To reduce the amount of allocations needed for creating a Curl_addrinfo
struct, make a single larger malloc instead of three separate smaller
ones.
Closes#5533
When the method is updated inside libcurl we must still not change the
method as set by the user as then repeated transfers with that same
handle might not execute the same operation anymore!
This fixes the libcurl part of #5462
Test 1633 added to verify.
Closes#5499
This test verifies the -w %json output and the test case includes a full
generated "blob". If there's no proxy support built into libcurl, it
will return an error for proxy related info variables and they will not
be included in the json, thus causing a mismatch and this test fails.
Reported-by: Marc Hörsken
Fixes#5501Closes#5502
... and free it as soon as the transfer is done. It removes the extra
alloc when a new size is set with setopt() and reduces memory for unused
easy handles.
In addition: the closure_handle now doesn't use an allocated buffer at
all but the smallest supported size as a stack based one.
Closes#5472
GCC 10 warns about this with warning: implicit conversion
from 'SANITIZEcode' to 'CURLcode' [-Wenum-conversion]
Since 'expected_result' is not really of type 'CURLcode' and
it is not exposed in any way, we can just use 'SANITIZEcode'.
Reviewed-by: Daniel Stenberg
Reviewed-by: Marcel Raad
Closes#5476
Since curl_setup.h now makes use of curlx_win32_fopen for Win32
builds with USE_WIN32_LARGE_FILES or USE_WIN32_SMALL_FILES defined,
we need to include the relevant files for tests using fopen,
because the libtest sources are also including curl_setup.h
Reviewed-by: Marcel Raad
Reviewed-by: Daniel Stenberg
Follow up to #3784 (ffdddb45d9)
Closes#5475
They're only limited to the maximum string input restrictions, not to
256 bytes.
Added test 1178 to verify
Reported-by: Will Roberts
Fixes#5448Closes#5449
Fixed the alt-svc parser to treat a newline as end of line.
The unit tests in test 1654 were done without CRLF and thus didn't quite
match the real world. Now they use CRLF as well.
Reported-by: Peter Wu
Assisted-by: Peter Wu
Assisted-by: Jay Satiro
Fixes#5445Closes#5446
curl would previously show "curl: Saved to filename 'name from header'"
if -J was used and a name was picked from the Content-Disposition
header. That output could interfer with other stdout output, such as -w.
This commit removes that output line.
Bug: https://curl.haxx.se/mail/archive-2020-05/0044.html
Reported-by: Коваленко Анатолий Викторович
Closes#5435
- Stick to a single unified way to use structs
- Make checksrc complain on 'typedef struct {'
- Allow them in tests, public headers and examples
- Let MD4_CTX, MD5_CTX, and SHA256_CTX typedefs remain as they actually
typedef different types/structs depending on build conditions.
Closes#5338
If it can't be found in the request. Also support --cmdfile to set it to
a custom file name.
runtests.pl always writes this file with the test number in it since a
while back.
Added test 971 to verify that the list is in sync with the files in
cmdline-opts. The check also verifies that .d-files that uses Added:
specify the same version number as the options-in-versions file does.
Closes#5381
* runtests.pl:
- Fix out-of-tree build under CMake when srcdir is not set. Default
srcdir to the location of runtests.pl.
- Add a hack to allow CMake to use the TFLAGS option as documented
in tests/README and used in scripts/travis/script.sh.
* Bump CMake version to 3.2 for USES_TERMINAL, dropping Debian Jessie
support (no one should care, it is already EOL.).
* Remove CTest since it defines its own 'test' target with no tests
since all unittests are already broken and not built by default.
* Add new test targets based on the options from Makefile.am. Since
new test targets are rarely added, I opted for duplicating the
runtests.pl options as opposed to creating a new Makefile.inc file.
Use top-level target names (test-x) instead of x-test since that is
used by CI and others.
Closes#5358
The default target should only build libcurl and curl. Add a dedicated
'testdeps' target which will be used later when running tests. Note that
unittests are currently broken in CMake and already excluded.
Closes#5368
The tftpd server may still be busy if the total timeout of
25 seconds has not been reached or no sread error was received
during or after the execution of the timeout test 1238.
Once the next TFTP test comes around (eg. 1242 or 1243),
those will fail because the tftpd server is still waiting
on data from curl due to the UDP protocol being stateless
and having no connection close. On Linux this error may not
happen, because ICMP errors generated due to a swrite error
can also be returned async on the next sread call instead.
