- rely on the new flush to handle blocked sends. No longer
do simulated EAGAIN on (partially) blocked sends with their
need to handle repeats.
- fix some debug handling CURL_SMALLREQSEND env var
- add some assertings in request.c for affirming we do it right
- enhance assertion output in test_16 for easier analysis
Closes#14435
Before this patch `pkg-config`-based detection was ignored, and used
solely as a path hint for native detection.
- fix `pkg_search_module()` result prefix to match what code expects:
`_GSS` (was: `_GSS_PKG`). Update variable that were in sync with old
prefix.
- update the pkg-config codepath to use `_GSS_MODULE_NAME` to detect
GSS flavour. This requires CMake 3.16.
Otherwise fall back to the old method. (The old method doesn't seem to
work anymore (?) as of CMake 3.30.1. Documented
`<prefix>_<modulename>_VERSION` variable is defined, but empty.)
- update the pkg-config codepath to use `_GSS_VERSION` set by CMake.
Resort to the old code when this variable is empty. (The old code
doesn't seem to work anymore (?) as of CMake 3.30.1)
- fix pkg-config codepath to set the documented result variables.
- align native detection variable names with those generated by
`pkg_search_module()` in the pkg-config codepath.
- GHA/macos: enable GSS Heimdal in a cmake job.
Uses the native detection.
- GHA/linux: enable GSS Heimdal in cmake and autotools jobs.
CMake uses `pkg-config`-based detection.
- suppress test 2077 and 2078 results on Linux + Heimdal.
```
FAIL-IGNORED 2077: 'curl --fail --negotiate to unauthenticated service fails' HTTP, HTTP GET, GSS-API
FAIL-IGNORED 2078: 'curl --negotiate should not send empty POST request only' HTTP, HTTP GET, GSS-API
```
Failing with valgrind errors in both autotools and cmake builds:
https://github.com/curl/curl/actions/runs/10282222581/job/28453472068?pr=14430#step:38:3638https://github.com/curl/curl/actions/runs/10282222581/job/28453473398?pr=14430#step:38:7831Closes#14430
```
lib/krb5.c:343:39: warning: cast from 'void **' to 'unsigned char **' increases required alignment from 2 to 8 [-Wcast-align]
(unsigned char **)&_gssresp.value,
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
```
Seen on macOS Intel with Apple clang and brew heimdal 7.8.0_1.
Closes#14433
- use documented flavour of `xcrun` option.
- show SDK version with a dedicated command.
(Sometimes the SDK path is a symlink and doesn't tell the version.
This is not at the moment the case in CI, but handle it anyway.)
- align group header with reality.
Preinstalled vs. installed Homebrew packages can be recognized
by their directory timestamps. Installed ones have a current date.
Closes#14434
This disambiguates the source code being tested. The output format is
the same as when testing out of a git repo, but with no description and
a long hash.
Ref: #14363Closes#14429
(in debug-builds)
Fix implementation in curl using libuv to process parallel transfers.
Add pytest capabilities to run test cases with --test-event.
- fix uv_timer handling to carry correct 'data' pointing to uv context.
- fix uv_loop handling to reap and add transfers when possible
- fix return code when a transfer errored
Closes#14413
- QUIT is not an important FTP command
- curl only sends it "best effort", meaning it might not be sent
- it is a known "flaky" thing in test output because of this
Closes#14404
- make mentioning `-v` on the curl command line increase the
verbosity of the trace output
- related discussion https://github.com/curl/curl/discussions/13810
- make a single -v revert all previous -v+ changes
- make --no-verbose also reset all trace configs
Closes#13977
Follow-up to 9a0cf56471
- increase the buffer to handle 160 characters manpage lines
- add another assert
- if the line buffer gets full, abort
Ideally, we add another step in the build process that makes the build
fail if this long lines are used.
Closes#14422
- sync build-dir/source-dir header path order with autotools, by
including build-dir first, then source-dir.
This prevents out-of-tree builds breaking due to leftover generated
headers in the source tree.
- tests/unit: move `src` ahead of `libtest` in header path, syncing with
autotools.
- stop adding non-existing generated `include` dir to header path.
There are no generated `include` headers and this directory is either
missing in out-of-tree builds or the same as the one already added
globally via the root `CMakeLists.txt`.
- lib: stop adding a duplicate source include directory to the header
path.
It's already added globally via the root `CMakeLists.txt`.
- lib: stop adding the project root to the header path.
- docs/examples: drop internal header paths.
Examples do not and should not use internal headers.
- replace `curl_setup_once.h` in comments with `curl_setup.h`,
the header actually used, and also referred to in autotools comments.
- add comment why we need `src` in include path for `tests/server`.
- add quotes around header directories.
Closes#14416
- make `--with-apple-idn` override libidn2, in sync with cmake and
`lib/curl_setup.h`.
- sync detection function name with cmake.
- limit AppleIDN feature check to Darwin.
(also drop !WinIDN precondition check.)
Follow-up to 8de8fe8c98#14401Closes#14419
Replicate the method used by autotools and cmake to enable libidn2.
This way `lib/curl_setup.h` sets `USE_LIBIDN2` automatically.
