- enable quictls in autotools and cmake jobs. autotools requires
a workaround due to wrong libpath in the quictls pkg-config.
nghttp3 is offered by Homebrew, but not ngtcp2, to enable H3.
- install `libnghttp2` rather than `nghttp2`.
`libnghttp2` is preinstalled and smaller. It also avoids detecting
`nghttpx`, which confuses `pytest`.
- limit `brew unlink openssl` to libressl/quictls jobs.
Closes#16517
This now makes sure to trim off exact matches for curl symbols and long
curl commanad line options instead of using pattern matching as before.
This should catch typoed names (that still follow the pattern) better.
The cleanspell.pl script is no longer used. cleancmd.pl is used for all
markdown files.
Closes#16504
- replace `--parallel <n>` and `-j<n>` for individual commands with
`MAKEFLAGS`, for jobs not yet doing it.
This enables parallel builds in distcheck / maketgz-and-verify-in-tree,
where `-j` option was missing.
- add `--parallel` for iOS Xcode job for improved performance.
- drop redundant `-j5` for Android jobs.
- drop stray `cmake --config` options from single-target jobs (cygwin,
msys/mingw, dl-mingw, non-native). Drop redundant
`CMAKE_RUNTIME_OUTPUT_DIRECTORY_*` settings too.
- GHA/windows: add timeout for package install steps where missing.
- GHA/non-native: specify target type explicitly for iOS cmake jobs.
Xcode default was already Debug, single-target default was generic,
now it's Release, with unity batch to keep it fast.
`MAKEFLAGS` is necessary for autotools jobs and CMake jobs using
the default (GNU Make) generator. It's ignored by Ninja and other tools.
`cmake --parallel` is still necessary for jobs with Visual Studio or
Xcode generators. Parallelism is 5 for GHA Linux and Windows runners,
4 for macOS, 3 for VMs, 2 for AppVeyor.
Closes#16502
- GHA/windows/WinCE:
- set `-O3 -DNDEBUG` C flags manually for the CMake mingw32ce build.
CMake doesn't recognize the platform and fails to add them. To match
autotools (using `-O2`), and hit similar compiler warnings.
- enable parallel builds for cmake.
- tune parallelism for cmake using unity batches.
- tune parallelism for autotools.
Follow-up to 2a292c3984#15975
- tests: fix potentially uninitialized value in `readline()` in
`getpart.c`. Detected by gcc 4.4.0 `-O2` (Windows CE) jobs:
```
tests/server/getpart.c: In function 'getpart':
tests/server/getpart.c:298: error: 'datalen' may be used uninitialized in this function
```
Ref: https://github.com/curl/curl/actions/runs/13522595237/job/37785147505?pr=16476#step:11:25
Follow-up to 592880a3ca
- vtls_scache: rework returning pointer to avoid compiler warning seen
with `-O3` gcc 4.4.0 builds (Windows CE/schannel):
```
lib/vtls/schannel.c: In function 'schannel_connect_step1':
lib/vtls/vtls_scache.c:975: error: dereferencing pointer 'old_cred.4474' does break strict-aliasing rules
lib/vtls/vtls_scache.c:985: error: dereferencing pointer 'old_cred.4474' does break strict-aliasing rules
lib/vtls/schannel.c:959: note: initialized from here
```
Ref: https://github.com/curl/curl/actions/runs/13523868335/job/37789610845#step:9:25
Follow-up to fa0ccd9f1f#15774Closes#16476
`./configure` mingw32ce builds enable C99 mode automatically, that
triggers compiler warnings in gcc 4.4.0. We initially worked it around
in CI by suppressing the detection of C99 with `ac_cv_prog_cc_c99=no`.
