]> git.ipfire.org Git - thirdparty/curl.git/log
thirdparty/curl.git
6 months agobuild: stop detecting `sched_yield()` on Windows
Viktor Szakats [Thu, 16 Jan 2025 19:39:13 +0000 (20:39 +0100)] 
build: stop detecting `sched_yield()` on Windows

On Windows a successful `sched_yield()` detection requires mingw-w64
built with POSIX threads (not Win32 threads) and GCC (not llvm/clang).
(linking to `winpthread` via custom options may also work.)

In CMake builds, it was pre-cached as unavailable before this patch.

When detected (via autotools), it got only used for Windows XP or older
targets combined with a non-GCC, non-clang compiler that doesn't support
`__builtin_ia32_pause()`, or with the Intel C compiler. According to
`lib/easy_lock.h`.

mingw-w64 only supports GCC and clang, leaving a very narrow chance when
`shed_yield()` gets called on Windows. Even then, `sched_yield()` is
implemented in `winpthread` as `Sleep(0)`, which may or not be a useful.
It's also trivial to implement locally if it is, and such rare build
combination is also deemed useful.

Thus, this patch marks `sched_yields()` permanently unavailable on the
Windows platform also with autotools, and instead of pre-caching, skip
this feature check with CMake.

This syncs `HAVE_SCHED_YIELDS` between builds methods on Windows.

Follow-up to 9b517c8b69a1f365fdb6f54f7153561182285b6c #11973
Follow-up to 23af112f5556d6a785c17e09f2422ac931405f61 #8680

Closes #16037

6 months agocmake: detect mingw-w64 version, pre-fill `HAVE_STRTOK_R`
Viktor Szakats [Thu, 16 Jan 2025 17:10:31 +0000 (18:10 +0100)] 
cmake: detect mingw-w64 version, pre-fill `HAVE_STRTOK_R`

Replace `strtok_r()` detection with detection mingw-w64 version.
The use this version to pre-fill `HAVE_STRTOK_R`.

This gives use mingw-w64 version information for free.

Closes #16022

6 months agocmake: pre-fill `HAVE_STDATOMIC_H`, `HAVE_ATOMIC` for mingw-w64
Viktor Szakats [Thu, 16 Jan 2025 14:21:23 +0000 (15:21 +0100)] 
cmake: pre-fill `HAVE_STDATOMIC_H`, `HAVE_ATOMIC` for mingw-w64

`stdatomic.h` and `_Atomic` were first available in gcc 4.9.0 and
llvm/clang 3.6. Set detection values accordingly and save these two
detections on configure runs.

Closes #16036

6 months agoTLS: check connection for SSL use, not handler
Stefan Eissing [Fri, 17 Jan 2025 10:57:00 +0000 (11:57 +0100)] 
TLS: check connection for SSL use, not handler

Protocol handler option PROTOPT_SSL is used to setup a connection
filters. Once that is done, used `Curl_conn_is_ssl()` to check if
a connection uses SSL.

There may be other reasons to add SSL to a connection, e.g. starttls.

Closes #16034

6 months agopytest: remove 'repeat' parameter
Stefan Eissing [Fri, 17 Jan 2025 10:16:37 +0000 (11:16 +0100)] 
pytest: remove 'repeat' parameter

With the use of the pytest-repeat plugin, this parameter is no
longer necessary.

Closes #16033

6 months agowindows: drop dupe macros, detect `CURL_OS` for WinCE ARM, indentation
Viktor Szakats [Thu, 16 Jan 2025 14:31:54 +0000 (15:31 +0100)] 
windows: drop dupe macros, detect `CURL_OS` for WinCE ARM, indentation

- setup-win32: drop duplicate `WIN32_LEAN_AND_MEAN` and `NOGDI` macros
  (set earlier in `curl_setup.h`)

- config-win32ce: detect ARM for `CURL_OS` macro.
  Based on: https://cegcc.sourceforge.net/docs/details.html
  It should also apply to MSVC. (untested)

- indentation.

Closes #16029

6 months agoGHA: update `ubuntu-24.04` to `ubuntu-latest`
Viktor Szakats [Fri, 17 Jan 2025 02:29:26 +0000 (03:29 +0100)] 
GHA: update `ubuntu-24.04` to `ubuntu-latest`

`ubuntu-latest` is an alias for `ubuntu-24.04` now.

6 months agoconfigure: do not inline 'dnl' comments [ci skip]
Viktor Szakats [Fri, 17 Jan 2025 01:34:47 +0000 (02:34 +0100)] 
configure: do not inline 'dnl' comments [ci skip]

(Did not cause issues in this particular case.)

Follow-up to a1184525a6fc144cccf81cab406880519777bd21 #15971

6 months agoconfigure: UWP and Android follow-up fixes
Viktor Szakats [Fri, 17 Jan 2025 00:09:48 +0000 (01:09 +0100)] 
configure: UWP and Android follow-up fixes

- limit UWP detection to native Windows.
- add missing double-quotes to variable.
- drop interlock and sync order in buildinfo flags.

Follow-up to 56a74fac47a27cfbfe2f49976938dab7a1b9b4f2 #16014
Follow-up to f7bb6c1f64049dcdaee970de6759db6f8597879e #16020

Closes #16027

6 months agoandroid: add CI jobs, buildinfo, cmake docs, disable `CURL_USE_PKGCONFIG` by default
Viktor Szakats [Wed, 15 Jan 2025 14:46:32 +0000 (15:46 +0100)] 
android: add CI jobs, buildinfo, cmake docs, disable `CURL_USE_PKGCONFIG` by default

- GHA/non-native: add Android builds, both cmake and autotools,
  both NDK 21 (oldest available) and 35 (newest available)
  https://github.com/actions/runner-images/blob/main/images/ubuntu/Ubuntu2404-Readme.md
  It comes with a maintenance burden to bump the oldest/latest values
  with CI runner updates.

- cmake: disable `CURL_USE_PKGCONFIG` by default for Android.
  To avoid picking up system package by default.

- build: add `ANDROID-<NDK-LEVEL>` flag to `buildinfo.txt`.
  Also detect NDK level with the CMake built-in build method:
  https://cmake.org/cmake/help/latest/manual/cmake-toolchains.7.html#cross-compiling-for-android

- INSTALL.md: add CMake build instructions for Android.

- INSTALL.md: make NDK levels consistent in `./configure` example.

Closes #16014

6 months agoGHA: start using ARM Linux runners
Viktor Szakats [Thu, 16 Jan 2025 18:12:02 +0000 (19:12 +0100)] 
GHA: start using ARM Linux runners

- GHA/linux: allow per-job runner image override.
- GHA/linux: add an arm version of an existing job.
  Add workaround for broken `PATH` in the arm runner image.
- GHA/non-native: add CPU arch to job name where missing.
- GHA/checkdocs: switch a linter job to arm.

Performance looks a little bit better than Intel
(presumably with lower power consumption).

Test jobs, with openssl, tests, pytests, examples:
- arm:
  - https://github.com/curl/curl/actions/runs/12816430794/job/35737374521 4m7s
  - https://github.com/curl/curl/actions/runs/12816201136/job/35736615144 4m5s
- Intel:
  - https://github.com/curl/curl/actions/runs/12816430794/job/35737374118 4m32s
  - https://github.com/curl/curl/actions/runs/12816201136/job/35736614764 4m13s

Ref: https://github.blog/changelog/2025-01-16-linux-arm64-hosted-runners-now-available-for-free-in-public-repositories-public-preview/

Closes #16023

6 months agoGHA/linux: enable wolfSSH in a wolfSSL job
Viktor Szakats [Thu, 16 Jan 2025 21:19:51 +0000 (22:19 +0100)] 
GHA/linux: enable wolfSSH in a wolfSSL job

With tests. (Most SFTP tests fail with wolfSSH.)

Also:
- disable pseudo-terminal in wolfSSH builds (also in CircleCI).
  Not needed for curl.
- test642: add "compressed" to the description.
  To avoid being the same as test600.
  Follow-up to b7b4dc0d49543175ab0d9bb1cdc257a2d7f7cf0a #1735

Closes #16025

6 months agowebsocket: fix message send corruption
Stefan Eissing [Thu, 2 Jan 2025 15:34:52 +0000 (16:34 +0100)] 
websocket: fix message send corruption

- Fix a bug in EAGAIN handling when sending frames that led to a
  corrupted last byte of the frame sent.

- Restore sanity to curl_ws_send() behaviour:

  - Partial writes are reported as OK with the actual number of
    payload bytes sent.

  - CURLE_AGAIN is only returned when none of the payload bytes
    (or for 0-length frames, not all of the frame header bytes)
    could be sent.

  - curl_ws_send() now behaves like a common send() call.

- Change 'ws-data' test client to allow concurrent send/recv
  operations and vary frame sizes and repeat count.

- Add DEBUG env var CURL_WS_CHUNK_EAGAIN to simulate blocking
  after a chunk of an encoded websocket frame has been sent.

- Add tests.

Prior to this change data corruption may occur when sending websocket
messages due to two bugs:

1) 3e64569a (precedes 8.10.0) caused a data corruption bug in the last
   byte of frame of large messages.

2) curl_ws_send had non-traditional send behavior and could return
   CURLE_AGAIN with bytes sent and expect the caller to adjust buffer
   and buflen in a subsequent call. That behavior was not documented.

Reported-by: na-trium-144@users.noreply.github.com
Fixes https://github.com/curl/curl/issues/15865
Fixes https://github.com/curl/curl/issues/15865#issuecomment-2569870144
Closes https://github.com/curl/curl/pull/15901

6 months agoTODO: write an SMB test server to replace impacket
Daniel Stenberg [Thu, 16 Jan 2025 19:27:07 +0000 (20:27 +0100)] 
TODO: write an SMB test server to replace impacket

Closes #15697
Closes #16024

6 months agoasyn-ares: acknowledge CURLOPT_DNS_SERVERS set to NULL
Daniel Stenberg [Thu, 16 Jan 2025 07:45:45 +0000 (08:45 +0100)] 
asyn-ares: acknowledge CURLOPT_DNS_SERVERS set to NULL

Since c-ares has no function to restore this value to default, it
instead closes and re-inits the c-ares channel.

Reported-by: Deniz Sökmen
Fixes #16015
Closes #16016

6 months agoGHA/labeler: fix `INSTALL-CMAKE.md` references
Viktor Szakats [Thu, 16 Jan 2025 16:41:05 +0000 (17:41 +0100)] 
GHA/labeler: fix `INSTALL-CMAKE.md` references

Follow-up to 0f4c19b66ad5c646ebc3c4268a0f3ad9c15bf57c #12772

Closes #16021

6 months agoautotools: add support for mingw UWP builds
Viktor Szakats [Thu, 16 Jan 2025 15:47:15 +0000 (16:47 +0100)] 
autotools: add support for mingw UWP builds

To match cmake builds.

- GHA/windows: allow autotools UWP builds.
- detect UWP and add to `buildinfo.txt`.
  Consider it enabled if `CPPFLAGS` contains `-DWINSTORECOMPAT`.
- disable telnet with UWP.
- enable Unicode with UWP.
- do not use `wldap32` with UWP.
- do not enable `USE_WIN32_CRYPTO` with UWP.
- make sure to link to `ws2_32` in UWP builds.
  To fix `undefined reference to `in6addr_any'` when linking
  `tests/server` programs. More in the comment.

Closes #16020

6 months agodocs/EXPERIMENTAL.md: add a mention of HTTPSRR as experimental
Daniel Stenberg [Thu, 16 Jan 2025 09:52:39 +0000 (10:52 +0100)] 
docs/EXPERIMENTAL.md: add a mention of HTTPSRR as experimental

Closes #16017

6 months agocmake: move mingw UWP workaround from GHA to `CMakeLists.txt`
Viktor Szakats [Wed, 15 Jan 2025 13:28:19 +0000 (14:28 +0100)] 
cmake: move mingw UWP workaround from GHA to `CMakeLists.txt`

CMake (as of 3.31.2) doesn't fully recognize mingw-w64 with
`CMAKE_SYSTEM_NAME=WindowsStore`.
The manual logic works around it.

Also move existing DJGPP workaround to the same block.

Closes #16019

6 months agotidy-up: extend `CURL_O_BINARY` to lib and tests
Viktor Szakats [Wed, 15 Jan 2025 09:52:02 +0000 (10:52 +0100)] 
tidy-up: extend `CURL_O_BINARY` to lib and tests

Move `CURL_O_BINARY` definition from src to lib and use it from lib and
tests code.

