To allow making per-job variations for SSH backends.
Also:
- fix Cygwin builds to not ignore per-job `install:` items.
It worked by accident before this patch.
Follow-up to 66313cc036#16629Closes#16911
curl now has a working GnuTLS CI job, with tests, with MSYS2.
The MultiSSL build scenario is now tested on macOS.
The vcpkg GnuTLS package seems to have a deep dependency tree with large
packages that need to be rebuilt relatively frequently. Since they can't
fit into to the time limit, these cause CI failures.
To stabilize CI, drop the `shiftmedia-libgnutls` dependency.
Partial revert of e86f99824c#16623
Ref: https://github.com/curl/curl/actions/runs/14192680124/job/39760753274?pr=16902Closes#16904
Setting a server-side file read-only by `chmod 0444` has does not
prevent overwriting it via SFTP upload (as tested in CI).
Fix it by setting its MS-DOS read-only attribute in addition. It
requires the Cygwin tool `chattr`.
Also unignore in CI.
Fixes:
```
test 0615...[SFTP put remote failure]
curl returned 0, when expecting 9
615: exit FAILED
=== Start of file stderr615
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 30 0 0 100 30 0 93 --:--:-- --:--:-- --:--:-- 95
100 30 0 0 100 30 0 92 --:--:-- --:--:-- --:--:-- 92
=== End of file stderr615
```
Ref: https://github.com/curl/curl/actions/runs/14037991918/job/39300723214#step:12:1269Closes#16818
Use the `PATH` `/usr/bin` to avoid any Windows system or 3rd-party tool
installed on the runner machine that may interfere with or add undesired
dependencies to the builds and tests.
Follow-up to d838d43430#16465
Ref: #16437Closes#16814
Install on drive `D:` which has much better write performance than `C:`,
on GitHub Windows runner machines.
- It's bringing down `dl-mingw` installation steps to 5-15s per job,
from 15s-130s before this patch.
- Saving 30-90s per job in the Cygwin install step.
The before values were fluctuating, but it seems reasonable to expect
saving at least a couple of minutes for each workflow run.
Closes#16813
- on native Windows (also when using MSYS2 openssh), the group and other
permissions do not end up as requested by Perl's chmod:
```diff
--- log/8/check-expected
+++ log/8/check-generated
@@ -1,3 +1,3 @@
d????????? N U U N ??? N NN:NN asubdir[LF]
--rw?rw?rw? 1 U U 37 Jan 1 2000 plainfile.txt[LF]
+-rw?r-?r-? 1 U U 37 Jan 1 2000 plainfile.txt[LF]
-r-?r-?r-? 1 U U 47 Dec 31 2000 rofile.txt[LF]
```
Ref: https://github.com/curl/curl/actions/runs/14004029192/job/39215359241?pr=16781#step:15:1596
Fix it by ignoring group and other attributes.
- fix failing postprocess cleanup by making the read-only test file
writeable again before deleting it. Fixing:
```
Directory not empty at ../../tests/libtest/test613.pl line 83.
```
(seen on Windows with Git for Windows `perl.exe`)
- unignore in GHA/windows.
Closes#16791
It accidentally worked on all CI-tested operating systems, except on
native Windows.
Fixing:
```
=== Start of file stderr612
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
[...]
curl: (21) rm command failed: Operation failed
```
Ref: https://github.com/curl/curl/actions/runs/14004029192/job/39215359241?pr=16781#step:15:1424
Also remove this test from the ignore list in GHA/windows.
Closes#16801
SFTP/SCP tests were failing in CI with WinCNG libssh2 since we first
added such job. With `curl: (67) Authentication failure`.
The reason is that the default `ssh-keygen` RSA private key format
changed to OpenSSH (RFC4716) in 2018. libssh2 does not support this
format with some of its crypto backends.
Fix it by generating keys explicitly in PEM format as necessary via
the `-m` option. This format is universally recognized for RSA keys.
2018-08-24: https://www.openssh.com/txt/release-7.8: OpenSSH format becomes default
2010-08-23: https://www.openssh.com/txt/release-5.6: `-m` option first supported
This fixed the auth issue, just to reveal a known flakiness issue in
libssh2 + WinCNG, causing:
```
curl: (2) Failure establishing ssh session: -8, Unable to exchange encryption keys
```
Ref: https://github.com/curl/curl/actions/runs/14000494428/job/39205633258?pr=16781#step:15:1796
Tracked here: https://github.com/libssh2/libssh2/issues/804
Mitigated in libssh2 tests by retrying them.
Due to this, keep ignoring these test results.
Also:
- add an env to customize key format: `CURL_TEST_SSH_KEY_FORMAT`
- display the generated format in the log.