Replace it with automatically silencing the bogus warnings in C99 mode,
for all build systems:
```
lib/ftp.c: In function 'Curl_GetFTPResponse':
lib/ftp.c:726: error: format '%zd' expects type 'signed size_t', but argument 4 has type 'ssize_t'
lib/ws.c: In function 'ws_dec_pass_payload':
lib/ws.c:304: error: format '%zd' expects type 'signed size_t', but argument 3 has type 'ssize_t'
lib/ws.c: In function 'ws_enc_write_head':
lib/ws.c:581: error: format '%zd' expects type 'signed size_t', but argument 3 has type 'long int'
lib/vtls/schannel.c: In function 'schannel_connect_step1':
lib/vtls/schannel.c:1122: error: format '%zd' expects type 'signed size_t', but argument 3 has type 'ssize_t'
lib/vtls/schannel.c: In function 'schannel_connect_step2':
lib/vtls/schannel.c:1311: error: format '%zd' expects type 'signed size_t', but argument 3 has type 'ssize_t'
lib/vtls/schannel.c: In function 'schannel_send':
lib/vtls/schannel.c:1793: error: format '%zd' expects type 'signed size_t', but argument 3 has type 'ssize_t'
lib/vtls/schannel.c:1810: error: format '%zd' expects type 'signed size_t', but argument 3 has type 'ssize_t'
lib/vtls/schannel.c: In function 'schannel_shutdown':
lib/vtls/schannel.c:2286: error: format '%zd' expects type 'signed size_t', but argument 4 has type 'ssize_t'
lib/vtls/vtls.c: In function 'ssl_cf_recv':
lib/vtls/vtls.c:1422: error: format '%zd' expects type 'signed size_t', but argument 5 has type 'ssize_t'
```
Ref: https://github.com/curl/curl/actions/runs/13533841306/job/37821720902?pr=16492#step:9:20
Also: simplify Windows CE job configuration in GHA/windows.
Follow-up to 2a292c3984#15975Closes#16492
This PR updates the CMake build/install docs in `docs/INSTALL-CMAKE.md`,
in particular focusing on the use of libcurl from CMake using
`find_package` as well as the newly added features/protocols support via
using `COMPONENTS` or `OPTIONAL_COMPONENTS` with `find_package`.
See #15854 for initial discussion and the corresponding PR #15858 that
was merged.
Some additional best-practices notes are added, for example:
* Encouraging building out-of-source
* Using `--config` with `cmake --build` for multi-config CMake
generators, not `CMAKE_BUILD_TYPE`
We also add a CURL CMake-specific tip on using `CMAKE_INSTALL_PREFIX`
during configure time to set the install prefix, not using `--prefix`
when running `cmake --install` so `curl-config` output is consistent.
Closes#16329
- appveyor: restore VS2008 job, after fixing its issues.
Enable OpenSSL in it. It takes 1 minute.
Follow-up to 9b0467b169#16453
Follow-up to edfa537100#16456
- appveyor: make a copy of OpenSSL DLLs to have them picked up as an
artifact (disabled by default) to aid local tests.
- appveyor: dump CMake configuration logs on failure.
- appveyor: tidy up job parameter defaults.
- GHA/windows: add pre-fill check option for dl-mingw jobs.
- GHA/windows: fix pre-fill check option for MSYS jobs by installing
`diffutils`.
Follow-up to e7adf3e837#15841
- GHA/windows: de-duplicate to `PATH` commands for Cygwin.
- GHA/windows: drop `$SYSTEMROOT/System32` from `PATH` for Cygwin
configure. It's not needed.
Follow-up to 36fd2dd6ee#13599
- list `.pdb` files in curl version step for MSVC.
Ref: #16439
Cherry-picked from #16394Closes#16458
Support multi-target cmake builds via `CURL_DIRSUFFIX` env. For example:
`export CURL_DIRSUFFIX=Debug/`.
Multi-target generators place their output to `src/<subdir>/`,
`lib/<subdir>/`, `tests/server/<subdir>`, `tests/libtest/<subdir>` and
`tests/unit/<subdir>/` by default. Before this patch, `runtests.pl`
couldn't run on such builds because it expected the binaries under the
their `<subdir>`-less directories. This patch allows to set such subdir
and make `runtests.pl` find the binaries. In CI we use multi-target
builds with tests for MSVC. It also helps Xcode-generator builds, though
in CI we don't have such job running tests.
There may be better solutions to configure this, but passing a custom
value to `runtests.pl` including its subprocesses is somewhat tricky.
The reason the configuration value expects the slash at the end is
because MSYS is automagically expanding the env to a (wrong) absolute
path if the slash is in the front.
Also:
- drop the `-DCMAKE_RUNTIME_OUTPUT_DIRECTORY_*=` workaround from CI.
- replace `resolve` references in tests with a new `%RESOLVE` variable.
It didn't use a filename extension before. After this patch it uses
`exe_ext('TOOL')`. I'm not sure if this is the correct choice vs.