Therefore we will now just kill the tftpd server after test
1238 to make sure that the following tests are not affected.
This enables us to no longer ignore tests 1242, 1243, 2002
and 2003 on the CI platforms CirrusCI and AppVeyor.
Assisted-by: Peter Wu
Closes#5364
setjmp.h should only be included if HAVE_SETJMP_H is defined.
Add additional log statements to see wether reads and writes
are blocking or finishing before an alarm signal is received.
Assisted-by: Peter Wu
Part of #5364
OpenSSH for Windows shows group and other/world permissions as *,
because those concepts do not exist on Windows. It also does not
show the current or parent directory, so we just ignore those.
Reviewed-by: Daniel Stenberg
Closes#5328
A common set of functions instead of many separate implementations for
creating buffers that can grow when appending data to them. Existing
functionality has been ported over.
In my early basic testing, the total number of allocations seem at
roughly the same amount as before, possibly a few less.
See docs/DYNBUF.md for a description of the API.
Closes#5300
Remove many one second sleeps that were done *after* each newly started
test server already has been verified. They should not have any purpose
there.
Closes#5323
This makes sure that select_ws behaves similar to real select
which stops waiting on a signal handler being triggered.
This makes it possible to gracefully stop sockfilt.exe on
Windows with taskkill /IM sockfilt.exe (without /F force flag).
Reviewed-by: Jay Satiro
Part of #5260
This commit adds a global exit event to the test servers that
Windows-specific wait routines can use to get triggered if the
program was signaled to be terminated, eg. select_ws in sockfilt.c
The exit event will be managed by the signal handling code and is
set to not reset automatically to support multiple wait routines.
Reviewed-by: Jay Satiro
Closes#5260
Use raise to trigger signal handler instead of calling it
directly and causing potential unexpected control flow.
Reviewed-by: Jay Satiro
Part of #5260
OpenSSH for Windows requires paths in the format of /C:/
instead of the pseudo-POSIX paths /cygdrive/c/ or just /c/
Reviewed-by: Daniel Stenberg
Closes#5298
Several tests referred to specific server ports even when the test
didn't actually use that server or specify that it's needed. In such
cases, the test harness substitutes the text "[not running]" as the port
number which causes many such tests to fail due to the inability to
parse the URL. These tests are changed to use %NOLISTENPORT which will
always be substituted correctly.
More connection cache accesses are protected by locks.
CONNCACHE_* is a beter prefix for the connection cache lock macros.
Curl_attach_connnection: now called as soon as there's a connection
struct available and before the connection is added to the connection
cache.
Curl_disconnect: now assumes that the connection is already removed from
the connection cache.
Ref: #4915Closes#5009
As stunnel is an external tool and it has no specific option to export
the actually used port number when asked to listen to 0, runtests
instead iterates over ten randomly picked high number ports and sticks
to the first one stunnel can listen to.
Closes#5267
These were commented out in e9dd099870 when Makefile.inc was included
instead. 11 years have passed since then and the commented code is of
course very outdated. Remove it to avoid confusion.
Closes#5311
Since the http2 test server is a mere proxy that needs to know about the
dynamic port the HTTP server is using, it too needs to get restarted
when the http server is killed.
A regression caused by 80d6515.
Fixes#5289Closes#5291
Prior to this change if there was a 303 reply to a PUT request then
the subsequent request to respond to that redirect would also be a PUT.
It was determined that was most likely incorrect based on the language
of the RFCs. Basically 303 means "see other" resource, which implies it
is most likely not the same resource, therefore we should not try to PUT
to that different resource.
Refer to the discussions in #5237 and #5248 for more information.
Fixes https://github.com/curl/curl/issues/5237
Closes https://github.com/curl/curl/pull/5248
The purpose with this variable is to provide a port number that is
reasonably likely to not have a listener on the local host so that tests
can try connect failures against it. It uses port 47 - "reserved"
according to IANA.
Updated six tests to use it instead of the previous different ports.
Assisted-by: Emil Engler
Closes#5270
This fixes the error: 'for' loop initial declaration used outside C99
mode by declaring the loop increment variable in the beginning of the
block instead of inside the for loop.