- GHA/linux: document the wolfSSH error code causing it to fail tests:
```
curl: (79) wolfssh SFTP connect error -1051 / WS_MATCH_KEY_ALGO_E / cannot match key algo with peer
```
Follow-up to 4911e7af11#16735
Follow-up to 0ec72c1ef8#16672
Follow-up to e53523fef0#14859
Follow-up to e26cbe20cb#13979Closes#16781
On suspect of strain on the runtime env/pipes, disable this test, which
is flaky due to `runtests` detecting a 2009 result code from curl, while
curl is returning the expected 56:
```
test 0498...[Reject too large HTTP response headers on endless redirects]
curl returned 2009, when expecting 56
498: exit FAILED
== Contents of files in the log/5/ dir after test 498
[...]
0 0 0 0 0 0 0 0 --:--:-- 0:00:10 --:--:-- 0
curl: (56) Too large response headers: 6144086 > 6144000
```
In such cases the number of log lines for this single test is 4800. In
comparison the total number of log lines for a clear test run is 3800.
Seen with mingw, dl-mingw, msvc CI jobs.
Follow-up to 4911e7af11#16735
Ref: https://github.com/curl/curl/discussions/14854#discussioncomment-12503065Closes#16748
Skipping these tests saves time and reduces test logs from 11500 lines
to 3800.
Tests are permanently broken due to `curl: (67) Authentication failure`.
This libssh2 is built with WinCNG. Builds using libcrypto from OpenSSL
work fine.
Closes#16735
After restricting OpenSSH-Windows to a single job, and bumping it to
the pre-release version, that job started hanging then timing out with
reasonable consistency.
Since we saw similar hangs before with OpenSSH-Windows stable, in all
jobs, drop OpenSSH-Windows from CI, and replace it with MSYS openssh.
After this patch, all Windows jobs use MSYS2 or Cygwin openssh.
Follow-up to 0ec72c1ef8#16672Closes#16704
MSVC:
- switch jobs to standard openssh server. Reduce exceptions.
- make the SCP/SFTP ignore list more specific and comment with details.
- keep using OpenSSH-Windows for the OpenSSL job, and bump to the
prerelease version.
- disable `ENABLE_DEBUG` for BoringSSL to have such build tested. (This
is the first Windows non-ENABLE_DEBUG build with test runs.)
Takeaways:
- test 612 broken on Windows.
- test 613 broken on Windows with the standard openssh server.
- test 614 broken with libssh and OpenSSH-Windows.
- test 3022 broken with libssh2 and OpenSSH-Windows.
- tests broken with OpenSSH-Windows:
601 603 617 619 621 641 665 2004.
- vcpkg `libssh2[core,zlib]` broken due to:
curl: (67) Authentication failure
MSVC prep steps:
- install base msys2 package to simplify configuration, align with other
jobs and allow to use msys2 packages for tests.
- add support for msys2 openssh server. Keep OpenSSH-Windows as per-job
option. Add support for OpenSSH prerelease versions.
Prerelease does not make a difference in test results, but, stable was
last updated in 2019 (v8.0.0.1) and it seems better to use maintained
release track, with its latest from April 2024 (v9.5.0).
https://community.chocolatey.org/packages/openssh/8.0.0.1https://community.chocolatey.org/packages/opensshhttps://github.com/PowerShell/Win32-OpenSSHhttps://github.com/PowerShell/openssh-portable
- add 'libssh' to its job name.
- make `ENABLE_DEBUG` a per-job option.
msys/mingw:
- install `openssh` later and only when necessary.
- downgrade msys2 runtime later. (to follow other jobs)
- disable `CheckSpace` earlier. Also to untie it from the runtime
downgrade step, which we would hopefully drop.
Closes#16672
Replace `Cwd::abs_path()` with `File::Spec->rel2abs()`. The former
requires the file to exist, but in some cases, it's missing.
Seen in MSVC vcpkg jobs using Chocolatey OpenSSH v8.0.0.1 ending up with
`$path=/d/a/curl/curl/bld/tests/log/3/server/ssh_server.pid`, which does
not exist while converting to an absolute path (the path is already
absolute, but the conversion is done unconditionally):
```
Use of uninitialized value in subroutine entry at D:/a/curl/curl/tests/pathhelp.pm line 128.
can't convert empty path at D:/a/curl/curl/tests/pathhelp.pm line 128.
```
Ref: https://github.com/curl/curl/actions/runs/13747741797/job/38444844173#step:14:1233 (master)
Ref: https://github.com/curl/curl/actions/runs/13751862952/job/38453816737#step:14:3185 (trace)
Also ignore 3 new libssh2 jobs failing due to memleak.
Partial revert of 1bd5ac998b#16570Closes#16636
It fixes test 2302, 2303, 2307 with MSVC and clang on Windows.
GCC Windows builds were not affected.
Failure was caused by stack overflow due to a 1MB+ sized test struct on
stack. Replace it with dynamic allocation.
Also unignore affected tests in GHA/windows.
As seen under WINE with llvm-mingw:
```
$ wine64 libtests.exe lib2302 ws://127.0.0.1:59964/2302 > stdout2302 2> stderr2302
Test: lib2302
URL: ws://127.0.0.1:59964/2302
wine: Unhandled stack overflow at address 000000014007486A (thread 0024), starting debugger...