`exe_ext('SRV')`.
- fix `-c` option format in manual.
- fix some whitespace.
Note, in CI we still tweak `CMAKE_RUNTIME_OUTPUT_DIRECTORY_*` in jobs
which share steps between `./configure` and cmake. It's easier that way.
Ref: #15000
Cherry-picked from #16394Closes#16452
When running curl event based, connect attempts stalled as the 'done'
check was using the wrong state in gnutls.
Add event based pytest runs to all http3 jobs and the openssl and
mbedtls ones on linux.
Closes#16423
Make it possible to build curl for Windows CE using the CeGCC toolchain.
With both CMake and autotools, including tests and examples, also in CI.
The build configuration is the default one with Schannel enabled. No
3rd-party dependencies have been tested.
Also revive old code to make Schannel build with Windows CE, including
certificate verification.
Builds have been throughougly tested. But, I've made no functional tests
for this PR. Some parts (esp. file operations, like truncate and seek)
are stubbed out and likely broken as a result. Test servers build, but
they do not work on Windows CE. This patch substitutes `fstat()` calls
with `stat()`, which operate on filenames, not file handles. This may or
may not work and/or may not be secure.
About CeGCC: I used the latest available macOS binary build v0.59.1
r1397 from 2009, in native `mingw32ce` build mode. CeGCC is in effect
MinGW + GCC 4.4.0 + old/classic-mingw Windows headers. It targets
Windows CE v3.0 according to its `_WIN32_WCE` value. It means this PR
restores portions of old/classic-mingw support. It makes the Windows CE
codepath compatible with GCC 4.4.0. It also adds workaround for CMake,
which cannot identify and configure this toolchain out of the box.
Notes:
- CMake doesn't recognize CeGCC/mingw32ce, necessitating tricks as seen
with Amiga and MS-DOS.
- CMake doesn't set `MINGW` for mingw32ce. Set it and `MINGW32CE`
manually as a helper variable, in addition to `WINCE` which CMake sets
based on `CMAKE_SYSTEM_NAME`.
- CMake fails to create an implib for `libcurl.dll`, due to not
recognizing the platform as a Windowsy one. This patch adds the
necessary workaround to make it work.
- headers shipping with CeGCC miss some things curl needs for Schannel
support. Fixed by restoring and renovating code previously deleted
old-mingw code.
- it's sometime non-trivial to figure out if a fallout is WinCE,
mingw32ce, old-mingw, or GCC version-specific.
- WinCE is always Unicode. With exceptions: no `wmain`,
`GetProcAddress()`.
- `_fileno()` is said to convert from `FILE *` to `void *` which is
a Win32 file `HANDLE`. (This patch doesn't use this, but with further
effort it probably could be.)
https://stackoverflow.com/questions/3989545/how-do-i-get-the-file-handle-from-the-fopen-file-structure
- WinCE has no signals, current directory, stdio/CRT file handles, no
`_get_osfhandle()`, no `errno`, no `errno.h`. Some of this stuff is
standard C89, yet missing from this platform. Microsoft expects
Windows CE apps to use Win32 file API and `FILE *` exclusively.
- revived CeGCC here (not tested for this PR):
https://building.enlyze.com/posts/a-new-windows-ce-x86-compiler-in-2024/
On `UNDER_CE` vs. `_WIN32_WCE`: (This patch settled on `UNDER_CE`)
- A custom VS2008 WinCE toolchain does not set any of these.
The compiler binaries don't contain these strings, and has no compiler
option for targeting WinCE, hinting that a vanilla toolchain isn't
setting any of them either.
- `UNDER_CE` is automatically defined by the CeGCC compiler.
https://cegcc.sourceforge.net/docs/details.html
- `UNDER_CE` is similar to `_WIN32`, except it's not set automatically
by all compilers. It's not supposed to have any value, like a version.
(Though e.g. OpenSSL sets it to a version)
- `_WIN32_WCE` is the CE counterpart of the non-CE `_WIN32_WINNT` macro.
That does return the targeted Windows CE version.
- `_WIN32_WCE` is not defined by compilers, and relies on a header
setting it to a default, or the build to set it to the desired target
version. This is also how `_WIN32_WINNT` works.
- `_WIN32_WCE` default is set by `windef.h` in CeGCC.
