]> git.ipfire.org Git - thirdparty/curl.git/log
thirdparty/curl.git
17 months agosocket: support binding to interface *AND* IP
Orgad Shaneh [Fri, 17 May 2024 11:44:44 +0000 (14:44 +0300)] 
socket: support binding to interface *AND* IP

Introduce new notation for CURLOPT_INTERFACE / --interface:
ifhost!<interface>!<host>

Binding to an interface doesn't set the address, and an interface can
have multiple addresses.

When binding to an address (without interface), the kernel is free to
choose the route, and it can route through any device that can access
the target address, not necessarily the one with the chosen address.

Moreover, it is possible for different interfaces to have the same IP
address, on which case we need to provide a way to be more specific.

Factor out the parsing part of interface option, and add unit tests:
1663.

Closes #13719

17 months agosocketpair: add `eventfd` and use `SOCK_NONBLOCK` for `socketpair()`
Andy Pan [Sat, 1 Jun 2024 18:26:11 +0000 (02:26 +0800)] 
socketpair: add `eventfd` and use `SOCK_NONBLOCK` for `socketpair()`

Currently, we use `pipe` for `wakeup_create`, which requires ***two***
file descriptors. Furthermore, given its complexity inside, `pipe` is a
bit heavyweight for just a simple event wait/notify mechanism.

`eventfd` would be a more suitable solution for this kind of scenario,
kernel also advocates for developers to use `eventfd` instead of `pipe`
in some simple use cases:

    Applications can use an eventfd file descriptor instead of a pipe
    (see pipe(2) in all cases where a pipe is used simply to signal
    events. The kernel overhead of an eventfd file descriptor is much
    lower than that of a pipe, and only one file descriptor is required
    (versus the two required for a pipe).

This change adds the new backend of `eventfd` for `wakeup_create` and
uses it where available, eliminating the overhead of `pipe`. Also, it
optimizes the `wakeup_create` to eliminate the system calls that make
file descriptors non-blocking by moving the logic of setting
non-blocking flags on file descriptors to `socketpair.c` and using
`SOCK_NONBLOCK` for `socketpair(2)`, `EFD_NONBLOCK` for `eventfd(2)`.

Ref:
https://man7.org/linux/man-pages/man7/pipe.7.html
https://man7.org/linux/man-pages/man2/eventfd.2.html
https://man7.org/linux/man-pages/man2/socketpair.2.html
https://www.gnu.org/software/gnulib/manual/html_node/eventfd.html

Closes #13874

17 months agoci: update github/codeql-action digest to 2e230e8
renovate[bot] [Tue, 4 Jun 2024 15:52:27 +0000 (15:52 +0000)] 
ci: update github/codeql-action digest to 2e230e8

Closes #13881

17 months agoexamples/threaded-ssl: remove locking callback code
Jay Satiro [Sat, 1 Jun 2024 18:41:41 +0000 (14:41 -0400)] 
examples/threaded-ssl: remove locking callback code

- Remove the locking callback code that demonstrates how to meet
  requirements of threading libraries (mainly OpenSSL).

Locking callback code has not been needed for many years. According to
the documentation for OpenSSL and GnuTLS they are thread-safe by design,
assuming support for the underlying OS threading API is built-in.

Ref: https://github.com/curl/curl/pull/13850#issuecomment-2143538458

Closes https://github.com/curl/curl/pull/13851

17 months agotests: delete redundant `!MSDOS` guard
Viktor Szakats [Tue, 4 Jun 2024 11:40:13 +0000 (13:40 +0200)] 
tests: delete redundant `!MSDOS` guard

This fix was supposed to be committed earlier, but ended up missing from
the final commit.

Follow-up to e9a7d4a1c8377dbcf9a2d94365f60e3e5dff48f8 #12376
Closes #13878

17 months agolib: fix thread entry point to return `DWORD` on WinCE
Viktor Szakats [Tue, 4 Jun 2024 00:41:26 +0000 (02:41 +0200)] 
lib: fix thread entry point to return `DWORD` on WinCE

We already do this in `tests/server/util.c`:
https://github.com/curl/curl/blob/97e5e37cc8269660bc5d4a1936f10f2390b97c5a/tests/server/util.c#L604-L606
and in `sockfilt.c`, `lib3026.c`.

Before this patch it returned `unsigned int`.

Closes #13877

17 months agosocket: use SOCK_NONBLOCK to eliminate extra system call
Andy Pan [Sun, 2 Jun 2024 06:40:42 +0000 (14:40 +0800)] 
socket: use SOCK_NONBLOCK to eliminate extra system call

Every time function `cf_socket_open()` is called to create a socket,
`curlx_nonblock()` is called to make that socket non-blocking. And
`curlx_nonblock()` will cost us 1 or 2 system calls (2 for `fcntl()`, 1
for `ioctl()`, etc.), meanwhile, tucking `SOCK_NONBLOCK` and
`SOCK_CLOEXEC` into the `type` argument for `socket()` is widely
supported across UNIX-like OS: Linux, *BSD, Solaris, etc. With that
ability, we can save 1 or 2 system calls on each socket.

Another change in this PR is to eliminate the redundant
`curlx_nonblock()` call on the socket in `cf_udp_setup_quic()` as that
socket created by `cf_socket_open()` is already non-blocking.

Ref:
https://man7.org/linux/man-pages/man2/socket.2.html
https://man.freebsd.org/cgi/man.cgi?socket(2)
https://man.dragonflybsd.org/?command=socket&section=2
https://man.netbsd.org/socket.2
https://man.openbsd.org/socket
https://docs.oracle.com/cd/E88353_01/html/E37843/socket-3c.html
https://illumos.org/man/3SOCKET/socket
...

Closes #13855

17 months agoGHA: show cmake error log in Windows and non-native workflows
Viktor Szakats [Tue, 4 Jun 2024 00:46:05 +0000 (02:46 +0200)] 
GHA: show cmake error log in Windows and non-native workflows

CMake configure doesn't fail often, but when it does, it helps to see
its `CMakeFiles/CMakeConfigureLog.yaml` output. This file is present
since CMake v3.26:
  https://cmake.org/cmake/help/v3.26/manual/cmake-configure-log.7.html

(Older CMake versions save similar contend to
`CMakeFiles\CMakeOutput.log` and
`CMakeFiles\CMakeError.log`. This patch doesn't deal with that because
the workflows touched are all running a newer CMake.)

After this patch, we dump the content if cmake fails. Syncing this with
autotools, where we already did that.

Closes #13872

17 months agoGHA: switch a Windows job to UCRT (gcc)
Viktor Szakats [Tue, 4 Jun 2024 09:20:20 +0000 (11:20 +0200)] 
GHA: switch a Windows job to UCRT (gcc)

Cherry-picked from #13870

17 months agocurl-config: revert to backticks to support old target envs
Viktor Szakats [Tue, 4 Jun 2024 02:19:32 +0000 (04:19 +0200)] 
curl-config: revert to backticks to support old target envs

Make an exception for `curl-config` because this script that may be
running on any target system, including old ones, e.g. SunOS 5.10.