Unhandled exception: stack overflow in 64-bit code (0x000000014007486a).
```
Ref: #16629 (discovery)
Ref: 1bd5ac998b#16570Closes#16630
The GnuTLS MSVC/vcpkg build doesn't actually work on Windows. Let's
restore the build itself, to keep it fit for more testing. With disabled
tests (and examples) to keep it fast and not add to flakiness.
Also:
- enable GnuTLS in the MultiSSL job.
- limit building examples to one normal and one UWP job. It saves
6 x 1-1.5 minutes. Coverage remains the same, because example builds
only depend on the toolchain / target, not on the actual features
(except IPv6, but that's enabled for all.)
Closes#16623
- enable zstd in Cygwin and MSYS jobs.
- dl-mingw: use Ninja in the 9.5.0 (winlibs-mingw) job.
The download package is shipping with it. Saves 15s build time.
Keep testing GNU Makefiles with the two mingw-builds jobs.
- dl-mingw: split `env` prop to `env` and `ver` to aid integrating with
MSYS2.
- dl-mingw: install MSYS2 with options to make it quick (<20s).
It allows to use MSYS2 dependency packages with the downloaded
toolchains. It also makes configuration cleaner. Install libpsl.
- dl-mingw: enable mbedTLS in the 7.3.0 job.
(OpenSSL took a long time to install, wolfSSL misses features.)
Assisted-by: Jeremy Drake
Closes#16429
Re-enable running tests 19, 504, 704, 705, 1233 in CI MSYS jobs.
We carried over these exceptions from AppVeyor CI, where they have been
present for a long time. Cygwin jobs do not need these exceptions and
Cygwin and MSYS are similar envs. Time to re-evaluate if skipping them
is still necessary on MSYS.
Closes#16592
Bump msys2/setup-msys2 from 2.26.0 to 2.27.0. It brings the perf
regression experienced earlier with GfW and the pre-installed
MSYS2 on the GHA runner. Apply the runtime downgrade trick as
a workaround.
Fixes#16440Fixes#16547Closes#16574
These tests seem to be running no less stable now than others.
Stop ignoring their results to catch real issues.
These are consistently failing and remain on the ignore list:
in MSVC / vcpkg jobs:
```
FAIL-IGNORED 2302: 'WebSockets via callback (frame mode) + curl_ws_send()' WebSockets
FAIL-IGNORED 2303: 'WebSockets but gets a 200 back' WebSockets
FAIL-IGNORED 2307: 'WebSockets, overlong PING payload' WebSockets
```
https://github.com/curl/curl/actions/runs/13674664461/job/38233949942?pr=16570#step:14:4089
- Likely curl issues either in tests, server, or in WebSockets support.
in tests running under MSYS, affecting native mingw Windows builds only:
```
FAIL-IGNORED 612: 'SFTP post-quote remove file' SFTP, post-quote
[...]
curl: (21) rm command failed: Operation failed
```
https://github.com/curl/curl/actions/runs/13674664461/job/38233952699?pr=16570#step:14:1378
in tests running under MSYS, affecting both MSYS and native mingw Windows builds:
```diff
FAIL-IGNORED 613: 'SFTP directory retrieval' SFTP, directory
[...]
--- log/7/check-expected 2025-03-05 11:19:54.119658000 +0000
+++ log/7/check-generated 2025-03-05 11:19:54.119658000 +0000
@@ -1,3 +1,3 @@
d????????? N U U N ??? N NN:NN asubdir[LF]
--rw?rw?rw? 1 U U 37 Jan 1 2000 plainfile.txt[LF]
+-rw?r-?r-? 1 U U 37 Jan 1 2000 plainfile.txt[LF]
-r-?r-?r-? 1 U U 47 Dec 31 2000 rofile.txt[LF]
```
https://github.com/curl/curl/actions/runs/13674664461/job/38233950866?pr=16570#step:14:1316
- Possibly a curl test portabibility, Perl or MSYS issue.
in Cygwin tests:
```
FAIL-IGNORED 615: 'SFTP put remote failure' SFTP, SFTP put, FAILURE
```
https://github.com/curl/curl/actions/runs/13674664461/job/38233949428?pr=16570#step:12:3817
Follow-up to adcfd4fb3e#16553
Ref: #14854Closes#16570
To include the expected/generated diffs in the error log.
Also make it explicit for pure MSYS, though it was installed by `gcc`
before this patch.
Closes#16571
This job was never stable. Bumping to -j8 (from -j4) possibly made it
flakier: 032447e624#16271
Keep this job for build tests and drop running tests to improve the CI
experience and save CI time.
It's also a simple build with no dependencies. CI continues to build
a similar job with 9.5.0, which is more stable.
It remains a puzzle why builds with this toolchain (7.3.0 win32 threads
mingw-builds) is flakier and requires more test exceptions than the
indentical build with a slightly different build/version of
the toolchain (9.5.0 posix threads winlibs_mingw).
Ref: #14854Closes#16564
- 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
- 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
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
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
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