Closes #16009

6 months agoRELEASE-NOTES: synced
Daniel Stenberg [Thu, 16 Jan 2025 08:27:49 +0000 (09:27 +0100)] 
RELEASE-NOTES: synced

6 months agotool_getparam: ECH param parsing refix
sftcd [Wed, 15 Jan 2025 12:47:41 +0000 (12:47 +0000)] 
tool_getparam: ECH param parsing refix

Regression.

Fixes #16006
Reported-by: Milon Renatus
Closes #16010

6 months agocf-https-connect: look into httpsrr alpns when available
Stefan Eissing [Wed, 15 Jan 2025 15:45:25 +0000 (16:45 +0100)] 
cf-https-connect: look into httpsrr alpns when available

Improved the filter implementation to be flexible in which order h3 and
h2/h1 are attempted. When HTTPSRR is enabled, look at the ALPNs it found
and use the order given for connecting in default setups.

Closes #16012

6 months agoGHA/non-native: lower job timeouts
Viktor Szakats [Wed, 15 Jan 2025 15:15:19 +0000 (16:15 +0100)] 
GHA/non-native: lower job timeouts

New values have a 2-3x headroom.

6 months agodoh: send HTTPS RR requests for all HTTP(S) transfers
Daniel Stenberg [Wed, 15 Jan 2025 10:04:46 +0000 (11:04 +0100)] 
doh: send HTTPS RR requests for all HTTP(S) transfers

When enabled in the build.

Update test2100: verify with HTTPS RR included

Adjust runtests and server/disabled.c to include "HTTPSRR" as a feature
in the test suite.

Also, decode the ALPN list in HTTPS records straight into IDs. There's
no point in storing everything in string format. Skip ALPNs we do not
support.

Closes #16007

6 months agodoh: cleanups and extended HTTPS RR code
Daniel Stenberg [Wed, 15 Jan 2025 08:03:47 +0000 (09:03 +0100)] 
doh: cleanups and extended HTTPS RR code

In preparation for using HTTPS outside of ECH, the parser now also
extracts the port number.

Plus other minor cleanups.

Closes #16007

6 months agoINSTALL.md: tweak to the MSVC version line [ci skip]
Viktor Szakats [Wed, 15 Jan 2025 01:33:36 +0000 (02:33 +0100)] 
INSTALL.md: tweak to the MSVC version line [ci skip]

Follow-up to 308437ac53a142be2c3470330e7de3737915bb23 #15992

6 months agoCI: add/extend curl binary info, VS2010 32-bit, misc improvements
Viktor Szakats [Sun, 12 Jan 2025 19:07:09 +0000 (20:07 +0100)] 
CI: add/extend curl binary info, VS2010 32-bit, misc improvements

- fix `find` commands to not miss items.

- call `file` on the built files in `curl -V` steps.
  To give more feedback on what was built.

- add `curl info` step for cross-jobs that can't do a `curl -V`.
  It lists the files built and calls `file` on them.

- appveyor: make a VS2010 32-bit to match the VS2008 job it replaced.
  Follow-up to d34aeecb08cdbb98cb81420605a78b83e4aee3c2 #15934

- GHA/windows: drop the word "old" from standalone mingw-w64 jobs to not
  conflate it with "old mingw" we no longer support (while also keeping
  it short).

Cherry-picked from #15975
Closes #16001

6 months agoeasy: make curl_easy_perform() return error if connection still there
Daniel Stenberg [Tue, 14 Jan 2025 14:47:21 +0000 (15:47 +0100)] 
easy: make curl_easy_perform() return error if connection still there

This typically happens if CURL_CONNECT_ONLY is used and a second
curl_easy_perform() is attempted.

A connection "taken over" with CURL_CONNECT_ONLY cannot be ended any
other way than a curl_easy_cleanup() on the easy handle that holds it.

Add test 696 to verify.

Closes #16003

6 months agoCURLOPT_CONNECT_ONLY.md: an easy handle with this option set cannot be reused
Daniel Stenberg [Tue, 14 Jan 2025 13:53:42 +0000 (14:53 +0100)] 
CURLOPT_CONNECT_ONLY.md: an easy handle with this option set cannot be reused

Closes #16002

6 months agoINSTALL.md: document VS2008 and mingw-w64
Viktor Szakats [Mon, 13 Jan 2025 18:52:50 +0000 (19:52 +0100)] 
INSTALL.md: document VS2008 and mingw-w64

Ref: #15972
Closes #15992

6 months agosha256/hmac: use these with TLS builds now
Daniel Stenberg [Mon, 13 Jan 2025 22:49:09 +0000 (23:49 +0100)] 
sha256/hmac: use these with TLS builds now

Used by the SSL session cache code.

Bug: https://curl.se/mail/lib-2025-01/0041.html
Reported-by: Aleksander Mazur
Closes #15994

6 months agotidy-up: drop parenthesis around `return` expression
Viktor Szakats [Mon, 13 Jan 2025 17:52:32 +0000 (18:52 +0100)] 
tidy-up: drop parenthesis around `return` expression

Closes #15990

6 months agoCI: run pytest in github CI colored
Stefan Eissing [Tue, 14 Jan 2025 09:27:05 +0000 (10:27 +0100)] 
CI: run pytest in github CI colored

Add colors to pytest runs in github workflows and see how test clutch
copes with it!

Closes #15998

6 months agopytest: use httpd/apache2 directly, no apachectl
Stefan Eissing [Tue, 14 Jan 2025 09:49:35 +0000 (10:49 +0100)] 
pytest: use httpd/apache2 directly, no apachectl

Since the script 'apachectl' from the httpd project is severly mutilated
on several distros, use the executable httpd/apache2 directly in pytest
runs.

Remove detection of apachectl form autoconf and cmake.

Closes #16000

6 months agocmdline-opts/version.md: describe multissl, mention SSLS-EXPORT
Daniel Stenberg [Tue, 14 Jan 2025 08:29:35 +0000 (09:29 +0100)] 
cmdline-opts/version.md: describe multissl, mention SSLS-EXPORT

Closes #15996

6 months agotransfer: fix CURLOPT_CURLU override logic
Jay Satiro [Mon, 13 Jan 2025 08:57:45 +0000 (03:57 -0500)] 
transfer: fix CURLOPT_CURLU override logic

- Change setopt and pretransfer to always reset URL related variables
  for a CURLU handle set CURLOPT_CURLU.

This change is to ensure we are in compliance with the doc which says
CURLU handles must be able to override a URL set via CURLOPT_URL and
that if the contents of the CURLU handle changes between transfers then
the updated contents must be used.

Prior to this change, although subsequent transfers appear to be
performed correctly in those cases, the work URL `data->state.url` was
not updated. CURLINFO_EFFECTIVE_URL returns data->state.url to the user
so it would return the URL from the initial transfer which was the wrong
URL. It's likely there are other cases as well.

Ref: https://curl.se/libcurl/c/CURLOPT_CURLU.html

Reported-by: Nicolás San Martín
Fixes https://github.com/curl/curl/issues/15984
Closes https://github.com/curl/curl/pull/15985

6 months agomprintf: terminate snprintf output on windows
Jay Satiro [Tue, 14 Jan 2025 09:21:38 +0000 (04:21 -0500)] 
mprintf: terminate snprintf output on windows

- Null terminate the end of the snprintf output buffer on Windows.

Old versions of the Windows CRT (which are often found on later versions
of Windows) do not terminate the snprintf output buffer if the output
reaches the max size.

This is a follow-up to parent 7e32f656 which made the same change but
limited it to mingw, however it is a CRT version issue irrespective of
compiler.

Ref: https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/snprintf-snprintf-snprintf-l-snwprintf-snwprintf-l?view=msvc-170#remarks

Closes https://github.com/curl/curl/pull/15997

6 months agomprintf: fix integer handling in float precision
Daniel Stenberg [Mon, 13 Jan 2025 12:24:31 +0000 (13:24 +0100)] 
mprintf: fix integer handling in float precision

In the double output function when an extremely large width and
precision is set that reaches the libcurl maximum (325), the handling of
the precision part would do wrong which could lead to bad output.

Also: work-around for single-byte buffer snprintf overflow with mingw.

Extend test 557 to verify.

Coverity CID 1638751.

Closes #15988

6 months agoGHA: enable the SSL session cache in a few builds
Daniel Stenberg [Mon, 13 Jan 2025 17:32:34 +0000 (18:32 +0100)] 
GHA: enable the SSL session cache in a few builds

Closes #15989

6 months agocurl_easy_ssls_export/import.md: made for TLS protocols
Daniel Stenberg [Mon, 13 Jan 2025 22:13:37 +0000 (23:13 +0100)] 
curl_easy_ssls_export/import.md: made for TLS protocols

Which then makes the generated man page also include details about the
specific backends that support this feature.

Follow-up to 515a21f350b89f0676e5df7f904c62c8f67be377

Closes #15993

6 months agocmake: enable SSLS-EXPORT in the build
Daniel Stenberg [Mon, 13 Jan 2025 18:32:53 +0000 (19:32 +0100)] 
cmake: enable SSLS-EXPORT in the build

Follow-up to 515a21f350b89f0676e5df7f904c62c8f67be377

Closes #15991

6 months agolib/easy.c: bring back the vtls/vtls_scache.h include
Daniel Stenberg [Mon, 13 Jan 2025 17:26:19 +0000 (18:26 +0100)] 
lib/easy.c: bring back the vtls/vtls_scache.h include

6 months agotelnet: handle single-byte input option
Daniel Stenberg [Mon, 13 Jan 2025 11:16:19 +0000 (12:16 +0100)] 
telnet: handle single-byte input option

Coverity CID 1638753 correctly identies this code misbehaved if the
passed in suboption is exactly one byte long by substracting two from
the unsigned size_t variable.

Closes #15987

6 months agoaltsvc: return error on dot-only name
Daniel Stenberg [Mon, 13 Jan 2025 11:05:02 +0000 (12:05 +0100)] 
altsvc: return error on dot-only name

Because it is a trailing dot that otherwise leads to a zero length name.

Coverity CID 1638755.

Closes #15986

6 months agoRELEASE-NOTES: synced
Daniel Stenberg [Mon, 13 Jan 2025 09:41:12 +0000 (10:41 +0100)] 
RELEASE-NOTES: synced

6 months agomultihandle: add an ssl_scache here
Daniel Stenberg [Sun, 12 Jan 2025 16:35:39 +0000 (17:35 +0100)] 
multihandle: add an ssl_scache here

The TLS session cache is now held by the multi handle unless it is
shared, so that all easy handles within a multi handle get the benefit
of sharing the same, larger, cache.

The multi handle session cache size is set to 25, unless it is the
internal one used for the easy interface - which still uses only 3.

Closes #15982

6 months agochecksrc: use 'banfunc' proper in more places
Daniel Stenberg [Mon, 13 Jan 2025 07:26:15 +0000 (08:26 +0100)] 
checksrc: use 'banfunc' proper in more places

6 months agoexamples: use return according to code style
Daniel Stenberg [Mon, 13 Jan 2025 07:24:00 +0000 (08:24 +0100)] 
examples: use return according to code style

6 months agochecksrc: check for return with parens around a value/name
Daniel Stenberg [Sun, 12 Jan 2025 21:52:13 +0000 (22:52 +0100)] 
checksrc: check for return with parens around a value/name

Ref: #15979
Closes #15983

6 months agocurl-functions.m4: fix indentation in `CURL_SIZEOF()`
Viktor Szakats [Sat, 11 Jan 2025 11:55:02 +0000 (12:55 +0100)] 
curl-functions.m4: fix indentation in `CURL_SIZEOF()`

Closes #15981

6 months agocmake: pick a better IPv6 feature flag when assembling the feature list
Viktor Szakats [Sat, 11 Jan 2025 12:27:51 +0000 (13:27 +0100)] 
cmake: pick a better IPv6 feature flag when assembling the feature list

Before this patch it used `ENABLE_IPV6`, the configuration intent.
Replace with `USE_IPV6` which is the actual setting passed to C.

The two can be different for targets without IPv6 support.

Closes #15980

6 months agotidy-up: drop parenthesis around `return` values
Viktor Szakats [Sat, 11 Jan 2025 17:46:39 +0000 (18:46 +0100)] 
tidy-up: drop parenthesis around `return` values

Closes #15979

6 months agoconfig: drop unused code and variables
Viktor Szakats [Sun, 12 Jan 2025 10:57:24 +0000 (11:57 +0100)] 
config: drop unused code and variables

- cmake, config-*: drop unused `PACKAGE*`, `VERSION` variables.
- config-win32: indentation
- config-win32ce: drop mingw-specific code.
  This header is not used with MinGW.
- config-win32ce: `_WIN64` is never true for Windows CE, drop.

Closes #15978

6 months agocmake: drop VS2010 "Dialog Hell" workaround added in 2013
Viktor Szakats [Sun, 12 Jan 2025 02:25:36 +0000 (03:25 +0100)] 
cmake: drop VS2010 "Dialog Hell" workaround added in 2013

Delete the workaround added via a94a68a3c1d04ccb53e46baa69753bbf6354ee14
(2013-02-04). The commit message has no details. The comment mentions
"Dialog Hell", and seems to fix CMake missing to regenerate `CURL.sln`
with VS2010. It also added a FIXME saying the workaround can be deleted
with future versions of CMake.

At the time CMake's latest version was v2.8.10.

curl now requires v3.7 (2018) minimum, and v3.24 (2022) was the
latest CMake natively supporting VS2010. Assume this has since been
fixed.

Also: format an MSVC version reference in comment.

Closes #15973

6 months agoconfigure: streamline Windows large file feature check
Viktor Szakats [Sat, 11 Jan 2025 23:55:32 +0000 (00:55 +0100)] 
configure: streamline Windows large file feature check

Before this patch the `CURL_CHECK_WIN32_LARGEFILE` feature check was
running an `AC_COMPILE` snippet that always succeeded. (except for
Windows CE, which isn't supported in other parts of `./configure` yet.)

The only Windows toolchain autotools supports is mingw. Of them, curl
only supports mingw-w64. All mingw-w64 versions support large files.
This allows to drop the check and assume it supported on Windows. To not
lose Windows CE support, rework that too, without using `AC_COMPILE`.

Drop the feature check altogether for non-Windows targets.

Ref: https://github.com/curl/curl/pull/15968#discussion_r1912158201
Follow-up to 7eb4ddb850d3757ac9e0b60b0198bbb11e83135e #15968

Closes #15971

6 months agosystem.h: fix indentation
Viktor Szakats [Sun, 12 Jan 2025 02:53:07 +0000 (03:53 +0100)] 
system.h: fix indentation

Closes #15974

6 months agomsvc: require VS2005 for large file support
Viktor Szakats [Thu, 9 Jan 2025 15:17:08 +0000 (16:17 +0100)] 
msvc: require VS2005 for large file support

Large file support requires `_fseeki64()`. This function is offered in
VS2005 and upper.

VS2003 has it in the static CRT only, with declaration missing from
headers, so it's not usable.

Ref: https://archive.org/details/X10-38445 (MS Visual Studio .NET 2003)
Ref: 8b76a8aeb21c8ae2261147af1bddd0d4637c252c #15526

Closes #15958

6 months agocurl_setup: fix missing `ADDRESS_FAMILY` type in rare build cases
Viktor Szakats [Fri, 10 Jan 2025 15:58:48 +0000 (16:58 +0100)] 
curl_setup: fix missing `ADDRESS_FAMILY` type in rare build cases

Build failed when both `ADDRESS_FAMILY` and `sockaddr_un` stuct were
missing from the Windows SDK, with UnixSockets enabled.

Seen with GNU 4.4.0 in CeGCC 0.59.1:
```
lib/curl_setup.h:983: error: expected specifier-qualifier-list before 'ADDRESS_FAMILY'
lib/curl_setup.h:985: warning: struct has no members
```

Also reported with VS2003:
https://datagirl.xyz/posts/wolfssl_curl_w2k.html

Closes #15969

6 months agowindows: drop redundant `USE_WIN32_SMALL_FILES` macro
Viktor Szakats [Sat, 11 Jan 2025 02:22:10 +0000 (03:22 +0100)] 
windows: drop redundant `USE_WIN32_SMALL_FILES` macro

In effect it meant `_WIN32 && !USE_WIN32_LARGE_FILES`.
Replace it with these macros.

Also:
- configure: delete tautological check for small file support.
- configure: delete stray `_MSC_VER` reference. autotools does not
  support MSVC.
- drop tautological checks for WinCE in `config-win32*.h` when setting
  `USE_WIN32_LARGE_FILES`.
- merge related PP logic.
- prefer `#ifdef`, fix whitespace.

Suggested-by: Marcel Raad
Report: https://github.com/curl/curl/pull/15952#issuecomment-2580092328

Closes #15968

6 months agotidy-up: `curl_setup.h`, `curl_setup_once.h`, `config-win32ce.h`
Viktor Szakats [Fri, 10 Jan 2025 23:36:10 +0000 (00:36 +0100)] 
tidy-up: `curl_setup.h`, `curl_setup_once.h`, `config-win32ce.h`

- drop unused `LIBIDN_REQUIRED_VERSION` macro.
  Unused since 9c91ec778104ae3b744b39444d544e82d5ee9ece

- drop compatibility error for `CURL_WANTS_CA_BUNDLE_ENV`.
  This macro was once set by `Makefile.mk` and Watcom makefiles.
  They are no longer supported, making the compatibility message moot.

  Follow-up to 7d7346519d8bff513f9ed9f0c5e709782b159e5a #1542 (2017)
  Follow-up to c2aeb1b3ba09af8b2e555eafa94dd9c534426bfb #1538 (2017)

- document last MSVC supporting the `!_MT` condition
  Ref: https://stackoverflow.com/questions/2278919/are-the-time-functions-of-msvc-thread-safe

- flatten an `#if` tree, prefer `#ifdef`.

- fix/adjust indentation, whitespace.

Closes #15967

6 months agomsvc: tidy up `_CRT_*_NO_DEPRECATE` definitions
Viktor Szakats [Thu, 9 Jan 2025 17:19:35 +0000 (18:19 +0100)] 
msvc: tidy up `_CRT_*_NO_DEPRECATE` definitions

Dedupe and migrate MSVC-specific warning suppressions to `curl_setup.h`.
Make cmake set `_CRT_SECURE_NO_DEPRECATE` for examples and standalone
tests, and stop setting `_CRT_NONSTDC_NO_DEPRECATE` for them.