Reported-by: Alejandro R. Sedeño
Ref: https://github.com/curl/curl/pull/13307#issuecomment-2146427358
Follow-up to fa69b41c7790fab86fd363242c81d8ef2e89e183 #13307
Closes #13871

17 months agombedtls: v3.6.0 workarounds
Stefan Eissing [Fri, 31 May 2024 11:01:17 +0000 (13:01 +0200)] 
mbedtls: v3.6.0 workarounds

- add special sauce to disable unwanted peer verification by mbedtls
  when negotiating TLS v1.3
- add special sauce for MBEDTLS_ERR_SSL_RECEIVED_NEW_SESSION_TICKET
  return code on *writing* TLS data. We assume the data had not been
  written and EAGAIN.
- return correct Curl error code when peer verification failed.
- disable test_08_05 with 50 HTTP/1.1 connections, as mbedtls reports a
  memory allocation failed during handshake.
- bump CI mbedtls version to 3.6.0

Fixes #13653
Closes #13838

17 months agognutls: support CA caching
Stefan Eissing [Mon, 27 May 2024 14:50:15 +0000 (16:50 +0200)] 
gnutls: support CA caching

- similar to openssl, use a shared 'credentials' instance
  among TLS connections with a plain configuration.
- different to openssl, a connection with a client certificate
  is not eligible to sharing.
- document CURLOPT_CA_CACHE_TIMEOUT in man page

Closes #13795

17 months agotests: don't log buffer length in throwing away message
Dan Fandrich [Tue, 4 Jun 2024 05:19:14 +0000 (22:19 -0700)] 
tests: don't log buffer length in throwing away message

It's not available at that point, and it will be written in the
non-error case right afterward.

17 months agotests: log "Throwing away" messages before throwing away
Dan Fandrich [Mon, 3 Jun 2024 20:25:29 +0000 (13:25 -0700)] 
tests: log "Throwing away" messages before throwing away

In case the read that follows hangs we'll get a clue as to what it was
doing.

17 months agoCI: reduce memory request for FreeBSD builds
Dan Fandrich [Fri, 31 May 2024 01:39:12 +0000 (18:39 -0700)] 
CI: reduce memory request for FreeBSD builds

Also, add a comment with link to the Cirrus credit page since it's not
easy to find otherwise.

17 months agotcpkeepalive: support setting TCP keep-alive parameters on Solaris <11.4
Andy Pan [Mon, 3 Jun 2024 07:15:55 +0000 (15:15 +0800)] 
tcpkeepalive: support setting TCP keep-alive parameters on Solaris <11.4

Solaris didn't support TCP_KEEPIDLE and TCP_KEEPINTVL until 11.4,
before that it use TCP_KEEPALIVE_THRESHOLD and TCP_KEEPALIVE_ABORT_THRESHOLD
as the substitute. Therefore, for Solaris <11.4 we need to use this substitute
for setting TCP keep-alive parameters.

Ref:
https://docs.oracle.com/cd/E86824_01/html/E54777/tcp-7p.html
https://docs.oracle.com/cd/E88353_01/html/E37851/tcp-4p.html

Closes #13864

17 months agoKNOWN_BUGS: quiche: QUIC connection is draining
Daniel Stenberg [Mon, 3 Jun 2024 08:10:47 +0000 (10:10 +0200)] 
KNOWN_BUGS: quiche: QUIC connection is draining

Closes #12037
Closes #13867

17 months agoKNOWN_BUGS: aws-sigv4 has problems with particular URLs
Daniel Stenberg [Mon, 3 Jun 2024 08:04:50 +0000 (10:04 +0200)] 
KNOWN_BUGS: aws-sigv4 has problems with particular URLs

Closes #13058
Closes #13866

17 months agoKNOWN_BUGS: aws-sigv4 does not handle multipart/form-data correctly
Daniel Stenberg [Mon, 3 Jun 2024 08:02:33 +0000 (10:02 +0200)] 
KNOWN_BUGS: aws-sigv4 does not handle multipart/form-data correctly

Closes #13351
Closes #13866

17 months agoRELEASE-NOTES: synced
Daniel Stenberg [Mon, 3 Jun 2024 08:43:06 +0000 (10:43 +0200)] 
RELEASE-NOTES: synced

17 months agoGHA: fix old mingw-w64 32-bit job
Viktor Szakats [Sun, 2 Jun 2024 23:11:47 +0000 (01:11 +0200)] 
GHA: fix old mingw-w64 32-bit job

This toolchain resides in the `mingw32` directory. Make sure to
configure `PATH` accordingly.

Before this patch, it pointed to a non-existing `mingw64` directory,
making the job use the wrong compiler (gcc 12, 64-bit).

Follow-up to e838b341a08b44d4a8486fb0d3f15d12fc794c62 #12927
Closes #13863

17 months agotool_cb_hdr: return error for failed header writes
Daniel Stenberg [Sun, 2 Jun 2024 15:16:58 +0000 (17:16 +0200)] 
tool_cb_hdr: return error for failed header writes

By checking that fflush() works.

Reported-by: Sebastian Andersson
Fixes #13836
Closes #13859

17 months agoGHA: bump all build jobs to nproc+1
Viktor Szakats [Tue, 28 May 2024 07:01:49 +0000 (09:01 +0200)] 
GHA: bump all build jobs to nproc+1

- bump rest of the workflows (windows, macos, distrocheck).

- non-native virtualized envs have 2 CPUs, bump down accordingly.
  (for `vmactions/omnios-vm` it's just a guess.)

- bump all to nproc + 1.

Follow-up to e838b341a08b44d4a8486fb0d3f15d12fc794c62 #12927
Closes #13807

17 months agoGHA: disable MQTT and WebSocket tests in Windows jobs
Viktor Szakats [Sun, 2 Jun 2024 17:30:19 +0000 (19:30 +0200)] 
GHA: disable MQTT and WebSocket tests in Windows jobs

Trying to figure out which category is causing the remaining hangs.

Follow-up to def7d05382743ea7aa1d356d1e41dcb22ecdd4d7
Closes #13860

17 months agolib/v*: tidy up types and casts
Viktor Szakats [Mon, 13 May 2024 15:56:19 +0000 (17:56 +0200)] 
lib/v*: tidy up types and casts

Also add a couple of negative checks.

Cherry-picked from #13489
Closes #13622

17 months agoGHA: fix caching old mingw-w64 toolchains in the Windows workflow
Viktor Szakats [Sun, 2 Jun 2024 16:22:48 +0000 (18:22 +0200)] 
GHA: fix caching old mingw-w64 toolchains in the Windows workflow

- stop altering the `PATH` via `GITHUB_ENV`. This confused the
  `actions/cache` post-job, which needs to run in the exact same
  environment as its pre-job, to have a consistent cache entry "version"
  hash. Altering the `PATH` via `GITHUB_ENV` spills into the the
  post-job and breaks this hash. GHA doesn't reset the env automatically
  and I have not found a way to do it manually.

- add double-quotes where missing.

- move cache directory under `USERPROFILE` to not rely on absolute
  paths.

- make cache directory flatter and versionless.