- `_WIN32_WCE` isn't set to a default by MSVC Windows CE headers (the
ones I checked at least).
- CMake sets `_WIN32_WCE=<ver>`, `UNDER_CE`, `WINCE` for MSVC WinCE.
- `_WIN32_WCE` seems more popular in other projects, including CeGCC
itself. `zlib` is a notable exception amongst curl dependencies,
which uses `UNDER_CE`.
- Since `_WIN32_WCE` needs "certain" headers to have it defined, it's
undefined depending on headers included beforehand.
- `curl/curl.h` re-uses `_WIN32_WCE`'s as a self-guard, relying on
its not-(necessarily)-defined-by-default property:
25b445e479/include/curl/curl.h (L77)
Toolchain downloads:
- Windows:
https://downloads.sourceforge.net/cegcc/cegcc/0.59.1/cegcc_mingw32ce_cygwin1.7_r1399.tar.bz2
- macOS Intel:
https://downloads.sourceforge.net/cegcc/cegcc/0.59.1/cegcc_mingw32ce_snowleopard_r1397.tar.bz2Closes#15975
They use Visual Studio generators, which are multi-target.
The build command does the Release/Debug selection via `--config`.
Also:
- appveyor: drop unnecessary conditional for 3 options.
To sync with GHA.
- appveyor: drop unused `-DCMAKE_INSTALL_PREFIX=`.
To sync with GHA.
- sync cmake option order between GHA and appveyor.
Closes#16372
- replace `add_compile_options()`, `add_definitions()` with directory
properties. To harmonize this across all scripts. The new commands are
verbose, but describe better how they work. The syntax is also closer
to setting target properties, helps grepping.
- prefer `CMAKE_INSTALL_PREFIX` over `--prefix` (in tests, CI).
- tidy up cmake invocations.
- formatting.
Closes#16238
TL;DR: Save 10 minutes of CI time for GHA/macos jobs using pre-fills and
add pre-fill verification for Apple and Windows. Also restores Xcode job
and saves 1.5-10 minutes configuring iOS jobs.
Pre-filling feature detection results can bring down the CMake configure
step to ~5 seconds on most GHA runners, ~10 seconds in slow envs like
Cygwin/MSYS2.
The potential savings per job are:
- 5-40 (average 19) seconds on GHA/macos (33 jobs)
- ~10 seconds on GHA for iOS GNU Makefile (1 job)
- 1.5-10 minutes on GHA for iOS Xcode generator (1 job)
- 10 seconds on GHA/linux with native Ubuntu (12 jobs)
- 40 seconds for Cygwin/MSYS2 (2 jobs)
- 5-10 seconds for virtualized BSDs, native CPU (3 jobs)
- ~60 seconds for virtualized BSDs, emulated CPU (1 job)
On native Windows pre-filling has been in place for a long time and
saving 8 minutes (VS2019-VS2015) to 1.5-2 minutes (VS2022), 3 minutes
(VS2022 UWP), and 30-60 seconds (MinGW), per CI job.
The downside is that detection results need to be manually collected and
filtered to those that universally apply to all platforms that they are
enabled on. Another downside is that by using a cache, we're not running
the actual detections, and thus won't catch regressions in them. It
means we must make sure that the cache is solid and matches with actual
detections results. An upside is that it gives a rough overview of which
features are available on which platforms. Another upside is pre-filled
values do work for feature detections skipped for cross-builds, e.g.
`HAVE_WRITABLE_ARGV`.
This PR adds a pre-fill cache that supports all Unixes (except OmniOS)
used in CI, and makes it usable with an internal option. It also enables
it for GHA/macos CI jobs, where the maximum savings are. And also for
the two iOS [1] and two Cygwin/MSYS2 jobs. The latters don't have
pre-fill checks and we can drop them if they turn into a hassle.
Saving:
- 10 minutes of CI time per GHA/macos workflow run. [2]
- ~80 seconds per GHA/windows workflow run with Cygwin/MSYS2.
(offsetting the cost of pre-fill verifications)
- 1.5-10 minutes per GHA/non-native runs with iOS jobs. [3]
You can enable pre-fill locally with `-D_CURL_PREFILL=ON`. It's
experimental, and if you experience a problem, file a PR or an Issue.