Fixes#5254
Reviewed-by: Daniel Gustafsson <daniel@yesql.se>
FTP, IMAP, POP3, SMTP and their IPv6 versions are now all on dynamic
ports
Test 842-845 are unfortunately a bit hard to move over to this concept
right now and require "default port" still...
The runtests script now always performs variable replacement on the
entire test source file before the test gets executed, and saves the
updated version in a temporary file (log/test[num]) so that all test
case readers/servers can use that version (if present) and thus enjoy
the powers of test case variable substitution.
This is necessary to allow complete port number freedom.
Test 309 is updated to work with a non-fixed port number thanks to this.
Since the Windows versions of both programs would write Windows
PIDs to their pidfiles which we cannot handle, we need to use
our known perl.exe Cygwin/msys PID together with exec() in order
to tie the spawned processes to the existance of our perl.exe
The perl.exe that is executing secureserver.pl and sshserver.pl
has a Cygwin/msys PID, because it is started inside Cygwin/msys.
Related to #5188
- Ignore intermediate progress updates.
- Support locales that use a character other than period as decimal
separator (eg 100,0%).
test1148 checks that the progress finishes at 100% and has the right
bar width. Prior to this change the test assumed that the only progress
reported for such a quick transfer was 100%, however in rare instances
(like in the CI where transfer time can slow considerably) there may be
intermediate updates. For example, below is stderrlog1148 from a failed
CI run with explicit \r and \n added (it is one line; broken up so that
it's easier to understand).
\r
\r################################## 48.3%
\r######################################################################## 100.0%
\n
Closes https://github.com/curl/curl/pull/5194
With this bit set, my mac successfully resolves "ip6-localhost" when in
fact there is no such host known to my machine! That in turn made test
241 wrongly execute and fail.
Closes#5202
test1294: "split request" being when the entire request isn't sent in
the first go, and the remainder is sent in the PERFORM state. A GET
request is otherwise not sending anything during PERFORM.
test1295: same kind of split but with POST
Closes#5197
This change inverse the order of processing for the --etag-compare and
--etag-save option to process first --etag-compare. This in turn allows
to use the same file name to compare and save an etag.
The original behavior of not failing if the etag file does not exists is
conserved.
Fixes#5179Closes#5180
Replace the incomplete workaround regarding FD_CLOSE
only signalling once by instead doing a pre-check with
standard select and storing the result for later use.
select keeps triggering on closed sockets on Windows while
WSAEventSelect fires only once with data still available.
By doing the pre-check we do not run in a deadlock
due to waiting forever for another FD_CLOSE event.
Fix race-condition of waiting threads finishing while events are
already being processed which lead to invalid or skipped events.
Use mutex to check for one event at a time or do post-processing.
In addition to mutex-based locking use specific event as signal.
Closes#5156
Load long values correctly (e.g. for http_code).
Use curl_off_t (not long) for:
- size_download (CURLINFO_SIZE_DOWNLOAD_T)
- size_upload (CURLINFO_SIZE_UPLOAD_T)
The unit for these values is bytes/second, not microseconds:
- speed_download (CURLINFO_SPEED_DOWNLOAD_T)
- speed_upload (CURLINFO_SPEED_UPLOAD_T)
Fixes#5131Closes#5152
Reported by the new script 'scripts/copyright.pl'. The script has a
regex whitelist for the files that don't need copyright headers.
Removed three (mostly usesless) README files from docs/
Closes#5141
- send more data to make problems more obvious
- don't start the data with minus, it makes diffs harder to read
- skip the headers in the stdout comparison
- save to a file name to also verify 'filename_effective'
Ref: #5131
Update smbserver.py and negtelnetserver.py to be compatible with
Python 3 while staying backwards-compatible to support Python 2.
Fix string encoding and handling of echoed and transferred data.
Tested with both Python 2.7.17 and Python 3.7.7
Reported-by: Daniel Stenberg
Assisted-by: Kamil Dudka
Reviewed-by: Marcel Raad
Fixes#5104Closes#5110
- Implement new option CURLSSLOPT_REVOKE_BEST_EFFORT and
--ssl-revoke-best-effort to allow a "best effort" revocation check.
A best effort revocation check ignores errors that the revocation check
was unable to take place. The reasoning is described in detail below and
discussed further in the PR.
---
When running e.g. with Fiddler, the schannel backend fails with an
unhelpful error message:
Unknown error (0x80092012) - The revocation function was unable
to check revocation for the certificate.
Sadly, many enterprise users who are stuck behind MITM proxies suffer
the very same problem.
This has been discussed in plenty of issues:
https://github.com/curl/curl/issues/3727,
https://github.com/curl/curl/issues/264, for example.
In the latter, a Microsoft Edge developer even made the case that the
common behavior is to ignore issues when a certificate has no recorded
distribution point for revocation lists, or when the server is offline.
This is also known as "best effort" strategy and addresses the Fiddler
issue.
Unfortunately, this strategy was not chosen as the default for schannel
(and is therefore a backend-specific behavior: OpenSSL seems to happily
ignore the offline servers and missing distribution points).
To maintain backward-compatibility, we therefore add a new flag
(`CURLSSLOPT_REVOKE_BEST_EFFORT`) and a new option
(`--ssl-revoke-best-effort`) to select the new behavior.
Due to the many related issues Git for Windows and GitHub Desktop, the
plan is to make this behavior the default in these software packages.
The test 2070 was added to verify this behavior, adapted from 310.
Based-on-work-by: georgeok <giorgos.n.oikonomou@gmail.com>
Co-authored-by: Markus Olsson <j.markus.olsson@gmail.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Closes https://github.com/curl/curl/pull/4981
Makes curl_easy_getinfo() of "variable" numerical content instead return
the number set in the env variable `CURL_TIME`.
Makes curl_version() of "variable" textual content. This guarantees a
stable version string which can be tested against. Environment variable
`CURL_VERSION` defines the content.
Assisted-by: Mathias Gumz
The test uses SRP to "a server not supporting it" but modern stunnel
versions will silently accept it and remain happy. The test is therefore
faulty.
I haven't figured out how to make stunnel explicitly reject SRP-using
connects.
Reported-by: Marc Hörsken
Fixes#5105Closes#5113
Users of the SMB tests will have to install impacket manually.
Reasoning: our in-tree version of impacket was quite outdated
and only compatible with Python 2 which is already end-of-life.
Upgrading to Python 3 and a compatible impacket version would
require to import additional Python-only and CPython-extension
dependencies. This would have hindered portability enormously.
Closes#5094
When extracting a <section> <part> and there's no </part> before
</section>, this now outputs an error and returns a wrong string to
make users spot the mistake.
Ref: #5070Closes#5071
Even though the existing code can be fixed to run on Python 3, the
tests will fail due to the Unicode transition the protocol is invalid.
Follow up to ee63837Closes#5085
Recent gcc warns when byte count of strncpy() equals the destination
buffer size. Since the destination buffer is previously cleared and
the source string is always shorter, reducing the byte count by one
silents the warning without affecting the result.
Closes#5059
When using maximum code optimization level (-O3), valgrind wrongly
detects uses of uninitialized values in strcmp().
Preset buffers with all zeroes to avoid that.
This test does A LOT of *wakeup() calls and then calls curl_multi_poll()
twice. The first *poll() is then expected to return early and the second
not - as the first is supposed to drain the socketpair pipe.
It turns out however that when given "excessive" amounts of writes to
the pipe, some operating systems (the Solaris based are known) will
return EAGAIN before the pipe is drained, which in our test case causes
the second *poll() call to also abort early.
This change attempts to avoid the OS-specific behaviors in the test by
reducing the amount of wakeup calls from 1234567 to 10.
Reported-by: Andy Fiddaman
Fixes#5037Closes#5058
New test 666 checks this is effective.
As upload buffer size is significant in this kind of tests, shorten it
in similar test 652.
Fixes#4860Closes#4833
Reported-by: RuurdBeerstra on github
Input buffer filling may delay the data sending if data reads are slow.
To overcome this problem, file and callback data reads do not accumulate
in buffer anymore. All other data (memory data and mime framing) are
considered as fast and still concatenated in buffer.
As this may highly impact performance in terms of data overhead, an early
end of part data check is added to spare a read call.
When encoding a part's data, an encoder may require more bytes than made
available by a single read. In this case, the above rule does not apply
and reads are performed until the encoder is able to deliver some data.
Tests 643, 644, 645, 650 and 654 have been adapted to the output data
changes, with test data size reduced to avoid the boredom of long lists of
1-byte chunks in verification data.
New test 667 checks mimepost using single-byte read callback with encoder.
New test 668 checks the end of part data early detection.
Fixes#4826
Reported-by: MrdUkk on github
In case a read callback returns a status (pause, abort, eof,
error) instead of a byte count, drain the bytes read so far but
remember this status for further processing.
Takes care of not losing data when pausing, and properly resume a
paused mime structure when requested.
New tests 670-673 check unpausing cases, with easy or multi
interface and mime or form api.
Fixes#4813
Reported-by: MrdUkk on github
New test 666 checks this is effective.
As upload buffer size is significant in this kind of tests, shorten it
in similar test 652.
Fixes#4860
Reported-by: RuurdBeerstra on github
Input buffer filling may delay the data sending if data reads are slow.
To overcome this problem, file and callback data reads do not accumulate
in buffer anymore. All other data (memory data and mime framing) are
considered as fast and still concatenated in buffer.
As this may highly impact performance in terms of data overhead, an early
end of part data check is added to spare a read call.
When encoding a part's data, an encoder may require more bytes than made
available by a single read. In this case, the above rule does not apply
and reads are performed until the encoder is able to deliver some data.
Tests 643, 644, 645, 650 and 654 have been adapted to the output data
changes, with test data size reduced to avoid the boredom of long lists of
1-byte chunks in verification data.
New test 664 checks mimepost using single-byte read callback with encoder.
New test 665 checks the end of part data early detection.
Fixes#4826
Reported-by: MrdUkk on github
In case a read callback returns a status (pause, abort, eof,
error) instead of a byte count, drain the bytes read so far but
remember this status for further processing.
Takes care of not losing data when pausing, and properly resume a
paused mime structure when requested.
New tests 670-673 check unpausing cases, with easy or multi
interface and mime or form api.
Fixes#4813
Reported-by: MrdUkk on github
Closes#4833
This reduces the HTTP/2 window size to 32 MB since libcurl might have to
buffer up to this amount of data in memory and yet we don't want it set
lower to potentially impact tranfer performance on high speed networks.
Requires nghttp2 commit b3f85e2daa629
(https://github.com/nghttp2/nghttp2/pull/1444) to work properly, to end
up in the next release after 1.40.0.
Fixes#4939Closes#4940
It is still possible to override the executable to run during the test,
using the <tool> tag, but this patch removes the requirement that the
tag must be present for unit tests.
It also removes the possibility of human error when existing test cases
are used as the basis for new tests, as recently witnessed in 81c37124.
Reviewed-by: Daniel Stenberg
Closes#4976
When doing a request with a body + Expect: 100-continue and the server
responds with a 417, the same request will be retried immediately
without the Expect: header.
Added test 357 to verify.
Also added a control instruction to tell the sws test server to not read
the request body if Expect: is present, which the new test 357 uses.
Reported-by: bramus on github
Fixes#4949Closes#4964
Note: The RCPT TO command isn't required to advertise to the server that
it contains UTF-8 characters, instead the server is told that a mail may
contain UTF-8 in any envelope command via the MAIL command.
Support the SMTPUTF8 extension when sending mailbox information in the
MAIL command (FROM and AUTH parameters). Non-ASCII domain names will
be ACE encoded, if IDN is supported, whilst non-ASCII characters in
the local address part are passed to the server.
Reported-by: ygthien on github
Fixes#4828
The dot character between the host and the tld was not being escaped,
which meant it specified a match of 'any' character rather than an
explicit dot separator.
Additionally removed the dot character from the host name as it allowed
the following to be specified as a valid address in our test cases:
<bad@example......com>
Both are typos from 98f7ca7 and 8880f84 :(
I can't remember whether my intention was to allow sub-domains to be
specified in the host or not with these additional dots, but by placing
it outside of the host means it can only be specified once per domain
and by placing a + after the new grouping support for sub-domains is
kept.
Closes#4912
The alt-svc cache survives a call to curl_easy_reset fine, but the file
name to use for saving the cache was cleared. Now the alt-svc cache has
a copy of the file name to survive handle resets.
Added test 1908 to verify.
Reported-by: Craig Andrews
Fixes#4898Closes#4902
RFC 7616 section 3.4 (The Authorization Header Field) states that "For
historical reasons, a sender MUST NOT generate the quoted string syntax
for the following parameters: algorithm, qop, and nc". This removes the
quoting for the algorithm parameter.
Reviewed-by: Steve Holme
Closes#4890
... as this is already done much earlier in the URL parser.