Follow-up to 0914d8aadddac0d1459673d5b7f77e8f3378b22b #13759
Closes #13856

17 months agoci: pin actions/github-script action to 60a0d83
renovate[bot] [Sat, 1 Jun 2024 08:58:26 +0000 (08:58 +0000)] 
ci: pin actions/github-script action to 60a0d83

Closes #13846

17 months agox509asn1: add some common ECDSA OIDs
Bo Anderson [Sun, 2 Jun 2024 13:17:10 +0000 (14:17 +0100)] 
x509asn1: add some common ECDSA OIDs

Closes #13857

17 months agoci: update rojopolis/spellcheck-github-actions digest to e36f662
renovate[bot] [Sat, 1 Jun 2024 18:44:11 +0000 (18:44 +0000)] 
ci: update rojopolis/spellcheck-github-actions digest to e36f662

Closes #13852

17 months agox509asn1: fallback to dotted OID representation
Bo Anderson [Sun, 2 Jun 2024 13:18:20 +0000 (14:18 +0100)] 
x509asn1: fallback to dotted OID representation

Reported-by: Luke Hamburg
Fixes #13845
Closes #13858

17 months agorequest.md: language fix
Lee Li [Sun, 2 Jun 2024 01:42:52 +0000 (09:42 +0800)] 
request.md: language fix

improved for better readability and correctness

Closes #13854

17 months agovtls: deprioritize Secure Transport
MonkeybreadSoftware [Tue, 7 May 2024 07:45:15 +0000 (09:45 +0200)] 
vtls: deprioritize Secure Transport

Moved Secure Transport behind OpenSSL, so we can build CURL with both
and prefer using OpenSSL over Secure Transport by default.

Closes #13547

17 months agourlapi: add CURLU_NO_GUESS_SCHEME
Daniel Stenberg [Sat, 1 Jun 2024 10:03:34 +0000 (12:03 +0200)] 
urlapi: add CURLU_NO_GUESS_SCHEME

Used for extracting:

- when used asking for a scheme, it will return CURLUE_NO_SCHEME if the
  stored information was a guess

- when used asking for a URL, the URL is returned without a scheme, like
  when previously given to the URL parser when it was asked to guess

- as soon as the scheme is set explicitly, it is no longer internally
  marked as guessed

The idea being:

1. allow a user to figure out if a URL's scheme was set as a result of
  guessing

2. extract the URL without a guessed scheme

3. this makes it work similar to how we already deal with port numbers

Extend test 1560 to verify.

Closes #13616

17 months agowolfssl: support CA caching
Daniel Stenberg [Sat, 1 Jun 2024 10:24:42 +0000 (12:24 +0200)] 
wolfssl: support CA caching

As a bonus, add SSLSUPP_CA_CACHE to let TLS backends signal its support
for this so that *setopt() return error if there is no support.

Closes #13786

17 months agosocket: change TCP keepalive from ms to seconds on DragonFly BSD
Andy Pan [Sat, 1 Jun 2024 10:36:04 +0000 (18:36 +0800)] 
socket: change TCP keepalive from ms to seconds on DragonFly BSD

DragonFly BSD changed the time unit for TCP keep-alive from milliseconds
to seconds since v5.8, thus setting the keepalive options with
milliseconds with curl/libcurl will result in unexpected behaviors on
DragonFlyBSD 5.8+

Distinguish the DragonFly BSD versions and use the proper time units
accordingly.

Ref:
  https://lists.dragonflybsd.org/pipermail/commits/2019-July/719125.html
  https://github.com/DragonFlyBSD/DragonFlyBSD/blob/965b380e960908836b97aa034fa2753091e0172e/sys/sys/param.h#L207

Fixes #13847
Closes #13848

17 months agocurlver.h: aiming for 8.9.0
Daniel Stenberg [Sat, 1 Jun 2024 10:26:20 +0000 (12:26 +0200)] 
curlver.h: aiming for 8.9.0

17 months agonoproxy: patterns need to be comma separated
Daniel Stenberg [Sat, 1 Jun 2024 10:25:12 +0000 (12:25 +0200)] 
noproxy: patterns need to be comma separated

or they will not parse correctly.

Mentioned in DEPRECATED since Janurary 2023 (in 7ad8a7ba9ebdedc).

Closes #13789

17 months agosectransp: remove large cipher table
Jan Venekamp [Wed, 29 May 2024 11:58:53 +0000 (13:58 +0200)] 
sectransp: remove large cipher table

Previously a large table of ciphers was used to determine the default
ciphers and to lookup manually selected ciphers names.

With the lookup of the manually selected cipher names moved to
Curl_cipher_suite_walk_str() the large table is no longer needed for
that purpose.

The list of manually selected cipher can now be intersected with the
ciphers supported by Secure Transport (SSLGetSupportedCiphers()),
instead of using the fixed table for that.

The other use of the table was to filter the list of all supported
ciphers offered by Secure Transport to create a list of ciphers to
use by default, excluding ciphers in the table marked as weak.

Instead of using a complement based approach (exclude weak), switch
to using an intersection with a smaller list of ciphers deemed
appropriate.

Closes #13823

17 months agoGHA: unify http3 workflows into one
Tatsuhiro Tsujikawa [Tue, 21 May 2024 09:50:57 +0000 (18:50 +0900)] 
GHA: unify http3 workflows into one

This commit unifies the following http3 workflows into http3-linux.yml:

- ngtcp2-linux.yml
- osslq-linux.yml
- quiche-linux.yml

The idea is better use of the build cache. Previously, they
independently create caches with the same key. Some of the caches
include source code and intermediate object files, which makes cache
quite large. In this commit, only built artifacts are cached, which
drastically reduces the cache size. OpenSSL v3, mod_h2 and quiche caches
still include all stuff, but they are left for the later improvement.
Because the contents of the cache have been changed, the cache keys are
also changed to include the word "http3".

Closes #13841

17 months agoopenSSL: fix hostname handling when using ECH
Stephen Farrell [Wed, 29 May 2024 11:32:32 +0000 (12:32 +0100)] 
openSSL: fix hostname handling when using ECH

Reported-by: vvb2060
Fixes #13818
Closes #13822

17 months agoci: update github/codeql-action digest to f079b84
renovate[bot] [Fri, 31 May 2024 09:55:28 +0000 (09:55 +0000)] 
ci: update github/codeql-action digest to f079b84

Closes #13837

17 months agoRELEASE-NOTES: synced
Daniel Stenberg [Sat, 1 Jun 2024 08:40:20 +0000 (10:40 +0200)] 
RELEASE-NOTES: synced

17 months agocurl_multi_poll.md: expand the example with an custom file descriptor
Daniel Stenberg [Fri, 31 May 2024 12:42:45 +0000 (14:42 +0200)] 
curl_multi_poll.md: expand the example with an custom file descriptor

Closes #13842

17 months agoDISTROS: add a link to the list archive
Christian Heusel [Fri, 31 May 2024 12:44:02 +0000 (14:44 +0200)] 
DISTROS: add a link to the list archive

Related to https://github.com/curl/curl/discussions/13833

Signed-off-by: Christian Heusel <christian@heusel.eu>
Closes #13843

17 months agoautoconf: remove 'deeper' checks for `AC_CHECK_FUNCS`
Matt Jolly [Thu, 30 May 2024 04:47:11 +0000 (14:47 +1000)] 
autoconf: remove 'deeper' checks for `AC_CHECK_FUNCS`

The net effect of the deeper checks is to raise implicit function decls
on modern compilers.

These checks appear to have been added ~20 years ago, relating to an
unverifiable claim about HP-UX. Autoconf support for the platform has
grown in leaps and bounds since.

It didn't cause a real problem here, but when investigating a FP this
came up. No evidence has been identified that this was actually broken
in the past, and there is no evidence that this is necessary now.

`-Werror=implicit-function-declarations` is enabled for both checks;
without a working prototype they will both fail regardless. In the
second case there will in fact never be a working prototype and
therefore it will always fail unconditionally.

`AC_CHECK_FUNCS` does effectively the same thing as the removed checks,
except it actually defines a dummy prototype to see if it links.

If `AC_CHECK_FUNCS` is broken on a given platform we have bigger
problems than trying to build cURL. This should also be faster.

Bug: https://bugs.gentoo.org/932827
Reviewed-By: Eli Schwartz <eschwartz93@gmail.com>
Closes #13830

17 months agocf-socket: improve SO_SNDBUF update for Winsock
Jay Satiro [Sun, 26 May 2024 18:56:24 +0000 (14:56 -0400)] 
cf-socket: improve SO_SNDBUF update for Winsock

- Rename: Curl_sndbufset => Curl_sndbuf_init

- Rename: win_update_buffer_size => win_update_sndbuf_size

- Save the last set SO_SNDBUF size to compare against so that we can
  avoid setsockopt calls every second.

This is a follow-up to 0b520e12 which moved the SO_SNDBUF update check
into cf-socket. This change improves it further by making the function
names easier to understand and reducing the amount of setsockopt calls.

Closes https://github.com/curl/curl/pull/13827

17 months agotidy-up: use consistent casing for Windows directories
Viktor Szakats [Thu, 30 May 2024 10:26:12 +0000 (12:26 +0200)] 
tidy-up: use consistent casing for Windows directories

C:\Windows\System32

Closes #13832

17 months agoGHA: use ubuntu-latest with OmniOS job
Viktor Szakats [Thu, 30 May 2024 10:00:27 +0000 (12:00 +0200)] 
GHA: use ubuntu-latest with OmniOS job

It's the same as ubuntu-22.04.

Also update OmniOS package search link.

Closes #13831

17 months agoGHA: adjust parallel job counts
Ayesh Karunaratne [Mon, 12 Feb 2024 07:05:35 +0000 (14:05 +0700)] 
GHA: adjust parallel job counts

Adjusts the `make -j` flag to match the latest GitHub-hosted runner
hardware specs[^1]:

 - `ubuntu-latest` on 4 CPU cores
 - `macos-latest` on 3 CPU cores

The processor count is ideally obtained from `nproc`, but setting env
vars from the current CI yaml files is not possible because they expect
literal strings.

[^1]: https://docs.github.com/en/actions/using-github-hosted-runners/about-github-hosted-runners/about-github-hosted-runners#standard-github-hosted-runners-for-public-repositories

Closes #12927

17 months agoget.d: clarify the explanation
pszlazak [Sun, 19 May 2024 15:35:47 +0000 (17:35 +0200)] 
get.d: clarify the explanation

Closes #13706

17 months agocurl_url_set.md: libcurl only parses :// URLs
Daniel Stenberg [Wed, 29 May 2024 09:33:41 +0000 (11:33 +0200)] 
curl_url_set.md: libcurl only parses :// URLs

Make it clearer in the documentation.

Closes #13821

17 months agomulti: fix multi_wait() timeout handling
Stefan Eissing [Wed, 29 May 2024 15:13:34 +0000 (17:13 +0200)] 
multi: fix multi_wait() timeout handling

- determine the actual poll timeout *after* all sockets
  have been collected. Protocols and connection filters may
  install new timeouts during collection.
- add debug logging to test1533 where the mistake was noticed

Reported-by: Matt Jolly
Fixes #13782
Closes #13825

17 months agolib: prefer `var = time(NULL)` over `time(&var)`
Viktor Szakats [Tue, 28 May 2024 14:16:01 +0000 (16:16 +0200)] 
lib: prefer `var = time(NULL)` over `time(&var)`

Following up on previous occurrences showing up as gcc warnings, replace
the remaining `time(&var)` calls with `var = time(NULL)`, though these
aren't specifically causing compiler warnings. These are in the TFTP
client code (`lib/tftp.c`), except one which is in a debug branch in
`lib/http_aws_sigv4.c`.

What's unexplainable is that this patch seems to mitigate TFTP tests
often hanging or going into an infinite loop on GHA windows workflows
with MSYS2, mingw-w64 and MSVC (Cygwin is unaffected):
  https://github.com/curl/curl/pull/13599#issuecomment-2119372376
TFTP hangs did not entirely disappear though, so could be unrelated.

`time()` docs:
https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/time-time32-time64
https://manpages.debian.org/bookworm/manpages-dev/time.2.en.html

Follow-up to 58ca0a2f0743a586716ca357c382b29e3f08db69 #13800
Follow-up to d0728c9109629ee82b855b350a4c3f1f52ee61df #13643
Closes #13815

17 months agowinsock: move SO_SNDBUF update into cf-socket
Stefan Eissing [Fri, 24 May 2024 08:09:32 +0000 (10:09 +0200)] 
winsock: move SO_SNDBUF update into cf-socket

- Move the code that updates the SO_SNDBUF size for Windows to
  cf_socket_send.

Prior to this change the code was in readwrite_upload but the socket
filter is the more appropriate place because it applies to all sends.

Background:

For Windows users SO_SNDBUF (the total per-socket buffer size reserved
by Winsock for sends) is updated dynamically by libcurl during the
transfer. This is because Windows does not do it automatically for
non-blocking sockets and without it the performance of large transfers
may suffer.

Closes https://github.com/curl/curl/pull/13763

17 months agosectransp: use common code for cipher suite lookup
Jan Venekamp [Wed, 8 May 2024 18:40:29 +0000 (20:40 +0200)] 
sectransp: use common code for cipher suite lookup

Take advantage of the Curl_cipher_suite_walk_str() and
Curl_cipher_suite_get_str() functions introduced in commit fba9afe.

Closes #13521

17 months agoaws-sigv4: url encode the canonical path
Matthias Gatto [Mon, 27 May 2024 12:58:11 +0000 (14:58 +0200)] 
aws-sigv4: url encode the canonical path

Refactors canon_query, so it could use the encoding part of the function
to use it in the path.

As the path doesn't encode '/', but encode '=', I had to add some
conditions to know If I was doing the query or path encoding.

Also, instead of adding a `bool in_path` variable, I use `bool
*found_equals` to know if the function was called for the query or path,
as found_equals is used only in query_encoding.

Test 472 verifies.

Reported-by: Alexander Shtuchkin
Fixes #13754
Closes #13814

Signed-off-by: Matthias Gatto <matthias.gatto@outscale.com>
17 months agocd2nroff: use an empty "##" to signal end of .IP sequence
Daniel Stenberg [Tue, 28 May 2024 06:32:19 +0000 (08:32 +0200)] 
cd2nroff: use an empty "##" to signal end of .IP sequence

Like when we list a series of options and then want to add "normal" text
again afterwards.

Without this, the indentation level wrongly continues even after the
final "##" header, making following text wrongly appear to belong to the
header above.

Adjusted several curldown files to use this.

Fixes #13803
Reported-by: Jay Satiro
Closes #13806

17 months agoopenssl: fix %-specifier in infof() call
vvb2060 [Tue, 28 May 2024 17:11:44 +0000 (01:11 +0800)] 
openssl: fix %-specifier in infof() call

Closes #13816

17 months agocurl: make warnings and other messages aware of terminal width
Daniel Stenberg [Tue, 28 May 2024 07:44:27 +0000 (09:44 +0200)] 
curl: make warnings and other messages aware of terminal width

This removes unnecessary line wraps when the terminal is wider than 79
columns and it also makes messages look better in narrower terminals.

The get_terminal_columns() function is not split out into its own source
file.