This PR also adds a pre-fill checker for macOS and MinGW/MSVC Windows
GHA jobs to catch if the cache diverges from real detections. It also
adds this logic to AppVeyor, but doesn't enable it due to the perf
penalty of 2 minutes mininum.
The pre-fill checker works by configuring out-of-tree with and without
pre-fill, then diffing their `lib/curl_config.h` outputs.
Exceptions are 3 detection results exposed indirectly [4], and missing
to expose 2, of which one is the C89 header `stddef.h`. While we assume
the C99 `stdint.h` available outside iOS. We can expose them in the
future, if necessary.
The pre-fill checks cost in total:
- ~20 seconds for macOS
- ~40 seconds for MinGW on GHA
- ~80 seconds for MSVC on GHA (UWP would be 2x this)
An extra time saving potential is caching type sizes. They are
well-known, and seldom change, esp. in CI. GHA/Windows jobs spend 8-17
seconds per job on these ~12 feature checks. ~5s on Cygwin/MSYS2. Couple
of seconds on other platforms. (This PR doesn't make this optimization.)
Another opportunity is doing the same for autotools, which typically
spends more time in the configuration step than cmake.
[1] Xcode job restored as a
follow-up to be5f20202c#16302
[2] GHA/macos cmake configure times in seconds:
Job | Bef. | After | Gain
:----------------------------------------------- | ----: | ----: | ----:
CM clang GnuTLS !ldap krb5 | 21.2 | 4.5 | 16.7
CM clang LibreSSL !ldap heimdal c-ares +examples | 13.3 | 3.9 | 9.4
CM clang OpenSSL +static libssh +examples | 20.0 | 4.6 | 15.4
CM clang OpenSSL IDN clang-tidy~ (w/chkprefill) | 15.7 | 18.6 | -2.9
CM clang OpenSSL gsasl rtmp AppleIDN | 25.0 | 4.7 | 20.3
CM clang OpenSSL torture !FTP | 15.3 | 4.5 | 10.8
CM clang OpenSSL torture FTP | 25.0 | 5.9 | 19.1
CM clang SecureTransport debug | 18.0 | 3.8 | 14.2
CM clang macos-13 SecureTransport | 45.8 | 12.4 | 33.4
CM clang macos-14 SecureTransport | 15.8 | 4.6 | 11.2
CM clang macos-15 SecureTransport | 26.8 | 6.1 | 20.7
CM clang mbedTLS openldap brotli zstd | 15.1 | 6.5 | 8.6
CM clang wolfSSL !ldap brotli zstd | 27.0 | 4.4 | 22.6
CM gcc-12 GnuTLS !ldap krb5 | 39.1 | 8.7 | 30.4
CM gcc-12 LibreSSL !ldap heimdal c-ares +examples| 23.8 | 7.2 | 16.6
CM gcc-12 OpenSSL +static libssh +examples | 20.7 | 8.5 | 12.2
CM gcc-12 OpenSSL gsasl rtmp AppleIDN | 23.1 | 10.1 | 13.0
CM gcc-12 SecureTransport debug | 21.1 | 4.8 | 16.3
CM gcc-12 mbedTLS openldap brotli zstd | 21.4 | 5.8 | 15.6
CM gcc-12 wolfSSL !ldap brotli zstd | 21.1 | 6.9 | 14.2
CM gcc-14 macos-13 SecureTransport | 61.9 | 18.7 | 43.2
CM gcc-14 macos-14 SecureTransport | 30.5 | 6.4 | 24.1
CM gcc-14 macos-15 SecureTransport | 32.7 | 8.4 | 24.3
CM llvm@15 GnuTLS !ldap krb5 | 21.1 | 7.5 | 13.6
CM llvm@15 LibreSSL !ldap heimdal c-ares +exampl~| 24.6 | 6.8 | 17.8
CM llvm@15 OpenSSL +static libssh +examples | 19.0 | 6.4 | 12.6
CM llvm@15 OpenSSL gsasl rtmp AppleIDN | 19.0 | 8.2 | 10.8
CM llvm@15 SecureTransport debug | 18.0 | 5.4 | 12.6
CM llvm@15 macos-13 SecureTransport | 66.2 | 25.7 | 40.5
CM llvm@15 macos-14 SecureTransport | 31.9 | 6.1 | 25.8
CM llvm@15 mbedTLS openldap brotli zstd | 19.5 | 8.9 | 10.6
CM llvm@15 wolfSSL !ldap brotli zstd | 24.3 | 5.9 | 18.4
CM llvm@18 macos-15 SecureTransport | 33.8 | 6.4 | 27.4
Total | 856.8 | 257.3 | 599.5
Before: https://github.com/curl/curl/actions/runs/13311042735/job/37173478424
After: https://github.com/curl/curl/actions/runs/13313927119/job/37183206426?pr=15841
[3] iOS:
Before: https://github.com/curl/curl/actions/runs/13326401704?pr=15841
After: https://github.com/curl/curl/actions/runs/13332177764?pr=15841
[4] detection results exposed indirectly in `curl_config.h`:
- `HAVE_FILE_OFFSET_BITS` via `_FILE_OFFSET_BITS`
- `HAVE_GETHOSTBYNAME_R_*_REENTRANT` via `NEED_REENTRANT`
- `HAVE_SOCKADDR_IN6_SIN6_ADDR` via `USE_IPV6`
Closes#15841
Before this patch curl code was redefining `getaddrinfo` and
`freeaddrinfo` system symbols to plug in its debug wrappers. This was
causing pains to avoid applying the redefinitions to system headers
defining these functions, and to the local debug wrappers. Especially
in unity builds. It also required workarounds for systems where these
symbols are already macros.
Introduce curl-namespaced macros for these functions and use them.
This allows to drop all workarounds and makes it work in all envs,
local targets and unity/bundle combinations.
Also drop GHA/windows workaround and use the same unity batch across
all jobs. Follow-up to 29e4eda631#16272
Ref: #16272
Ref: 71cf0d1fca#14772
Ref: 3efba94f77#14765
Ref: f7d5f47059#14399Closes#16274
Since last week the Ubuntu arm runner became flaky while installing `stunnel`.
```
08:07:26 Setting up stunnel4 (3:5.72-1build2) ...
08:07:26 Failed to check if group stunnel4 already exists: Connection refused
08:07:26 Group stunnel4 not found.
08:07:28 Reload daemon failed: Failed to activate service 'org.freedesktop.systemd1': timed out (service_start_timeout=25000ms)
08:07:28 Created symlink /etc/systemd/system/multi-user.target.wants/stunnel.target -> /usr/lib/systemd/system/stunnel.target.
08:08:18 Failed to get unit file state for stunnel.target: Connection timed out
08:08:43 Failed to retrieve unit state: Connection timed out
08:08:43 stunnel.target is a disabled or a static unit, not starting it.
08:08:43 /bin/chown: invalid user: ‘stunnel4:stunnel4’
08:08:43 dpkg: error processing package stunnel4 (--configure):
08:08:43 installed stunnel4 package post-installation script subprocess returned error exit status 1
08:08:43 [...]
08:08:47 Errors were encountered while processing:
08:08:47 stunnel4
08:08:54 Error: Timeout was reached
08:08:55 E: Sub-process /usr/bin/dpkg returned an error code (1)
08:08:55 Error: Process completed with exit code 100.
```
Ref: https://github.com/curl/curl/actions/runs/13280736653/job/37078440398?pr=16300#step:2:94Closes#16303
Fix `HAVE_GETHOSTBYNAME_R_*` feature detections always failing with
`CURL_WERROR=ON` due to stripping a const.
Also fix the GHA/cmake-vs-configure to enable `CURL_WERROR=ON` to sync
this setting with `./configure` which enables it by default. With that,
CI detects this issue.
```
CMake/CurlTests.c:73:19: error: initialization discards ‘const’ qualifier from pointer target type [-Wdiscarded-qualifiers]
73 | char *address = "example.com";
| ^~~~~~~~~~~~~
```
Ref: https://github.com/curl/curl/actions/runs/13225827821/job/36916564107#step:33:4198Closes#16282
Default curl unity builds make a single unit for each target. It means
all target sources are batched together and built in a single compiler
invocation. With multi-core CPUs this doesn't always result in the best
possible performance. This patch enables smaller batches for jobs where
this resulted in shorter build times. These jobs are Cygwin, MSYS2,
MinGW, running on the Windows runners.
Use batch of 30 (meaning 30 sources batched into units), and 32 for
Cygwin/MSYS2 to avoid a unity fallout that's subject to a different PR.
(CMake allows to set the number of sources per unit, not the number
of units, though the latter may be more practical to max out CPU cores.)
Also override to not batch the `curlu` target because batching lost
a little bit of time there, due to the already existing parallelism when
building the `testdeps` targets.
For jobs on the macOS and Linux runners jobs were already mostly single
digit or below teen seconds, and batching didn't improve on them
noticeably. On VM jobs, the virtual CPUs are limited, so I didn't
make a try. In AppVeyor and GHA vcpkg jobs (using msbuild), batching
didn't result in conclusive or any gains.
Build times in seconds (curl + testdeps):
Job | Before | After w curlu=0 | Gain
:--------------------| :-------------- | :-------------- | :---
cygwin, CM | 19 + 32 = 51 | 12 + 32 = 44 | 7
msys2, CM | 7 + 15 = 22 | 5 + 14 = 19 | 3
mingw gcc U, CM | 19 + 30 = 49 | 13 + 32 = 45 | 4
mingw ucrt, CM | 32 + 42 = 74 | 15 + 43 = 58 | 16
mingw clang, CM | 15 + 21 = 36 | 8 + 21 = 29 | 7
mingw uwp, CM | 30 + 40 = 70 | 14 + 40 = 54 | 16
mingw gcc, CM | 20 + 31 = 51 | 12 + 31 = 43 | 8
mingw x86, CM | 35 + 40 = 75 | 15 + 38 = 53 | 22
dl-mingw, CM 9.5.0 | 88 + 99 = 187 | 42 + 101 = 143 | 44
dl-mingw, CM 7.3.0 U | 24 + 32 = 56 | 17 + 35 = 52 | 4
Total | | | 131
Total gain per GHA/windows workflow runs: 2m11s
Runs:
Before: https://github.com/curl/curl/actions/runs/13220256084/job/36904342259
After: https://github.com/curl/curl/actions/runs/13220383702/job/36904602981https://github.com/curl/curl/actions/runs/13220613141/job/36905170104https://github.com/curl/curl/actions/runs/13222019443/job/36908358550
With curlu tweak: https://github.com/curl/curl/actions/runs/13222239255/job/36908782462
Ref: 116950a250#16265Closes#16272
To match the rest of Windows jobs.
dl-mingw, CM 9.5.0-x86_64 schannel: 4m24s -> 2m56s
dl-mingw, CM 7.3.0-x86_64 schannel U: 4m37s -> 3m10s
(based on a few runs.)
Closes#16271
Use the last known good release of Git for Windows by installing it
manually. It restores `runtests.pl` performance to the levels before
the October 2024 and this week's fallouts which gradually deployed
MSYS2 runtimes with pipe/signal/concurrency issues.
Also:
- restore vcpkg job's test parallelism to `-j8` (from `-j4`).
- keep using the default shell for jobs not running tests.
To avoid the unnecessary Git for Windows install overhead.
Upsides:
- good performance again.
- easy to experiment with any version.
Downsides:
- installing the Git for Windows package takes 15-30 seconds.
- we're pinned to an old package version.
- no canary to tell when the issue is fixed on the runner images.
Unknown:
- stability. (no MSYS2 runtimes were ever stable and it's difficult
to quantify if a version improves or worsens stability/flakiness, and
intermittent env failures.
Follow-up to 1bf774df57#16217
Follow-up to 5f9411f953#15380Closes#16265
Add CMake test project consuming curl via these methods:
`FetchContent`, `add_subdirectory()`, `find_package()`.
Also:
- GHA/distcheck: run these tests in CI.
- cmakelint: exclude a warning for calling "wonky-cased" built-in
CMake functions, such as `FetchContent_Declare()`.
Closes#16126
We set this macro to silence a warning inside `openldap.h`. With this
warning now silenced by using `-isystem`, we can drop it. Also it never
had to be set to `1`.
Also enable OpenLDAP in a CMake GHA/macos job.
Follow-up to 445fb81237#14763
Follow-up to 751e168d93#12024Closes#16146
This issue was not addressed with CMake builds so far. curl-for-win
worked thanks to its `-Wl,--start-group` workaround. It affects
binutils `ld` linking statically. Shared linking and llvm's `lld`
doesn't need strict lib order, and are not affected.
The solution is to pass libs in dependency order, with least dependent
(e.g. system) libs last. In case of cyclic dependency, may pass libs
twice.
