deterministically dirty and every `ninja` invocation re-runs CMake. A pristine first configure is safe (the file does not exist yet, so the `if(EXISTS ...)` guard skips the copy), which is why the loop looks machine- or IDE-specific. ## Fix 1. Per review, re-vendor `third-party/curl` at the `curl-8_21_0` tag (previously `curl-8_3_0`), which carries the upstream probe fix plus two years of upstream development; `vendor.yaml` updated to match. Adjustments the version jump forced: - curl 8.15 removed the native macOS Secure Transport backend (`CURL_USE_SECTRANSP`), so macOS now builds curl against OpenSSL like Linux. The two macOS workflows install Homebrew `openssl@3` and export `OPENSSL_ROOT_DIR` (keg-only), and the macOS setup docs gained the same two lines. - `CURL_BROTLI` / `CURL_ZSTD` switched to AUTO-detection in curl 8.10; pinned OFF to keep the previous no-compression behavior and avoid silently linking whatever the CI images happen to have. - curl's new top-level `BUILD_EXAMPLES` cache option (default ON) leaked into discord-rpc's identically named option and broke configure at a nonexistent `examples/send-presence` directory; pinned OFF ahead of the third-party subdirectories. The diff is dominated by the mechanical tag-tree swap under `third-party/curl` (linguist-vendored, collapsed in review). The hand-written changes are `CMakeLists.txt`, the two macOS workflows, `docs/setup/system/macos.md`, and `vendor.yaml`. 2. Swap `configure_file()` for `file(COPY ...)`: the same clangd copy with no configure dependency registered. (`file(COPY_FILE ... ONLY_IF_DIFFERENT)` would be cleaner still but requires CMake 3.21, above the declared `cmake_minimum_required(VERSION 3.10)`.) ## Test plan - [x] Fresh `cmake --preset Release-windows-clang` (LLVM 22, no cache seeding) completes and logs `Enabled SSL backends: Schannel`; the FIONBIO probe passes without the previous `#error` - [x] Full Windows Release build from scratch in the branch worktree (all 1422 targets) - [x] goalc-test suite: 1509 passed, 0 failed - [x] Second consecutive configure with `compile_commands.json` present: the regen rule in `build.ninja` has no `compile_commands.json` input; `<src>/build/compile_commands.json` is still refreshed for clangd - [x] Repeated `ninja` invocations after a full build no longer re-run CMake - [x] macOS Intel and ARM CI green (first exercise of the OpenSSL backend switch) --- I work off a self-hosted forge, so this GitHub account is quiet; the configure logs and ninja dirty-node traces from the investigation are available if anyone wants the raw data. (AI-assisted)
3.7 KiB
Vendored
Generated
SSL problems
First, let's establish that we often refer to TLS and SSL interchangeably as SSL here. The current protocol is called TLS, it was called SSL a long time ago.
There are several known reasons why a connection that involves SSL might fail. This is a document that attempts to detail the most common ones and how to mitigate them.
CA certs
CA certs are used to digitally verify the server's certificate. You need a
"ca bundle" for this. See lots of more details on this in the SSLCERTS
document.
CA bundle missing intermediate certificates
When using said CA bundle to verify a server cert, you may experience problems if your CA store does not contain the certificates for the intermediates if the server does not provide them.
The TLS protocol mandates that the intermediate certificates are sent in the handshake, but as browsers have ways to survive or work around such omissions, missing intermediates in TLS handshakes still happen that browser users do not notice.
Browsers work around this problem in two ways: they cache intermediate certificates from previous transfers and some implement the TLS "AIA" extension that lets the client explicitly download such certificates on demand.
Protocol version
Some broken servers fail to support the protocol negotiation properly that SSL servers are supposed to handle. This may cause the connection to fail completely. Sometimes you may need to explicitly select an SSL version to use when connecting to make the connection succeed.
An additional complication can be that modern SSL libraries sometimes are built with support for older SSL and TLS versions disabled.
All versions of SSL and the TLS versions before 1.2 are considered insecure and should be avoided. Use TLS 1.2 or later.
Ciphers
Clients give servers a list of ciphers to select from. If the list does not include any ciphers the server wants/can use, the connection handshake fails.
curl has recently disabled the user of a whole bunch of seriously insecure ciphers from its default set (slightly depending on SSL backend in use).
You may have to explicitly provide an alternative list of ciphers for curl to use to allow the server to use a weak cipher for you.
Note that these weak ciphers are identified as flawed. For example, this includes symmetric ciphers with less than 128-bit keys and RC4.
Schannel in Windows XP is not able to connect to servers that no longer support the legacy handshakes and algorithms used by those versions, so we advise against building curl to use Schannel on really old Windows versions.
Reference: Prohibiting RC4 Cipher Suites
Allow BEAST
BEAST is the name of a TLS 1.0 attack that surfaced 2011. When adding means to mitigate this attack, it turned out that some broken servers out there in the wild did not work properly with the BEAST mitigation in place.
To make such broken servers work, the --ssl-allow-beast option was introduced. Exactly as it sounds, it re-introduces the BEAST vulnerability but on the other hand it allows curl to connect to that kind of strange servers.
Disabling certificate revocation checks
Some SSL backends may do certificate revocation checks (CRL, OCSP, etc) depending on the OS or build configuration. The --ssl-no-revoke option was introduced in 7.44.0 to disable revocation checking but currently is only supported for Schannel (the native Windows SSL library), with an exception in the case of Windows' Untrusted Publishers block list which it seems cannot be bypassed. This option may have broader support to accommodate other SSL backends in the future.
References: