It's causing false-positives with clang-tidy v21, in cases in system
headers (seen in `FD_ISSET()` with macOS SDK). In some cases in
tests/server, there was no distinct source line that was triggering it.
Example:
```
/Applications/Xcode_16.4.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX15.5.sdk/usr/include/sys/_types/_fd_def.h:83:10: error: Potential out of bound access to 'fds_read.fds_bits' with tainted index [clang-analyzer-security.ArrayBound,-warnings-as-errors]
83 | return _p->fds_bits[(unsigned long)_fd / __DARWIN_NFDBITS] & ((__int32_t)(((unsigned long)1) << ((unsigned long)_fd % __DARWIN_NFDBITS)));
| ^
[...]
/Users/runner/work/curl/curl/tests/server/socksd.c:679:5: note: Taking false branch
679 | if(rc < 0) {
| ^
```
Closes#18422
Make the:
- mbedTLS valgrind job finish under 14m, vs 15m before.
- OpenSSL -O3 valgrind job finish in 14m30, vs 16m17.
- OpenSSL libssh2 valgrind job finish in 16m, vs 17m30.
- long valgrind rustls job finish 1 minute earlier, in return
for spending 30s more on the other rustls job.
Keep using autotools for the less slow valgrind job to test this combo.
Closes#18290
Replace autotools with cmake to avoid libtool wrappers that are changing
`LD_LIBRARY_PATH` in a way incompatible with the thread sanitizer.
To fix the output when the sanitizier is finding something:
```
==51718==WARNING: Can't write to symbolizer at fd 7
/usr/bin/llvm-symbolizer-18: /home/runner/work/curl/curl/bld/lib/.libs/libcurl.so.4: no version information available (required by /usr/bin/llvm-symbolizer-18)
/usr/bin/llvm-symbolizer-18: symbol lookup error: /home/runner/openssl/lib/libcrypto.so.3: undefined symbol: __tsan_func_entry
```
Ref: https://github.com/curl/curl/actions/runs/16911402500/job/47913783729#step:39:4466
After:
```
13:50:04.117885 == Info:ThreadSanitizer: thread T1 finished with ignores enabled, created at:
closing connection #0#0 pthread_create <null> (libtests+0x6bc0f) (BuildId: 4fe889446291259934205ac03931c397aa0210d3)
#1 Curl_thread_create /home/runner/work/curl/curl/lib/curl_threads.c:73:6 (libcurl.so.4+0x55a76) (BuildId: cb0f14ba2ad68c9cab0c980d9a5d7a53cc0782da)
#2 async_thrdd_init /home/runner/work/curl/curl/lib/asyn-thrdd.c:500:26 (libcurl.so.4+0x1c153) (BuildId: cb0f14ba2ad68c9cab0c980d9a5d7a53cc0782da)
[...]
```
Ref: https://github.com/curl/curl/actions/runs/16939193922/job/48003405272?pr=18274#step:39:4018
Also:
- disable memory tracker which turned out to be incompatible with
the thread sanitizer and detaching threads.
Ref: #18263 and #curl IRC.
- the job is ~30 seconds faster after this patch.
Reported-by: Stefan Eissing
Bug: https://github.com/curl/curl/pull/18263#issuecomment-3179279440
Follow-up to a2bcec0ee0#14751Closes#18274
In CI we want to ensure that examples build cleanly, but we don't want
to actually run them there. Meaning it's enough to just compile, but not
link them in CI. Saving time up to 2-4x (MSVC), and disk space up
to 1.2GB (or 8-70x).
Add a new cmake target that compiles all examples without linking them
into runnable binaries. Keep a full build for a single example to test
if it links correctly.
Also:
- CI: switch over all `curl-examples` targets to `curl-examples-build`
- GHA/linux-old: build examples in one of the cmake builds.
Result highlights:
Job | Bef. | Bef. | Aft. | Aft. |
:------------------ | ---: | ----: | ---: |----: |
cygwin | 15s | 9MB | 10s | 1MB |
msys | 13s | 8MB | 7s | 1MB |
dl-mingw 15 | 39s | 113M | 34s | 2MB |
dl-mingw 9.5.0 | 49s | 115MB | 42s | 2MB |
dl-mingw 7.3.0 | 19s | 113MB | 14s | 2MB |
dl-mingw 6.4.0 | 9s | 12MB | 7s | 4MB |
Linux cross | 19s | 28MB | 19s | 2MB |
MSVC UWP | 65s | 374MB | 9s | 17MB |
MSVC x64 | 22s | 846MB | 9s | 17MB |
VS2010 | 48s | 105MB | 15s | 9MB |
VS2022 clang-cl | 195s | 1.2GB | 51s | 20MB |
iOS Xcode | 8s | | 5s | |
macOS LibreSSL | 16s | | 11s | |
Linux aws-lc | 3s | | 1s | |
Follow-up to dda251ef10#18232Closes#18209
Also:
- GHA/linux, macos: test `install` with autotools too.
- GHA/linux, macos: enable fish and zsh completion in an autotools job.
- scripts: drop dynamic perl checks redundant after this patch.
- scripts: drop two interim variables.
- autotools: make `build-certs` target check perl first.
- autotools: replace `$(PERL)` with `@PERL@` to match the pattern used
in most automake scripts. For consistency. This makes `PERL` be
defined at configure-time, as opposed to make-time, for these cases
now.
Closes#18141
- add startup check for 'danted' to avoid fails on low cpu
- rename 'sockd' to 'danted' everywhere to clarify what we use
- add proper defaults for 'danted' for debian
- install 'dante-server' in pytest ci runs
Closes#18075
Fixing:
- HTTPS-RR builds with c-ares and Linux MUSL.
- curl-for-win minimal builds with Linux MUSL.
It should fix all other kinds of entaglement between curl's redefintions
of system symbols and system (or 3rd-party) headers sensitive to that.
It also syncs memory override behavior between unity & non-unity builds,
thus reducing build variations.
The idea is to define and declare everything once in `curl_setup.h`,
without overriding any system symbols with curl ones yet. Then, like
before this patch, override them, if necessary, in each source file via
`curl_memory.h` and `memdebug.h`, after including system headers.
To ensure a clean slate with no overrides at the beginning of each
source file, reset all of them unconditionally at the end of
`curl_setup.h`, by including `curl_mem_undef.h`. (This assumes
`curl_setup.h` is always included first, which is already the case
throughout the codebase.)
`curl_mem_undef.h` can also be included explicitly wherever overrides
are causing problems. E.g. in tests which use unity-style builds and
a previously included `curl_memory.h`/`memdebug.h` can be spilling into
other source files.
The simplified role of the two override headers:
- `curl_memory.h`: overrides system memory allocator functions to
libcurl ones, when memory tracing (aka `CURLDEBUG`) is disabled.
- `memdebug.h`: overrides system memory allocator and some other
functions to curl debug functions, when memory tracing is enabled.
Changed made in this patch, step-by-step:
- curl_memory.h: move allocator typedefs and protos to `curl_setup.h`.
- memdebug.h: move `ALLOC_*` macros to `curl_setup.h`.
- memdebug.h: move allocator protos to `curl_setup.h`.
- memdebug.h: move `Curl_safefree()` macro to `curl_setup.h`.
(it's a regular macro, with a one-time, global, definition.)
- curl_memory.h: move system symbol undefs to a new, separate header:
`curl_mem_undef.h`.
- curl_setup.h: include `curl_mem_undef.h` at the end, unconditionally,
to reset system symbol macros after each inclusion.
- handle `sclose()` and `fake_sclose()` in `curl_setup.h`. They are not
system symbols, a one-time definition does the job.
Also:
- GHA/linux: enable unity mode for the HTTP-RR c-ares MUSL job.
Follow-up to 17ab4d62e6#16413
That said, I'd still find it better to avoid redefining system macros.
To communicate clearly the fact that they are not the original system
calls and they do behave differently. And, it would allow dropping the
undef/redef dance in each source file, and maintaining the logic with
it. The "last #include files should be in this order" comments in each
source would also become unnecessary. Also the trick of using
`(func)` (or interim macros) to call the non-overridden function where
required. This method works for printf and most everything else already.
For `_tcsdup`, socket and fopen functions this could work without
disturbing the codebase much.
Ref: #16428 (clean reboot of)
Closes#17827
- GHA/windows: disable building certs in the MSVC job that's not running
tests. Saves 4-5 seconds for MSVC, makes logs shorter for the rests.
- GHA/linux: build tests in two more jobs (LTO, CM Rustls), 5s each.
- GHA/linux: skip 'install test prereqs' for `skiprun` jobs.
(there were no such jobs before this patch.)
Closes#18034
- runtests: fix `codeset-utf8` feature detection. Before this patch it
detected if the calling environment had UTF-8 enabled. If not, UTF-8
tests were all skipped. After this patch, it detects if UTF-8 is
supported by the calling environment regardless of what's currently
enabled.
Follow-up to 0b70b23ef4#15039
- GHA/linux: sync `codeset-test` to also reset `LC_CTYPE` and
`LC_NUMBER`. To give it more spin.
Follow-up to c221c0ee59#17938
- GHA/macos: fix to actually enable `codeset-test`. Also set `LC_ALL`,
which seems necessary to trigger issues.
Follow-up to c221c0ee59#17938
- tests/data: replace `LC_CTYPE` env with `LC_ALL` in all tests
requiring a locale. Also to avoid potential issues with a blank or
unset `LC_ALL`, as seen earlier. And to ensure that the override works
on all platforms (as tested in CI.)
Slight downside is that this now resets the language/culture to `C`.
Ref: b4c9982382#4743
Ref: 23208e330a#4738
- replace `en_US.UTF-8` with `C.UTF-8` to be language/culture-agnostic.
- TEST-SUITE.md: drop `UTF-8` as a requirement for tests.
Tests shall work (or least be skipped) without UTF-8 support.
Tests requiring UTF-8 locale:
165, 962, 963, 964, 965, 966, 967, 1448, 1560, 2046, 2047
Tests requiring UTF-8 locale, but passing without one anyway:
955, 956, 957, 958, 959, 960, 961, 968, 1034, 1035
Spec 1997: https://pubs.opengroup.org/onlinepubs/7908799/xbd/envvar.html
Spec 2008: https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html
Ref: c221c0ee59#17938
Ref: 7cf8414fab
Ref: 4c140a5628
Ref: 28faaacee2#2436
Ref: ecd1d020abCloses#17988
To fix running test 1560 when `LC_ALL` is set to something unexpected
(e.g. `C`). Also syncing it with the rest of tests.
Also:
- GHA/linux: enable `libidn2` in more jobs.
Also to enable test 1560 reproducing this issue in more jobs.
- GHA/linux: run tests with `LC_ALL=C` in one of the jobs.
- GHA/linux: switch to the non-deprecated package name for libidn2.
- GHA/macos: run tests with non-default locale settings in one job.
- GHA/macos: enable AppleIDN in that job.
Ref: https://github.com/curl/curl/pull/17933#issuecomment-3074582840
Follow-up to f27262b179#10196Closes#17938
To fix test 1308 in MultiSSL builds.
Failure was caused by the random number generator virtual function being
NULL, instead of pointing to the implementation in the runtime-selected
TLS backend. This could happen in MultiSSL builds when a functionality
was asking for a random number without triggering a VTLS function table
initialization first. Such functionality is MIME, or form data via MIME.
The reason CI did not catch it in an earlier MultiSSL GHA/windows job,
is that it was a debug-enabled one. In debug-enabled builds the test
runner was overriding the random number generator for all tests.
Fixed this by moving the override to the tests requiring it, via
1fcf22585f#17971, enabling debug builds
to catch this issue.
Enable MultiSSL in two CI jobs, to verify this patch.
Fixing:
```
test 1308...[formpost tests]
libtests returned 44, when expecting 0
1308: exit FAILED
[...]
=== Start of file stderr1308
URL: log/3/test-1308
tests/libtest/lib1308.c:70 Assertion 'res == 0' FAILED: curl_formget returned error
tests/libtest/lib1308.c:72 Assertion 'total_size == 518' FAILED: curl_formget got wrong size back
tests/libtest/lib1308.c:88 Assertion 'res == 0' FAILED: curl_formget returned error
tests/libtest/lib1308.c:89 Assertion 'total_size == 899' FAILED: curl_formget got wrong size back
```
Ref: https://github.com/curl/curl/actions/runs/16387693424/job/46309536359?pr=17963#step:16:2515
Bug: https://github.com/curl/curl/pull/17963#issuecomment-3092282057Closes#17970
- circleci: pipe to tar.
- use long options uniformly.
- sync option order.
- set timeout where missing.
- set retry where missing.
- set `--retry-connrefused` where missing.
- set `--disable` where missing.
- lower 999s timeouts to 120s.
Closes#17851
Replace the `libstubgss.so`-based overload solution with one built into
libcurl at compile-time.
The previous, `LD_PRELOAD`-based, solution was non-portable, allowlisted
for Linux, BSD and Solaris. It also required non-debug builds, which
turned out to be an accidental condition:
7d342c723c. It also required a curl tool
built against a shared libcurl. Detecting this condition wasn't always
accurate, e.g. with certain cmake configurations.
The overload solution also didn't work on macOS, though it theoretically
should have:
- #17653
- #2394
Experiments on making the overload solution work in more envs:
- #17759
That revealed that it also did not work on NetBSD, in CI.
The replacement solution is overloading the necessary GSS-API functions
for test 2056 and 2057 at compile time. It requires a debug-enabled curl
build (due to its insecure nature).
This makes these tests run on all platforms. Including most GSS jobs in
CI, that are running tests. (the exception is old-linux, non-debug jobs,
where it felt overkill to enable debug for this.)
The refactored GSS stub code needs to overload less than before because
it's free to use the official GSS API. (This didn't work with
the overload solution on Alpine for example). It can also use libcurl
functions, allowing to replace `snprintf()` with `msnprintf()`.
OS/400 is also overloading GSS API functions. I haven't tested how this
works after this PR. In theory it should, because this PR doesn't rely
on preprocessor overrides.
Note that for future GSS tests, it may be necessary to stub these GSS
API functions: `gss_inquire_context()`, `gss_unwrap()`, `gss_wrap()`.
They are on codepaths not (yet) touched by tests.
Also:
- stub-gss: check for token buffer overrun.
- stub-gss: replace size macros with `sizeof()`.
- GHA: enable debug for some jobs with GSS.
- GHA/linux: ignore results for 2056 and 2057 in the valgrind job.
They leak the same way as seen with 2077 and 2078.
Ref: 7020ba7979#17462
Ref: 146759716c#14430
- GHA/linux: fix to ignore `gss_import_name()` leaks in valgrind builds.
only.
- lib/vauth/krb5_gssapi: reduce variable scope.
- lib/vauth/spnego_gssapi: reduce variable scope.
- tests/libtest: drop code and build logic dealing with `libstubgss`.
- runtests:
- drop `ld_preload` feature.
- drop special handling of `LD_PRELOAD` env in tests.
- drop logic dealing with shared curl tool detection.
- drop `LD_PRELOAD` envs from tests.
Follow-up to 56d949d31a#1687Closes#17752
Job `name:` now mandatory in zizmor.
Also:
- enclose `name:` values in single-quotes, for uniformity.
- drop `name: checkout` where set, for uniformity.
- dist: also install with cmake.
- dist: replace `make` with `cmake --build` for cmake.
(to make this make-tool agnostic)
- appveyor-status: double-quote shell arguments.
- tweak existing names to be shorter, to sync terms and style across
jobs and steps.
Ref: https://github.com/zizmorcore/zizmor/releases/tag/v1.10.0Closes#17773
Replace existing `mk-unity.pl` `--embed` workaround with running
`clang-tidy` manually on individual test source instead. This aligns
with how clang-tidy works and removes `mk-unity.pl` from the solution.
Also:
- mqttd: fix potentially uninitialized buffer by zero filling it.
```
tests/server/mqttd.c:484:41: error: The left operand of '<<' is a garbage value
[clang-analyzer-core.UndefinedBinaryOperatorResult,-warnings-as-errors]
484 | payload_len = (size_t)(buffer[10] << 8) | buffer[11];
| ^
[...]
tests/server/mqttd.c:606:45: error: The left operand of '<<' is a garbage value
[clang-analyzer-core.UndefinedBinaryOperatorResult,-warnings-as-errors]
606 | topiclen = (size_t)(buffer[1 + bytes] << 8) | buffer[2 + bytes];
| ^
```
- sockfilt: fix potential out-of-bound pointer:
```
tests/server/sockfilt.c:1128:33: error: The 2nd argument to 'send' is a buffer
with size 17010 but should be a buffer with size equal to or greater than
the value of the 3rd argument (which is 18446744073709551615)
[clang-analyzer-unix.StdCLibraryFunctions,-warnings-as-errors]
1128 | ssize_t bytes_written = swrite(sockfd, buffer, buffer_len);
| ^
```
- clang-tidy: suppress bogus `bzero()` warnings that happens
inside the notorious `FD_ZERO()` macros, on macOS.
Ref: https://github.com/curl/curl/pull/17680#issuecomment-2991730158Closes#17705
Fix building with `--disable-digest-auth`. Make test cases run in such
a build. Add such a build to our CI jobs.
Reported-by: Tristan Perrault
Fixes#17612Closes#17614
Make test bundles the default. Drop non-bundle build mode.
Also do all the optimizations and tidy-ups this allows, simpler builds,
less bundle exceptions, streamlined build mechanics.
Also rework the init/deinit macro magic for unit tests. The new method
allows using unique init/deinit function names, and calling them with
arguments. This is in turn makes it possible to reduce the use of global
variables.
Note this drop existing build options `-DCURL_TEST_BUNDLES=` from cmake
and `--enable-test-bundles` / `--disable-test-bundles` from autotools.
Also:
- rename test entry functions to have unique names: `test_<testname>`
This removes the last exception that was handled in the generator.