Details:
- drop version guards. On ancient MSVC version these macro are a no-op.
- move to `curl_setup.h` from `config-win32*.h`.
- sync macro values with CMake.
- cmake: stop setting them globally in favour of `curl_setup.h`.
- cmake: re-add these macros to `docs/examples` and `tests/http/clients`,
  which do not use `curl_setup.h`.
- cmake: drop `_CRT_NONSTDC_NO_DEPRECATE` for examples and tests.
  They build fine without.
- update comments.

Closes #15960

6 months agocmake: formatting [ci skip]
Viktor Szakats [Sat, 11 Jan 2025 12:34:43 +0000 (13:34 +0100)] 
cmake: formatting [ci skip]

6 months agoci: update actions/upload-artifact digest to 65c4c4a
renovate[bot] [Fri, 10 Jan 2025 17:22:48 +0000 (17:22 +0000)] 
ci: update actions/upload-artifact digest to 65c4c4a

Closes #15964

6 months agoci: update rojopolis/spellcheck-github-actions digest to 9e0a5fb
renovate[bot] [Tue, 7 Jan 2025 19:14:48 +0000 (19:14 +0000)] 
ci: update rojopolis/spellcheck-github-actions digest to 9e0a5fb

Closes #15933

6 months agosystem.h: drop duplicate and no-op code
Viktor Szakats [Fri, 10 Jan 2025 19:17:37 +0000 (20:17 +0100)] 
system.h: drop duplicate and no-op code

- drop duplicate declaration block under `__MVS__`.
- drop duplicate declaration block under `__IBMC__`.
- drop empty `#if` branches under `__IBMC__`.

Closes #15966

6 months agosystem.h: drop compilers lacking 64-bit integer type (Windows/MS-DOS)
Viktor Szakats [Thu, 9 Jan 2025 14:42:34 +0000 (15:42 +0100)] 
system.h: drop compilers lacking 64-bit integer type (Windows/MS-DOS)

- DJGPP 1.x (including `__GO32__`) (MS-DOS)
  DJGPP 2.x support remains unchanged.
- Salford C (Windows)
- Turbo C (Windows 16-bit)
- Borland C++ < 5.2 (Windows 16-bit?)
- Pelles C < 2.8 (Windows)

These targets mapped `curl_off_t` to `long`. On Windows and MS-DOS
`long` is always 32-bit.

curl requires C compilers supporting 64-bit `curl_off_t` type since
835682661c7a36b0e860c978204e8cd9b033be45 #10597 (v8.0.0).

Also: drop remaining `__GO32__` and Salford C guards.

Closes #15957

6 months agobuild: fix `-Wtrampolines` picky warning for gcc 4.x versions
Viktor Szakats [Fri, 10 Jan 2025 10:03:16 +0000 (11:03 +0100)] 
build: fix `-Wtrampolines` picky warning for gcc 4.x versions

Caused an error when using cmake with gcc 4.4 and 4.5.
Also 4.3 when using autotools.

Seen with GNU 4.4.0 in CeGCC 0.59.1:
```
cc1: error: unrecognized command line option "-Wtrampolines"
```

`-Wtrampolines` requires gcc 4.6 or upper.

Ref: https://gcc.gnu.org/onlinedocs/gcc-4.6.0/gcc/Warning-Options.html
Ref: https://master.dl.sourceforge.net/project/cegcc/cegcc/0.59.1/cegcc_mingw32ce_snowleopard_r1397.tar.bz2

Closes #15962

6 months agocmake: deprecate winbuild, add migration guide from legacy build methods
Viktor Szakats [Mon, 6 Jan 2025 22:34:19 +0000 (23:34 +0100)] 
cmake: deprecate winbuild, add migration guide from legacy build methods

We recommend migrating to CMake from winbuild and Visual Studio project
files. winbuild is deprecated and will be dropped in September 2025.

CMake supports all the features and options, with new ones added
promptly. It supports out-of-tree, unity and documentation builds.

- deprecate winbuild method in favour of CMake by September 2025.
- add migration guide from winbuild to CMake.
- add migration guide from Visual Studio Project Files to CMake.
- add deprecation message to winbuild.
  Need to ack with `WINBUILD_ACKNOWLEDGE_DEPRECATED=yes`
Authored-by: Jay Satiro
- mention `CMAKE_BUILD_TYPE` option in `INSTALL-CMAKE`.
- document missing `SSH_PATH` winbuild option.

Closes #15920

6 months agomsvc: assume `_INTEGRAL_MAX_BITS >= 64`
Viktor Szakats [Thu, 9 Jan 2025 13:58:24 +0000 (14:58 +0100)] 
msvc: assume `_INTEGRAL_MAX_BITS >= 64`

curl requires C compilers supporting 64-bit `curl_off_t` type since
835682661c7a36b0e860c978204e8cd9b033be45 #10597 (v8.0.0).

Assume the MSVC compiler offers the necessary support.

It makes curl require Visual Studio .NET 2003, v7.1 (`_MSC_VER = 1310`).
With the possibility that 1300 (Visual Studio .NET, v7.0, 2002), or 1200
(Visual C++, 32-bit, v6.0, 1998) may also work.

Follow-up to ca18198dd4d6517bd2d8b34a99fcd9d4f780d219 #15952
Closes #15955

6 months agossl session cache: change cache dimensions
Stefan Eissing [Thu, 9 Jan 2025 10:18:31 +0000 (11:18 +0100)] 
ssl session cache: change cache dimensions

Use a larger one when shared.

Closes #15953

6 months agoRELEASE-NOTES: bump cmdline options and number of public funcs
Daniel Stenberg [Fri, 10 Jan 2025 07:54:49 +0000 (08:54 +0100)] 
RELEASE-NOTES: bump cmdline options and number of public funcs

6 months agoRELEASE-NOTES: synced
Daniel Stenberg [Fri, 10 Jan 2025 07:47:27 +0000 (08:47 +0100)] 
RELEASE-NOTES: synced

6 months agocompletion.pl: add completion for paths after @ for fish
Asger Hautop Drewsen [Tue, 7 Jan 2025 14:56:52 +0000 (15:56 +0100)] 
completion.pl: add completion for paths after @ for fish

Closes #15928

6 months agoECH: update APIs to those agreed with OpenSSL maintainers
sftcd [Wed, 8 Jan 2025 20:08:15 +0000 (20:08 +0000)] 
ECH: update APIs to those agreed with OpenSSL maintainers

Closes #15945

6 months agocookie: cap expire times to 400 days
Daniel Stenberg [Wed, 8 Jan 2025 09:19:26 +0000 (10:19 +0100)] 
cookie: cap expire times to 400 days

The pending cookie RFC update (currently known as 6265bis draft-19) says

  Let cookie-age-limit be the maximum age of the cookie (which name of
  Max-Age and an attribute-value of expiry-time. SHOULD be 400 days or
  less.

This change makes received cookies over the wire get capped to 400 days.

It does not cap the expiry date of cookies loaded from file.

It does this by rounding the expire time to a even minute. This, to
allow the test suite to do the same and have a chance to get the same
number for stable testing without requiring a debug build.

The test script generates TWO numbers in the output file for each
%days[] used in the input test file, and the function that subsequently
compares and verifies output is fine with *either* of the two numbers.

This is done so that if the test case is generated the second
immediately before curl runs, that updated expiry number is also deemed
okay. It still checks for an exact match of either number.

Closes #15937

6 months agoGnuTLS: fix 'time_appconnect' for early data
Stefan Eissing [Thu, 9 Jan 2025 12:09:50 +0000 (13:09 +0100)] 
GnuTLS: fix 'time_appconnect' for early data

When using early data with GnuTLS, the the timer `appconnect`
had the value from the "pretended" connect, not when the actual
TLS handshake was done.

Closes #15954

6 months agovtls_scache: fix possible null ptr deref
Daniel Stenberg [Thu, 9 Jan 2025 07:31:32 +0000 (08:31 +0100)] 
vtls_scache: fix possible null ptr deref

Pointed out by CodeSonar. It is probably not reachable, but might as
well just add a precaution.

Closes #15950

6 months agotool_ssls: fix "ignored return value" warning
Daniel Stenberg [Thu, 9 Jan 2025 07:25:36 +0000 (08:25 +0100)] 
tool_ssls: fix "ignored return value" warning

Pointed out by CodeSonar. While harmless, we might as well address it.

Closes #15949

6 months agovtls_scache: avoid a "Redundant Condition"
Daniel Stenberg [Thu, 9 Jan 2025 07:20:22 +0000 (08:20 +0100)] 
vtls_scache: avoid a "Redundant Condition"

Pointed out by CodeSonar. "ssl_config" can in fact not be NULL here.
Made it an assert instead.

Closes #15948

6 months agomsvc: drop checks for `_MSC_VER >= 900`
Viktor Szakats [Thu, 9 Jan 2025 11:00:56 +0000 (12:00 +0100)] 
msvc: drop checks for `_MSC_VER >= 900`

It's Visual C++, 32-bit, version 2.0, released in 1993. Used to verify
if `_INTEGRAL_MAX_BITS` is available.

After this patch we assume `_INTEGRAL_MAX_BITS` is always available in
MSVC.

Closes #15952

6 months agomsvc: drop checks for ancient versions
Viktor Szakats [Thu, 9 Jan 2025 02:33:03 +0000 (03:33 +0100)] 
msvc: drop checks for ancient versions

- drop version guard for `__inline`.
  Supported since `_MSC_VER` 1000.
  Visual C++, 32-bit, version 4.0 (1996)

- drop version guard for `__declspec(noreturn)` and `__forceinline`.
  Supported since `_MSC_VER` 1200.
  Visual C++, 32-bit, version 6.0 (1998)

For ancient versions, it's possible to override the default behaviour
by setting these macros via `CPPFLAGS`: `CURL_NORETURN`, `CURL_INLINE`,
`CURL_FORCEINLINE`

Closes #15946

6 months agoGHA/windows: add wolfSSL to MSVC MultiSSL job
Tal Regev [Thu, 9 Jan 2025 05:13:39 +0000 (07:13 +0200)] 
GHA/windows: add wolfSSL to MSVC MultiSSL job

Follow-up to 98932f34879bba86339b8ca94ba04aa994c744f8 #15765

Closes #15438

6 months agocurl_trc: repair build with verbose strings disabled
Daniel Stenberg [Thu, 9 Jan 2025 07:49:15 +0000 (08:49 +0100)] 
curl_trc: repair build with verbose strings disabled

- moved the dummy functions into the C file, made them non-static
- added a Curl_trc_ssls dummy

Closes #15951

6 months agocmake: update a comment [ci skip]
Viktor Szakats [Thu, 9 Jan 2025 03:11:35 +0000 (04:11 +0100)] 
cmake: update a comment [ci skip]

Follow-up to 421e592db25cbbe4baadfeef8e6ed75a57579d0a #15832

6 months agotidy-up: delete stray references from autotools, docs
Viktor Szakats [Wed, 8 Jan 2025 18:37:33 +0000 (19:37 +0100)] 
tidy-up: delete stray references from autotools, docs

- autotools: delete stray `VC14_LIB*` references.
- autotools: delete (now) empty `CLEANFILES`.
- autotools: delete no longer used lib/src .inc includes in root makefile.
- autotools: delete stray `cygwinbin` target.
- autotools: delete stray `pkgadd` target (Solaris).
- lib, src: delete stray files from `.gitignore`.
- INSTALL.md: delete reference to non-existing `src/config-win32.h`.
- lib/config-win32ce.h: whitespace.
- lib/config-win32ce.h: sync comments with `config-win32.h`.

Closes #15944

6 months agobuild: delete `-Wsign-conversion` related FIXMEs
Viktor Szakats [Wed, 8 Jan 2025 11:19:11 +0000 (12:19 +0100)] 
build: delete `-Wsign-conversion` related FIXMEs

We decided last year not to pursue avoiding this warning, because it
adds noise and friction, while in most cases not revealing actual code
issues. We fixed the interesting portion of them throughout mid-2024.

Conclude this effort by deleting related FIXMEs and temporary comments.

Follow-up to 3829759bd042c03225ae862062560f568ba1a231 #12489
Closes #15939

6 months agoDEPRECATE: remove msh3 in six months
Daniel Stenberg [Tue, 7 Jan 2025 17:06:24 +0000 (18:06 +0100)] 
DEPRECATE: remove msh3 in six months

The msh3 backed for QUIC and HTTP/3 was introduced in April 2022 but has
never been made to work properly. It has seen no visible traction or
developer activity from the msh3 main author (or anyone else seemingly
interested) in two years. As a non-functional backend, it only adds
friction and "weight" to the development and maintenance.

Meanwhile, we have a fully working backend in the ngtcp2 one and we have
two fully working backends in OpenSSL-QUIC and quiche well on their way
of ending their experimental status in a future.

We remove msh3 support from the curl source tree in July 2025.

Closes #15931

6 months agoosslq: use SSL_poll to determine writeability of QUIC streams
Neil Horman [Fri, 3 Jan 2025 15:17:56 +0000 (10:17 -0500)] 
osslq: use SSL_poll to determine writeability of QUIC streams

This discussion:
https://github.com/openssl/openssl/discussions/23339#discussion-6094341

Specifically item number 2 (Send Blocking) was raised by the curl team,
noting that SSL_want_write returning false was not a good indicator of
when a stream is writeable. The suggestion in that discussion was to use
SSL_poll with an SSL_POLL_EVENT_W flag instead, as that is a proper
indication of when an SSL_object will allow writing without blocking.

While ssl_want_write updates its state based on the last error
encountered (implying a need to retry an operation to update the
last_error state again), SSL_poll checks stream buffer status during the
call, giving it more up to date information on request. This is the
method used by our guide demos (quic-hq-interop specifically), and it
works well.

This change has been run through the curl test suite, and shown to pass
all tests. However, given the initial problem description I'm not sure
if there is a test case that explicitly checks for blocking and
unblocking of streams. As such some additional testing may be warranted.

Closes #15909

6 months agoscripts/delta: add number of days since first httpget release
Daniel Stenberg [Wed, 8 Jan 2025 22:48:48 +0000 (23:48 +0100)] 
scripts/delta: add number of days since first httpget release

6 months agoHTTP/2: strip TE request header
Stefan Eissing [Wed, 8 Jan 2025 15:34:38 +0000 (16:34 +0100)] 
HTTP/2: strip TE request header

The TE request header field is invalid in HTTP/2. Since clients may not
know in advance if a connection negotiates HTTP/2, automatically strip
such a header when h2 is in play.

Add test_01_10 to verify.

Reported-by: Jiri Stary
Fixes #15941
Closes #15943

6 months agovtls: feature ssls-export for SSL session im-/export
Stefan Eissing [Tue, 7 Jan 2025 11:41:26 +0000 (12:41 +0100)] 
vtls: feature ssls-export for SSL session im-/export

Adds the experimental feature `ssls-export` to libcurl and curl for
importing and exporting SSL sessions from/to a file.

* add functions to libcurl API
* add command line option `--ssl-sessions <filename>` to curl
* add documenation
* add support in configure
* add support in cmake
+ add pytest case

Closes #15924

6 months agoappveyor: tidy up shell code
Viktor Szakats [Wed, 8 Jan 2025 12:03:20 +0000 (13:03 +0100)] 
appveyor: tidy up shell code

Closes #15940

6 months agoappveyor: bump VS2008 jobs to VS2010
Viktor Szakats [Tue, 7 Jan 2025 22:44:32 +0000 (23:44 +0100)] 
appveyor: bump VS2008 jobs to VS2010

VS2008 has been partly broken for a while with its shared-debug builds
crashing on startup. Its compiler output (UTF-16 HTML) was also barely
readable even after conversion. It's also the only platform in CI
missing `stdint.h`.

This patch migrates a VS2008 job to VS2010 and drops another that
already had a VS2010 equivalent.

We recommend switching to VS2010 or newer when using MSVC to build curl.

Ref: #15907
Closes #15934

6 months agoappveyor: always use cmake `-A` option to select x64
Viktor Szakats [Tue, 7 Jan 2025 23:04:20 +0000 (00:04 +0100)] 
appveyor: always use cmake `-A` option to select x64

The `Win64` generator suffix alternative was required by old CMake
versions (<3.1) only:
https://cmake.org/cmake/help/v3.22/generator/Visual%20Studio%2010%202010.html

Closes #15935

6 months agoappveyor: reorder builds to get useful results earlier
Viktor Szakats [Tue, 7 Jan 2025 18:32:38 +0000 (19:32 +0100)] 
appveyor: reorder builds to get useful results earlier

Also to align with existing VS2010. Keeping the VS2008 job first to give
a quick sniff test for MSVC builds.

Follow-up to 08ff33e483e15b003de169cef33ca7cb6b6a32c0 #15923
Follow-up to 50f6a6b1d419c9e6ebc4fb2a848fa20def622d38 #15926

Closes #15932

6 months agoGHA/windows: drop MSH3 job (broken after 0.7.0 bump)
Viktor Szakats [Tue, 7 Jan 2025 14:45:10 +0000 (15:45 +0100)] 
GHA/windows: drop MSH3 job (broken after 0.7.0 bump)

Starting GHA runner image 20250105.1.0.

As seen on Linux with 0.7.0:
```
/home/runner/msh3/include/msh3.h:377:18: error: width of ‘RESERVED’ exceeds its type
  377 |             bool RESERVED                 : 5;
      |                  ^~~~~~~~
/home/runner/msh3/include/msh3.h:490:18: error: width of ‘RESERVED’ exceeds its type
  490 |             bool RESERVED            : 7;
      |                  ^~~~~~~~
```
https://github.com/curl/curl/actions/runs/12655717818/job/35266716846#step:35:195

Bug: https://github.com/curl/curl/pull/15924#issuecomment-2575106711
Bug: https://github.com/curl/curl/pull/15930#issuecomment-2575842486

Closes #15927

6 months agocontent_encoding: namespace GZIP flag constants
Viktor Szakats [Tue, 7 Jan 2025 14:59:47 +0000 (15:59 +0100)] 
content_encoding: namespace GZIP flag constants

To avoid collision with a 3rd-party `RESERVED` symbols.

This symbol is used in the public header of MSH3 0.7.0.

Closes #15929

6 months agocookie: small variable type cleanups
Daniel Stenberg [Tue, 7 Jan 2025 08:05:58 +0000 (09:05 +0100)] 
cookie: small variable type cleanups

- bump cookie counter and "creation time" to use 'unsigned int'
- use BIT() for single-bit struct field
- make invalid_octets() return bool properly

Closes #15921

6 months agocmake: pre-fill `HAVE_STDINT_H` on Windows
Viktor Szakats [Tue, 7 Jan 2025 10:02:10 +0000 (11:02 +0100)] 
cmake: pre-fill `HAVE_STDINT_H` on Windows

Closes #15925