Suggested-by: Elliott Balsley
Fixes #13804
Closes #13808

17 months agoGHA: enable tests 1139, 1177, 1477 on Windows
Viktor Szakats [Tue, 28 May 2024 17:19:43 +0000 (19:19 +0200)] 
GHA: enable tests 1139, 1177, 1477 on Windows

These exclusions came from the AppVeyor CI config, but they do pass now
and they are static tests with no flakiness risk.

Follow-up to 0914d8aadddac0d1459673d5b7f77e8f3378b22b #13759
Closes #13817

17 months agoCI: Improve labeler tag detection
Dan Fandrich [Mon, 13 May 2024 19:58:44 +0000 (12:58 -0700)] 
CI: Improve labeler tag detection

Also, simplify patterns with a single glob.

17 months agoGHA: disable TFTP tests in Windows jobs
Viktor Szakats [Tue, 28 May 2024 11:07:34 +0000 (13:07 +0200)] 
GHA: disable TFTP tests in Windows jobs

Shot in the dark trying to find out which tests are
hanging / going to an infinite loop.

The ones failing after 45 minutes (mingw-w64) or 30 minutes (MSVC).

Ref: https://github.com/curl/curl/pull/13599#issuecomment-2119372376

17 months agoci: update vmactions/omnios-vm digest to a61ca1e
renovate[bot] [Mon, 27 May 2024 20:56:52 +0000 (20:56 +0000)] 
ci: update vmactions/omnios-vm digest to a61ca1e

Closes #13801

17 months agoopenssl/gnutls: rectify the TLS version checks for QUIC
Daniel Stenberg [Mon, 27 May 2024 21:12:27 +0000 (23:12 +0200)] 
openssl/gnutls: rectify the TLS version checks for QUIC

The versions check wrongly complained and return error if the *minimum*
version was set to something less than 1.3. QUIC is always TLS 1.3, but
that means minimum 1.2 is still fine to ask for.

This also renames the local variable to make the mistake harder to make
in the future.

Regression shipped in 8.8.0

Follow-up to 3210101088dfa3d6a125

Reported-by: fds242 on github
Fixes #13799
Closes #13802

17 months agognutls: improve TLS shutdown
Stefan Eissing [Mon, 27 May 2024 09:51:03 +0000 (11:51 +0200)] 
gnutls: improve TLS shutdown

local ftp upload tests sometimes failed with an invalid TLS record being
reported by gnutls. vsftp did log that the shutdown was not regarded as
clean, failing the control connection thereafter.

These changes make test_31_05 work reliable locally.

- on closing the SSL filter, shutdown READ *and* WRITE
- on closing, try a receive after shutdown is sent
- convert to DEBUGF to CURL_TRC_CF

Closes #13790

17 months agoRELEASE-NOTES: synced
Daniel Stenberg [Tue, 28 May 2024 08:29:05 +0000 (10:29 +0200)] 
RELEASE-NOTES: synced

17 months agotests: run with "--trace-config all" to provide even more info
Daniel Stenberg [Mon, 27 May 2024 12:06:37 +0000 (14:06 +0200)] 
tests: run with "--trace-config all" to provide even more info

in case of problems.

Closes #13791

17 months agobuild: untangle `CURLDEBUG` and `DEBUGBUILD` macros
Viktor Szakats [Mon, 20 May 2024 12:21:05 +0000 (14:21 +0200)] 
build: untangle `CURLDEBUG` and `DEBUGBUILD` macros

`CURLDEBUG` is meant to enable memory tracking, but in a bunch of cases,
it was protecting debug features that were supposed to be guarded with
`DEBUGBUILD`.

Replace these uses with `DEBUGBUILD`.

This leaves `CURLDEBUG` uses solely for its intended  purpose: to enable
the memory tracking debug feature.

Also:
- autotools: rely on `DEBUGBUILD` to enable `checksrc`.
  Instead of `CURLDEBUG`, which worked in most cases because debug
  builds enable `CURLDEBUG` by default, but it's not accurate.
- include `lib/easyif.h` instead of keeping a copy of a declaration.
- add CI test jobs for the build issues discovered.

Ref: https://github.com/curl/curl/pull/13694#issuecomment-2120311894
Closes #13718

17 months agoexamples: delete unused includes
Viktor Szakats [Fri, 24 May 2024 17:52:05 +0000 (19:52 +0200)] 
examples: delete unused includes

Delete a bunch of unnecessary-looking headers from some examples. This
is known to be tricky on AIX (perhaps also in other less-tested envs).

Let me know if any of this looks incorrect or outright fails on some
systems.

Follow-up to d4b85890555388bec212b75f47a5c1a48705b156 #13771
Closes #13785

17 months agoappveyor: fixup job name [ci skip]
Viktor Szakats [Mon, 27 May 2024 21:22:39 +0000 (23:22 +0200)] 
appveyor: fixup job name [ci skip]

Follow-up to fc8e0dee3045658f293452121f5290d81ba3aa1e #13694

17 months agocmake: fix `-Wredundant-decls` in unity/mingw-w64/gcc/curldebug/DLL builds
Viktor Szakats [Sun, 19 May 2024 14:55:45 +0000 (16:55 +0200)] 
cmake: fix `-Wredundant-decls` in unity/mingw-w64/gcc/curldebug/DLL builds

It affected cmake-unity shared-curltool curldebug mingw-w64 gcc builds
when building the `testdeps` target.

Apply the solution already used in `lib/base64.c` and `lib/dynbuf.c`
to fix it.

Also update an existing GHA CI job to test the issue fixed.

```
In file included from curl/lib/version_win32.c:35,
                 from curl/_bld/src/CMakeFiles/curl.dir/Unity/unity_0_c.c:145:
curl/lib/memdebug.h:52:14: error: redundant redeclaration of 'curl_dbg_logfile' [-Werror=redundant-decls]
   52 | extern FILE *curl_dbg_logfile;
      |              ^~~~~~~~~~~~~~~~
In file included from curl/src/slist_wc.c:32,
                 from curl/_bld/src/CMakeFiles/curl.dir/Unity/unity_0_c.c:4:
curl/lib/memdebug.h:52:14: note: previous declaration of 'curl_dbg_logfile' with type 'FILE *' {aka 'struct _iobuf *'}
   52 | extern FILE *curl_dbg_logfile;
      |              ^~~~~~~~~~~~~~~~
curl/lib/memdebug.h:55:44: error: redundant redeclaration of 'curl_dbg_malloc' [-Werror=redundant-decls]
   55 | CURL_EXTERN ALLOC_FUNC ALLOC_SIZE(1) void *curl_dbg_malloc(size_t size,
      |                                            ^~~~~~~~~~~~~~~
curl/lib/memdebug.h:55:44: note: previous declaration of 'curl_dbg_malloc' with type 'void *(size_t,  int,  const char *)' {aka 'void *(long long unsigned int,  int,  const char *)'}
   55 | CURL_EXTERN ALLOC_FUNC ALLOC_SIZE(1) void *curl_dbg_malloc(size_t size,
      |                                            ^~~~~~~~~~~~~~~
[...]
curl/lib/memdebug.h:110:17: error: redundant redeclaration of 'curl_dbg_fclose' [-Werror=redundant-decls]
  110 | CURL_EXTERN int curl_dbg_fclose(FILE *file, int line, const char *source);
      |                 ^~~~~~~~~~~~~~~
curl/lib/memdebug.h:110:17: note: previous declaration of 'curl_dbg_fclose' with type 'int(FILE *, int,  const char *)' {aka 'int(struct _iobuf *, int,  const char *)'}
  110 | CURL_EXTERN int curl_dbg_fclose(FILE *file, int line, const char *source);
      |                 ^~~~~~~~~~~~~~~
```
Ref: https://ci.appveyor.com/project/curlorg/curl/builds/49840554/job/a4aoet17e9qnqx1a#L362

After: https://ci.appveyor.com/project/curlorg/curl/builds/49843735/job/hbo2uah2vj0ns523

Ref: #13689 (CI testing this PR with `DEBUGBUILD`/`CURLDEBUG`/shared-static combinations)
Depends-on: #13694
Depends-on: #13800
Closes #13705

17 months agolib: fix gcc warning in certain debug builds
Viktor Szakats [Mon, 20 May 2024 12:21:05 +0000 (14:21 +0200)] 
lib: fix gcc warning in certain debug builds

```
curl/lib/http_aws_sigv4.c:536:10: error: 'clock' may be used uninitialized [-Werror=maybe-uninitialized]
  536 |   time_t clock;
      |          ^~~~~
```
Ref: https://github.com/curl/curl/actions/runs/9158755123/job/25177765000#step:13:79

Cherry-picked from #13718
Closes #13800

17 months agocmake: always build unit tests with the `testdeps` target
Viktor Szakats [Sat, 18 May 2024 00:15:32 +0000 (02:15 +0200)] 
cmake: always build unit tests with the `testdeps` target

Before this patch, the `testdeps` build target required `-DCURLDEBUG`
be set either via `ENABLE_DEBUG=ON` or `ENABLE_CURLDEBUG=ON` to build
the curl unit tests.

After fixing build issues in #13694, we can drop this requirement and
build unit tests unconditionally.

Depends-on: #13694
Depends-on: #13697 (fix unit test issue revealed by Old Linux CI job)
Follow-up to 39e7c22bb459c2e818f079984989a26a09741860 #11446
Closes #13698

17 months agoCI: disable dependency tracking in most autotools builds
Viktor Szakats [Mon, 27 May 2024 13:19:35 +0000 (15:19 +0200)] 
CI: disable dependency tracking in most autotools builds

For better build performance. Dependency tracking causes a build
overhead while compiling to help a subsequent build, but in CI there is
never one and the extra work is discarded.

Closes #13794

17 months agobuild: untangle `UNITTESTS` and `DEBUGBUILD` macros
Viktor Szakats [Thu, 16 May 2024 11:49:22 +0000 (13:49 +0200)] 
build: untangle `UNITTESTS` and `DEBUGBUILD` macros

- fix `DEBUGBUILD` guards that should be `UNITTESTS`, in libcurl code
  used by unit tests.
- fix guards for libcurl functions used in unit tests only.
- sync `UNITTEST` attribute between declarations and definitions.
- drop `DEBUGBUILD` guard from test `unit2600`.
- fix guards for libcurl HSTS code used by both a unit test (`unit1660`)
  and `test0446`.
- update an existing AppVeyor CI job to test the issues fixed.

This fixes building tests with `CURLDEBUG` enabled but `DEBUGBUILD`
disabled. This can happen when building tests with CMake with
`ENABLE_DEBUG=ON` in Release config, or with `ENABLE_CURLDEBUG=ON`
and _without_ `ENABLE_DEBUG=ON`. Possibly also with autotools
when using `--enable-curldebug` without `--enable-debug`.

Test results:
- before:
  https://ci.appveyor.com/project/curlorg/curl/builds/49835609
  https://ci.appveyor.com/project/curlorg/curl/builds/49898529/job/k8qpbs8idby70smw
  https://github.com/curl/curl/actions/runs/9259078835/job/25470318167?pr=13798#step:13:821
- after: https://ci.appveyor.com/project/curlorg/curl/builds/49839255
  (the two failures are unrelated, subject to PR #13705)

Ref: #13592 (issue discovery)
Ref: #13689 (CI testing this PR with `DEBUGBUILD`/`CURLDEBUG` combinations)
Closes #13694

17 months agoGHA: ignore flaky MQTT and FTP test results [ci skip]
Viktor Szakats [Mon, 27 May 2024 17:55:20 +0000 (19:55 +0200)] 
GHA: ignore flaky MQTT and FTP test results [ci skip]

MQTT / OmniOS:
```
TESTFAIL: These test cases failed: 1190 1198 3017
```
Ref: https://github.com/curl/curl/actions/runs/9258522297/job/25468730731?pr=13694#step:3:10251

MQTT / OmniOS:
```
TESTFAIL: These test cases failed: 1194 2200 2203 2205
```
Ref: https://github.com/curl/curl/actions/runs/9150523540/job/25155409832#step:3:10233

FTP / OmniOS:
```
TESTFAIL: These test cases failed: 1096
```
Ref: https://github.com/curl/curl/actions/runs/9150702711/job/25155793948#step:3:10247

FTP / OmniOS:
```
TESTFAIL: These test cases failed: 381
```
Ref: https://github.com/curl/curl/actions/runs/9163863822/job/25193897640#step:3:10230

FTP / OmniOS:
```
TESTFAIL: These test cases failed: 340
```
Ref: https://github.com/curl/curl/actions/runs/9233804752/job/25406671742?pr=13771#step:3:10245

Ref: https://github.com/curl/curl/pull/13583#issuecomment-2119376898

17 months agoCI: tidy up skipping tests build/run in Windows jobs
Viktor Szakats [Mon, 27 May 2024 16:48:32 +0000 (18:48 +0200)] 
CI: tidy up skipping tests build/run in Windows jobs

Simplify controlling whether to build and/run tests in a CI job.

Apply the TFLAGS='skipall' (do not build nor run tests) or
'skiprun' (build, but do not run) method already used with old-mingw-w64
and msvc jobs to existing Windows jobs in GHA and AppVeyor.

Also:
- add Cygwin/cmake test build and run steps while here.
- replace `DISABLED_TESTS` with `TFLAGS` in AppVeyor.

Closes #13796

17 months agocmake: use `APPLE` instead of `CMAKE_SYSTEM_NAME` string
Viktor Szakats [Sun, 19 May 2024 16:42:11 +0000 (18:42 +0200)] 
cmake: use `APPLE` instead of `CMAKE_SYSTEM_NAME` string

Follow-up to a86254b39307af1a53735b065a382567805cd9b8 #12515
Closes #13713

17 months agocmake: whitespace, formatting/tidy-up in comments
Viktor Szakats [Sun, 19 May 2024 16:49:42 +0000 (18:49 +0200)] 
cmake: whitespace, formatting/tidy-up in comments

Also correct casing in a few option descriptions.

Closes #13711

17 months agocmake: allow `ENABLE_CURLDEBUG=OFF` with `ENABLE_DEBUG=ON`
Viktor Szakats [Mon, 27 May 2024 12:33:54 +0000 (14:33 +0200)] 
cmake: allow `ENABLE_CURLDEBUG=OFF` with `ENABLE_DEBUG=ON`

Before this patch, `ENABLE_CURLDEBUG` (memory tracking) was
unconditionally enabled when `ENABLE_DEBUGBUILD` was set. This made
testing some build configurations complicated. To fix it, this patch
makes `ENABLE_CURLDEBUG` to receive the value of `ENABLE_DEBUG` by
default, while allowing free override by the user.

This allows to use the config:
`ENABLE_DEBUGBUILD=ON ENABLE_CURLDEBUG=OFF`
to enable debug features, without also enabling memory tracking.

This is important because some other build methods allow to set one of
these features but not the other. This patch allows to test any
combination with CMake.

This makes it unnecessary to use the workaround of passing
`-DDEBUGBUILD` via `CMAKE_C_FLAGS`. Which has the disadvantage that our
CMake logic cannot easily detect it, e.g. for disabling symbol hiding on
Windows for `ENABLE_DEBUG`/`DEBUGBUILD` builds.

Cherry-picked from #13718
Closes #13792

17 months agocmake: `ENABLE_DEBUG=ON` to always set `-DDEBUGBUILD`
Viktor Szakats [Sat, 11 May 2024 00:44:10 +0000 (02:44 +0200)] 
cmake: `ENABLE_DEBUG=ON` to always set `-DDEBUGBUILD`

Before this patch `ENABLE_DEBUG=ON` always enabled the TrackMemory
(aka `ENABLE_CURLDEBUG=ON`) feature, but required the `Debug` CMake
configration to actually enable curl debug features
(aka `-DDEBUGBUILD`).

Curl debug features do not require compiling with C debug options. This
also made enabling debug features unintuitive and complicated to use.
Due to other issues (subject to PR #13694) it also caused an error in
default (and `Release`/`MinSizeRel`/`RelWithDebInfo`) configs, when
building the `testdeps` target:
```
ld: CMakeFiles/unit1395.dir/unit1395.c.o: in function `test':
unit1395.c:(.text+0x1a0): undefined reference to `dedotdotify'
```
Ref: https://github.com/curl/curl/actions/runs/9037287098/job/24835990826#step:3:2483

Fix it by always defining `DEBUGBUILD` when setting `ENABLE_DEBUG=ON`.
Decoupling this option from the selected CMake configuration.

Note that after this patch `ENABLE_DEBUG=ON` unconditionally enables
curl debug features. These features are insecure and unsuited for
production. Make sure to omit this option when building for production
in default, `Release` (and other not-`Debug`) modes.

Also delete a workaround no longer necessary in GHA CI jobs.

Ref: 1a62b6e68c08c7e471ff22dd92932aba7e026817 (2015-03-03)
Ref: #13583
Closes #13592

17 months agoGHA: add autotools mingw-64, build-only job
Viktor Szakats [Mon, 27 May 2024 13:07:23 +0000 (15:07 +0200)] 
GHA: add autotools mingw-64, build-only job

Cherry-picked from #13718
Closes #13793

17 months agoGHA: add three MSVC jobs
Viktor Szakats [Fri, 24 May 2024 15:53:27 +0000 (17:53 +0200)] 
GHA: add three MSVC jobs

Continuing the theme, add 3 MSVC jobs with tests, matching
configurations used on AppVeyor. MSVC versions are identical:
19.39.33523.0 + Windows SDK 10.0.22621.0.

Also enable websockets, and build examples. Tests are run in parallel
(`-j14`), with improved performance.

Job performance:
```
                                                       AppVeyor  GHA
                                                                 w/examples
                                                       --------  ----------
CMake, VS2022, Debug, x64, Schannel, Static, Unicode   38m 4s    11m57s
CMake, VS2022, Debug, x64, no SSL, Static              35m15s    12m 6s
CMake, VS2022, Debug, x64, no SSL, Static, HTTP only   25m25s    10m36s
```
Based on these runs:
https://ci.appveyor.com/project/curlorg/curl/builds/49884748
https://github.com/curl/curl/actions/runs/9229448468

This is the first time examples are built in CI with MSVC: Fix all
warnings and errors that came up via
d4b85890555388bec212b75f47a5c1a48705b156 #13771.

Closes #13766

17 months agoGHA: add three old (gcc 6, 7, 9) mingw-w64 jobs
Viktor Szakats [Thu, 23 May 2024 12:54:49 +0000 (14:54 +0200)] 
GHA: add three old (gcc 6, 7, 9) mingw-w64 jobs

Re-implement old mingw-w64 jobs in GHA. This allows to use the latest
Windows runners, replacing Windows Server 2012 R2 (gcc 6) and Windows
Server 2016 (gcc 7, 9) with Windows Server 2022.

GHA runners are also significantly faster, and allow running tests in
parallel (`-j14`). It also offloads 3 more long-running jobs from
AppVeyor CI.

These jobs download (then cache) the mingw-w64 packages from their
original location, which allows flexibility in choosing which versions
and flavours (win32/POSIX, SEH/DWARF, 64/32-bit) we want to test in CI.
The new jobs use these distros:
- https://sourceforge.net/projects/mingw-w64/files/Toolchains%20targetting%20Win64/Personal%20Builds/mingw-builds/ (for gcc 7, same as on AppVeyor)
- https://sourceforge.net/projects/mingw-w64/files/Toolchains%20targetting%20Win32/Personal%20Builds/mingw-builds/ (for gcc 6, same as on AppVeyor)
- https://winlibs.com/ (for gcc 9)

I matched existing AppVeyor job configs, with these differences:
- gcc 6.4.0 instead of 6.3.0.
  (same distro as on AppVeyor, but the latest bugfix release)
- gcc 9.5.0 instead of 9.1.0 and a different (but compatible) binary distro.
  (in AppVeyor this relies on an old MSYS2 pre-installed on the runner)
- using win32 builds instead of posix for gcc 6.4.0 and 7.3.0.
- websockets enabled.
- always build examples.
- always build tests (this wasn't done for 6.4.0 with AppVeyor CI).

I did not replicate existing test exclusions, and oddly enough the few
failures (so far) were different from MSYS2 jobs and also from their
AppVeyor CI counterparts.

Also:
- delete redundant (default) `-u` option from `cygpath` calls.
- allow matrix options to override default ones in CMake.
- detect and use Windows-supplied curl for `TFLAGS` `-ac` option.
  (it's available in modern runners.)
- delete the 3 AppVeyor CI jobs now replicated in GHA.
- appveyor: prefer `SYSTEMROOT` over `WINDIR`.
- tidy-up quotes.

Job performance:
```
                                                                 AppVeyor  GHA
                                                                           w/examples
                                                                           w/tests
                                                                 --------  ----------
CMake, mingw-w64, gcc 6, Debug, x86, Schannel, Static, no-unity   1m25s     8m50s
CMake, mingw-w64, gcc 7, Debug, x64, Schannel, Static, Unicode   31m45s     9m39s
CMake, mingw-w64, gcc 9, Debug, x64, Schannel, Static            28m25s    13m38s
```
Based on these runs:
https://ci.appveyor.com/project/curlorg/curl/builds/49880799
https://github.com/curl/curl/actions/runs/9218292508

Notice that building examples and tests is time consuming.

We can tweak any build parameter as necessary to make them more useful
and/or without clogging the job queue or introducing flakiness.