Before this patch, `USE_LIBIDN2` was enabled directly, shortcutting
internal logic prioritizing IDN backends.
(This is academic now because `Makefile.mk` no longer supports
other IDN backends. But still useful for clarity.)
Closes#14421
The documented and mandated step has been to not use buildconf but to
invoke 'autoreconf -fi' for four years already.
This change only drops buildconf from the release tarball, it remains
present in git for now.
Follow-up to 85868537d6Closes#14412
Turns out CMake supports numeric comparison with hexadecimal values.
Confirmed in GHA/linux-old with CMake 3.7.2. I could not find
documentation about this, but our CMakeLists.txt already used it before
this patch.
Extend that method to two more comparisons.
Also pad the value in the existing one to 4 digits.
The padding/lowercasing logic when setting `HAVE_WIN32_WINNT` is no
longer required, but keep it anyway for uniform log output.
Follow-up to 2100d9fde2#12044Closes#14409
- extend existing Linux workflow with CMake support.
Including running pytest the first time with CMake.
- cmake: generate `tests/config` and `tests/http/config.ini`.
Required for pytest tests.
Uses basic detection logic. Feel free to take it from here.
Also dump config files in a CI step for debugging purposes.
- cmake: build `tests/http/clients` programs.
- fix portability issues with `tests/http/clients` programs.
Some of them use `getopt()`, which is not supported by MSVC.
Fix the rest to compile in CI (old-mingw-w64, MSVC, Windows).
- GHA/linux: add CMake job matching an existing autotools one.
- GHA/linux: test `-DCURL_LIBCURL_VERSIONED_SYMBOLS=ON`
in the new CMake job.
- reorder testdeps to build server, client tests first and then
libtests and units, to catch errors in the more complex/unique
sources earlier.
- sort list in `tests/http/clients/Makefile.inc`.
Closes#14382
Also use an `#undef` hack for CMake Unity builds to avoid the previously
included `memdebug.h` header messing up the declarations pulled in by
`uv.h`:
```
In file included from ~/curl/bld/src/CMakeFiles/curl.dir/Unity/unity_0_c.c:88:
In file included from ~/curl/src/tool_operate.c:54:
In file included from /usr/local/Cellar/libuv/1.48.0/include/uv.h:71:
In file included from /usr/local/Cellar/libuv/1.48.0/include/uv/unix.h:34:
/Library/Developer/CommandLineTools/SDKs/MacOSX13.1.sdk/usr/include/netdb.h:269:7: error: expected parameter declarator
void freeaddrinfo(struct addrinfo *);
^
~/curl/lib/memdebug.h:167:31: note: expanded from macro 'freeaddrinfo'
curl_dbg_freeaddrinfo(data, __LINE__, __FILE__)
^
```
Follow-up to 38d334e3e1#14298Closes#14399
If a request containing two headers that have equivalent prefixes (ex.
"x-amz-meta-test:test" and "x-amz-meta-test-two:test2") AWS expects the
header with the shorter name to come first. The previous implementation
used `strcmp` on the full header. Using the example, this would result
in a comparison between the ':' and '-' chars and sort
"x-amz-meta-test-two" before "x-amz-meta-test", which produces a
different "StringToSign" than the one calculated by AWS.
Test 1976 verifies
Closes#14370
Bring setting ciphers with WolfSSL in line with other SSL backends,
to make the curl interface more consistent across the backends.
Now the tls1.3 ciphers are set with the --tls13-ciphers option, when
not set the default tls1.3 ciphers are used. The tls1.2 (1.1, 1.0)
ciphers are set with the --ciphers option, when not set the default
tls1.2 ciphers are used. The ciphers available for the connection
are now a union of the tls1.3 and tls1.2 ciphers.
This changes the behaviour for WolfSSL when --ciphers is set, but
--tls13-ciphers is not set. Now the ciphers set with --ciphers
are combined with the default tls1.3 ciphers, whereas before solely
the ciphers of --ciphers were used.
Thus before when no tls1.3 ciphers were specified in --ciphers,
tls1.3 was completely disabled. This might not be what the user
expected, especially as this does not happen with OpenSSL.
Closes#14385
Bring setting ciphers with mbedTLS in line with other SSL backends,
to make the curl interface more consistent across the backends.
Now the tls1.3 ciphers are set with the --tls13-ciphers option, when
not set the default tls1.3 ciphers are used. The tls1.2 (1.1, 1.0)
ciphers are set with the --ciphers option, when not set the default
tls1.2 ciphers are used. The ciphers available for the connection
are now a union of the tls1.3 and tls1.2 ciphers.
This changes the behaviour for mbedTLS when --ciphers is set, but
--tls13-ciphers is not set. Now the ciphers set with --ciphers
are combined with the default tls1.3 ciphers, whereas before solely
the ciphers of --ciphers were used.
Thus before when no tls1.3 ciphers were specified in --ciphers,
tls1.3 was completely disabled. This might not be what the user
expected, especially as this does not happen with OpenSSL.
Closes#14384
add --with-libuv to configure to (optionally) use it in debug-builds to
drive the event-based API
Use curl_multi_socket_action() and friends to drive parallel transfers.
tests/README has brief documentation for this
Closes#14298