Fix most issues by moving Windows system libs `ws2_32` and `bcrypt`
last, and move SSH libs first due to their dependence on crypto
backends and zlib compression.
Also:
- modify an existing Linux curl-for-win job to use gcc.
- add a specific Windows gcc job to test this. Make it use different
options than the default to extend build coverage too: `libssh`,
`zlib-ng`, 32-bit.
- prefer CMake imported targets for OpenSSL and ZLIB.
Examples of issues fixed:
Windows LibreSSL, libpsl vs. ws2_32:
```
x86_64-w64-mingw32-ld: curl/libressl/lib/libcrypto.a(bss_sock.c.obj):bss_sock.c:(.text$sock_ctrl[sock_ctrl]+0x59): undefined reference to `__imp_shutdown'
x86_64-w64-mingw32-ld: curl/libressl/lib/libcrypto.a(gcm128.c.obj):gcm128.c:(.text$CRYPTO_gcm128_init[CRYPTO_gcm128_init]+0x65): undefined reference to `__imp_ntohl'
x86_64-w64-mingw32-ld: curl/libpsl/_x64-win-ucrt/usr/lib/libpsl.a(psl.o):(.text$psl_is_cookie_domain_acceptable+0xef): undefined reference to `__imp_WSAStringToAddressW'
```
Ref: https://github.com/curl/curl/actions/runs/13157579253/job/36718144881?pr=16182#step:3:5354
Linux libssh2 vs. zlib:
```
/usr/lib/gcc-cross/aarch64-linux-gnu/12/../../../../aarch64-linux-gnu/bin/ld: curl/libssh2/_a64-linux-gnu-libressl/usr/lib/libssh2.a(unity_0_c.c.o): in function `comp_method_zlib_dtor':
(.text.comp_method_zlib_dtor+0x8c): undefined reference to `deflateEnd'
/usr/lib/gcc-cross/aarch64-linux-gnu/12/../../../../aarch64-linux-gnu/bin/ld: curl/libssh2/_a64-linux-gnu-libressl/usr/lib/libssh2.a(unity_0_c.c.o): in function `comp_method_zlib_comp':
(.text.comp_method_zlib_comp+0x50): undefined reference to `deflate'
/usr/lib/gcc-cross/aarch64-linux-gnu/12/../../../../aarch64-linux-gnu/bin/ld: curl/libssh2/_a64-linux-gnu-libressl/usr/lib/libssh2.a(unity_0_c.c.o): in function `comp_method_zlib_init':
(.text.comp_method_zlib_init+0x8c): undefined reference to `deflateInit_'
```
Ref: https://github.com/curl/curl/actions/runs/13157270420/job/36717189086?pr=16182#step:3:5285
Windows libssh vs. ws2_32 and LibreSSL:
```
/usr/bin/i686-w64-mingw32-ld: curl/libssh/_x86-win-ucrt-libressl/usr/lib/libssh.a(connect.c.obj):(.text$ssh_connect_host_nonblocking+0x92): undefined reference to `WspiapiGetAddrInfo@16'
/usr/bin/i686-w64-mingw32-ld: curl/libssh/_x86-win-ucrt-libressl/usr/lib/libssh.a(connect.c.obj):(.text$ssh_connect_host_nonblocking+0x3d9): undefined reference to `gai_strerrorA'
/usr/bin/i686-w64-mingw32-ld: curl/libssh/_x86-win-ucrt-libressl/usr/lib/libssh.a(kex.c.obj):(.text$ssh_client_select_hostkeys+0xd2): undefined reference to `FIPS_mode'
/usr/bin/i686-w64-mingw32-ld: curl/libssh/_x86-win-ucrt-libressl/usr/lib/libssh.a(options.c.obj):(.text$ssh_options_set+0x942): undefined reference to `FIPS_mode'
```
Ref: https://github.com/curl/curl/actions/runs/13163986294/job/36739557888?pr=16182#step:3:5127
Ref: https://github.com/curl/curl/actions/runs/13163986294/job/36739557888?pr=16182#step:3:5121Closes#16182
libssh 0.9.0 was shipped on June 28 2019 and is the first version
featuring the knownhosts API
Drop libssh from the GHA/linux-old CI job since it gets a libssh 0.7.3
version, too old for us now.
Closes#16200