Also add test case 894 that verifies that pop3 with an encodedd CR in
the user name is rejected.
Closes#4887
In the "scheme-less" parsing case, we need to strip off credentials
first before we guess scheme based on the host name!
Assisted-by: Jay Satiro
Fixes#4856Closes#4857
Previously it was stored in a global state which contributed to
curl_global_init's thread unsafety. This boolean is now instead figured
out in curl_multi_init() and stored in the multi handle. Less effective,
but thread safe.
Closes#4851
- Add new error code CURLE_QUIC_CONNECT_ERROR for QUIC connection
errors.
Prior to this change CURLE_FAILED_INIT was used, but that was not
correct.
Closes https://github.com/curl/curl/pull/4754
When using randomized features of runtests (-R and --shallow) it is
useful to have a fixed random seed to make sure for example extra
commits in a branch or a rebase won't change the seed that would make
repeated runs work differently.
As it is also useful to change seed sometimes, the default seed is now
determined based on the current month (and first line curl -V
output). When the month changes, so will the random seed.
The specific seed is also shown in the standard test suite top header
and it can be set explictly with the new --seed=[num] option so that the
exact order of a previous run can be achieved.
Closes#4734
It was removed for output containing ' =' via `s/ =.*//`. With classic
MinGW, this made lines with `free()` end with CRLF, but lines with e.g.
`malloc()` end with only LF. The tests expect LF only.
Closes https://github.com/curl/curl/pull/4788
Previously it would end up with an uninitialized memory buffer that
would lead to a crash or junk getting output.
Added test 1271 to verify.
Reported-by: Brian Carpenter
Closes#4786
Prior to this change the swsbounce check in service_connection could
fail because prevtestno and prevpartno were not set, which would cause
the wrong response data to be sent to some tests and cause them to fail.
Ref: https://github.com/curl/curl/pull/4717#issuecomment-570240785
Prior to this change tests that required NTLM feature did not require
SSL feature.
There are pending changes to cmake builds that will allow enabling NTLM
in non-SSL builds in Windows. In that case the NTLM auth strings created
are different from what is expected by the NTLM tests and they fail:
"The issue with NTLM is that previous non-SSL builds would not enable
NTLM and so the NTLM tests would be skipped."
Assisted-by: marc-groundctl@users.noreply.github.com
Ref: https://github.com/curl/curl/pull/4717#issuecomment-566218729
Closes https://github.com/curl/curl/pull/4768
Even if the initial request line wasn't found. With the fix to 1455, the
test number is now detected correctly.
(Problem found when running tests in random order.)
Closes#4744
On my current Debian Unstable with libidn2 2.2.0, I get an error if
LC_ALL is set to blank. Then curl errors out with:
curl: (3) Failed to convert www.åäö.se to ACE; could not convert string to UTF-8
Closes#4738
Fixup the test to instead not compare the port number. It sometimes
caused problems like this:
"curl: (45) bind failed with errno 98: Address already in use"
Closes#4733
When set, shallow mode limits runtests -t to make no more than NUM fails
per test case. If more are found, it will randomly discard entries until
the number is right. The random seed can also be set.
This is particularly useful when running MANY tests as then most torture
failures will already fail the same functions over and over and make the
total operation painfully tedious.
Closes#4699
This enables the use of Windows Subsystem for Linux (WSL) to run the
testsuite against Windows binaries while using Linux servers.
This commit introduces the following environment variables:
- CURL_TEST_EXE_EXT: set the executable extension for all components
- CURL_TEST_EXE_EXT_TOOL: set it for the curl tool only
- CURL_TEST_EXE_EXT_SSH: set it for the SSH tools only
Later testcurl.pl could be adjusted to make use of those variables.
- CURL_TEST_EXE_EXT_SRV: set it for the test servers only
(This is one of several commits to support use of WSL for the tests.)
Closes https://github.com/curl/curl/pull/3899
Keys created on Windows Subsystem for Linux (WSL) require it for some
reason.
(This is one of several commits to support use of WSL for the tests.)
Ref: https://github.com/curl/curl/pull/3899
Bash in Windows Subsystem for Linux (WSL) requires it for some reason.
(This is one of several commits to support use of WSL for the tests.)
Ref: https://github.com/curl/curl/pull/3899
It could accidentally let the connection get used by more than one
thread, leading to double-free and more.
Reported-by: Christopher Reid
Fixes#4544Closes#4557