- fix `make dist` to not miss test sources with test bundles enabled.
- sync and merge `tests/mk-bundle.pl` into `scripts/mk-unity.pl`.
- mk-unity.pl: add `--embed` option and use it when `CURL_CLANG_TIDY=ON`
to ensure that `clang-tidy` does not miss external test C sources.
(because `clang-tidy` ignores code that's #included.)
- tests/unit: drop no-op setup/stop functions.
- tests: reduce symbol scopes, global macros, other fixes and tidy-ups.
- tool1621: fix to run, also fix it to pass.
- sockfilt: fix Windows compiler warning in certain unity include order,
by explicitly including `warnless.h`.
Follow-up to 6897aeb105#17468Closes#17590
These libraries do not support TLS 1.3 and have been marked for removal
for over a year. We want to help users select a TLS dependency that is
future-proof and reliable, and not supporting TLS 1.3 in 2025 does not
infer confidence. Users who build libcurl are likely to be served better
and get something more future-proof with a TLS library that supports
1.3.
Closes#16677
zizmor 1.9.0 effectively bans using GHA macros within shell script
snippets. Rework them to pass GHA macro values via envs and use those
envs in shell snippets. `${{ env.* }}` macros could be converted
to native env references after making their names shell-compatible.
Envs and shell commands can no longer be used in GHA macro values. Most
cases could be fixed by using literals. Passing quoted values with
spaces combined with other args also doesn't work anymore. This was
replaced by passing them separately.
Despite the initial complications, avoiding GHA macros in scripts does
seems to make the CI code reasonable cleaner. It also makes it possible
to analyze the scripts with shellcheck, finding subtle issues that went
unnoticed till now.
Also:
- un-ignore and fix three existing zizmor `template-injection` issues.
- add script to extract and shellcheck all shell code stored within GHA
and Circle CI YAML files.
- add CI job to run this script.
- fix shellcheck issues discovered.
- fix minor differences between cmake and autotools FreeBSD jobs.
- merge cmake/autotools FreeBSD jobs to avoid developing unwanted
differences again.
- fix/sync quote use across shell code.
- replace `$HOME` with `~` or literal where it made sense.
- replace most `brew --prefix` with literals.
- move all curl install tests to the `curl-install*` prefix.
- add missing curl install tests to cygwin/msys/mingw/*bsd.
- pipe to `tar` instead of storing downloads on disk.
- drop unnecessary `PKG_CONFIG_PATH` when building nghttp3.
Ref: https://github.com/curl/curl/actions/runs/15461461371/job/43523546041
Ref: https://github.com/zizmorcore/zizmor/releases/tag/v1.9.0
Follow-up to e522f47986#17278Closes#17537
- test1510: it was marked flaky, then disabled. Keep the flaky flag
and allow it to run again.
Ref: ef02da3156
Ref: 0110cced2a
- GHA/linux: extend information about the heimdal valgrind failures.
- GHA/macos: unignore 2037 2041 in torture tests. They no longer
appear flaky.
Follow-up to 4b791dca37#14667Closes#17462
Reduce Curl_ossl_ctx_init() complexity by splitting it up into
sub functions.
While splitting if ECH, add pytest fixed for AWS-LC and enable
it in CI.
Closes#17404
To avoid redundant work in CI and to avoid a single checksrc issue make
all autotools jobs fail. After this patch checksrc issues make fail
the checksrc job, the `dist / verify-out-of-tree-autotools-debug`,
`dist / maketgz-and-verify-in-tree` jobs and the fuzzer job (if run).
Of these, the `dist` jobs replicate local builds, also testing the build
logic.
Also add a script to check the complete local repository, optionally
with the build tree to verify generated C files.
Also:
- automatically run checksrc in subdirectories having a `checksrc`
target. (examples, OS400, tests http/client, unit and tunit)
- tests/libtest: make sure to run `checksrc` on generated `lib1521.c`.
(requires in-tree autotools build.)
- tests: run `checksrc` on targets also for non-`DEBUGBUILD`
builds. It ensures to check `lib1521.c` in CI via job
`dist / maketgz-and-verify-in-tree`.
- src: drop redundant `$(builddir)` in autotools builds.
- scripts: add `checksrc-all.sh` script to check all C sources and
the build directory as an option.
- use the above from CI, also make it verify all generated sources.
- silence `checksrc` issues in generated C sources.
- checksrc: add `-v` option to enable verbose mode.
- checksrc: make verbose mode show checked filename and fix to only
return error on failure.
- make sure that generated C files pass `checksrc`.
Assisted-by: Daniel Stenberg
Closes#17376