Closes #13759

17 months agoTODO: remove some old, clarify, add something
Daniel Stenberg [Mon, 27 May 2024 08:38:43 +0000 (10:38 +0200)] 
TODO: remove some old, clarify, add something

Closes #13788

17 months agoTODO: Add "Share CA cache" + "CA caching to more TLS backends"
Daniel Stenberg [Mon, 27 May 2024 08:13:40 +0000 (10:13 +0200)] 
TODO: Add "Share CA cache" + "CA caching to more TLS backends"

Closes #13787

17 months agoruntests: sort test IDs in summary lines
Viktor Szakats [Sat, 25 May 2024 06:54:27 +0000 (08:54 +0200)] 
runtests: sort test IDs in summary lines

Changing this output:
```
TESTFAIL: These test cases failed: 2301 2303 2302 2307
```
Ref: https://github.com/curl/curl/actions/runs/9228638364/job/25393106631#step:6:21181

To:
```
TESTFAIL: These test cases failed: 2301 2302 2303 2307
```

Cherry-picked from #13766
Closes #13774

17 months agoexamples: fix compiling with MSVC
Viktor Szakats [Fri, 24 May 2024 16:38:32 +0000 (18:38 +0200)] 
examples: fix compiling with MSVC

- `websocket.c`: use `Sleep()` on Windows.
  `sleep()` and `unistd.h` are not available in MSVC.

- `http2-upload.c`: use local `gettimeofday()` implementation when
  compiled with MSVC.
  (Alternate solution is to disable the trace function for MSVC.)
  Public domain code copied and adapted from libssh2:
  https://github.com/libssh2/libssh2/blob/e973493f992313b3be73f51d3f7ca6d52e288558/src/misc.c#L719-L743

- silence compiler warning for deprecated `inet_addr()`.
  Also drop duplicate winsock2 include.
  ```
  curl\docs\examples\externalsocket.c(125,32): error C2220: the following warning is treated as an error [curl\bld\docs\examples\curl-example-externalsocket.vcxproj]
  curl\docs\examples\externalsocket.c(125,32): warning C4996: 'inet_addr': Use inet_pton() or InetPton() instead or define _WINSOCK_DEPRECATED_NO_WARNINGS to disable deprecated API warnings [curl\bld\docs\examples\curl-example-e
  ```
  Ref: https://github.com/curl/curl/actions/runs/9227337318/job/25389073450#step:4:95

- silence an MSVC compiler warning. This is in conflict with `checksrc`
  rules, so silence the rule in favour of the warning-free C syntax.
  ```
  curl\docs\examples\multi-legacy.c(152,1): error C2220: the following warning is treated as an error [curl\bld\docs\examples\curl-example-multi-legacy.vcxproj]
  curl\docs\examples\multi-legacy.c(152,1): warning C4706: assignment within conditional expression [curl\bld\docs\examples\curl-example-multi-legacy.vcxproj]
  ```
  Ref: https://github.com/curl/curl/actions/runs/9227337318/job/25389073450#step:4:226

- do not use `sys/time.h` and `unistd.h` in Windows builds.
  Some of these includes look unnecessary. Subject to another PR.

Cherry-picked from #13766
Closes #13771

17 months agodocs/cmdline-opts: fix mail-auth example TLD typo
Jonathan Matthews [Sun, 26 May 2024 18:50:44 +0000 (19:50 +0100)] 
docs/cmdline-opts: fix mail-auth example TLD typo

Closes: #13784
Reviewed-by: Daniel Gustafsson <daniel@yesql.se>
17 months agolibssh: remove CURLOPT_SSL_VERIFYHOST check
Daniel Stenberg [Sat, 25 May 2024 21:30:51 +0000 (23:30 +0200)] 
libssh: remove CURLOPT_SSL_VERIFYHOST check

It was never meant for SSH: it should rely on the knownhosts file (if
set) in the same way libssh2 already does.

Reported-by: James Abbatiello
Fixes #13767
Closes #13781

17 months agomulti: add multi->proto_hash, a key-value store for protocol data
Stefan Eissing [Thu, 11 Apr 2024 10:34:40 +0000 (12:34 +0200)] 
multi: add multi->proto_hash, a key-value store for protocol data

- add `Curl_hash_add2()` that passes a destructor function for
  the element added. Call element destructor instead of hash
  destructor if present.
- multi: add `proto_hash` for protocol related information,
  remove `struct multi_ssl_backend_data`.
- openssl: use multi->proto_hash to keep x509 shared store
- schannel: use multi->proto_hash to keep x509 shared store
- vtls: remove Curl_free_multi_ssl_backend_data() and its
  equivalents in the TLS backends

Closes #13345

17 months agotests: add pytest for --ciphers and --tls13-ciphers options
Jan Venekamp [Sat, 25 May 2024 12:56:53 +0000 (14:56 +0200)] 
tests: add pytest for --ciphers and --tls13-ciphers options

Closes #13530

17 months agotool_operate: avoid explicitly setting verifypeer to 1
Orgad Shaneh [Sun, 19 May 2024 08:58:44 +0000 (11:58 +0300)] 
tool_operate: avoid explicitly setting verifypeer to 1

Also for the proxy verison. It is the default, just like verifyhost,
since a long time.

Closes #13704

17 months agotests: extend user/password parsing test1620
Orgad Shaneh [Thu, 23 May 2024 07:01:38 +0000 (10:01 +0300)] 
tests: extend user/password parsing test1620

Closes #13756

17 months agoconfigure: use `$EGREP` in place of `grep -E`
Alejandro R. Sedeño [Sat, 25 May 2024 16:04:18 +0000 (12:04 -0400)] 
configure: use `$EGREP` in place of `grep -E`

`$EGREP` is set based on an earlier test in configure so that we can
work with systems that have `egrep` and a `grep` that does not support
`-E`.

Closes #13780

17 months agoci: update dependency awslabs/aws-lc to v1.28.0
renovate[bot] [Fri, 24 May 2024 19:22:28 +0000 (19:22 +0000)] 
ci: update dependency awslabs/aws-lc to v1.28.0

Closes #13770

17 months agotests: test_17_ssl_use.py clarify mbedtls TLSv1.3 support
Jan Venekamp [Sat, 25 May 2024 13:07:58 +0000 (15:07 +0200)] 
tests: test_17_ssl_use.py clarify mbedtls TLSv1.3 support

Closes #13779

17 months agohttp: write last header line late
Stefan Eissing [Thu, 23 May 2024 10:21:46 +0000 (12:21 +0200)] 
http: write last header line late

- HEADERFUNCTIONS might inspect response properties like
  CURLINFO_CONTENT_LENGTH_DOWNLOAD_T on seeing the last header line. If
  the line is being written before this is initialized, values are not
  available.

- write the last header line late when analyzing a HTTP response so that
  all information is available at the time of the writing.

- add test1485 to verify that CURLINFO_CONTENT_LENGTH_DOWNLOAD_T works
  on seeing the last header.

Fixes #13752
Reported-by: Harry Sintonen
Closes #13757

17 months agotests: use exec when spawning nghttpx
Dan Fandrich [Fri, 24 May 2024 21:46:19 +0000 (14:46 -0700)] 
tests: use exec when spawning nghttpx

This stops keeping perl and shell processes around that are no longer
needed, plus it eliminates an unneeded shell message when the server is
later terminated.

Closes #13772

17 months agoGHA: ignore flaky test 3017 (MQTT) on OpenBSD
Viktor Szakats [Fri, 24 May 2024 21:47:15 +0000 (23:47 +0200)] 
GHA: ignore flaky test 3017 (MQTT) on OpenBSD

```
TESTFAIL: These test cases failed: 3017
```
Ref: https://github.com/curl/curl/actions/runs/9223543272/job/25376999226?pr=13759#step:3:16326
Ref: https://github.com/curl/curl/actions/runs/9230183764/job/25397883193?pr=13766#step:3:16345

Ref: https://github.com/curl/curl/pull/13583#issuecomment-2119376898