When reusing a connection, make sure the unix domain
socket option matches.

Follow-up to a4d88885. Changing the window size in that commit resulted
in a different HTTP2-Settings upgrade header, causing test 1800 to fail.

- Improve performance by using a huge HTTP/2 window size.

Bug: https://github.com/curl/curl/issues/1102


Reported-by:  <afrind@users.noreply.github.com>
Assisted-by: Tatsuhiro Tsujikawa

and warn on unrecognized lines

... and warn for too long --help lines

- In Curl_http2_switched don't call memcpy when src is NULL.

Curl_http2_switched can be called like:

Curl_http2_switched(conn, NULL, 0);

.. and prior to this change memcpy was then called like:

memcpy(dest, NULL, 0)

.. causing address sanitizer to warn:

http2.c:2057:3: runtime error: null pointer passed as argument 2, which
is declared to never be null

Reviewed-by: Nick Zitzmann

Closes #1105

Fixes test 1139 failures

Follow-up to f82bbe01

Brought in ee4f7660

Added test case 1280 to verify

Reported-by: Dave Reisner

Bug: https://github.com/curl/curl/commit/ee4f76606cfa4ee068bf28edd37c8dae7e8db317#commitcomment-19823146

Exit with an error on the first transfer error instead of continuing to
do the rest of the URLs.

Discussion: https://curl.haxx.se/mail/archive-2016-11/0038.html

Now Curl_rand() is made to fail if it cannot get the necessary random
level.

Changed the proto of Curl_rand() slightly to provide a number of ints at
once.

Moved out from vtls, since it isn't a TLS function and vtls provides
Curl_ssl_random() for this to use.

Discussion: https://curl.haxx.se/mail/lib-2016-11/0119.html

See MANPAGE.md for the description of how this works. Each command line
option is now described in a separate .d file.

Blah, I accidentally wrote size_t instead of time_t for two variables.

Reported-by: Dave Reisner

... as long is still 32bit on modern 64bit windows machines, while
time_t is generally 64bit.

This stops the compiler from potentially making invalid assumptions
about the immutability of sdp and sap across the longjmp boundary.

Previously, the [host] part was just ignored which made libcurl accept
strange URLs misleading users. like "file://etc/passwd" which might've
looked like it refers to "/etc/passwd" but is just "/passwd" since the
"etc" is an ignored host name.

Reported-by: Mike Crowe
Assisted-by: Kamil Dudka

1.1.0+ does more of this by itself so we can avoid extra processing this
way.

and remove RAND_screen from configure since nothing is using that
function

Discussed: https://curl.haxx.se/mail/lib-2016-11/0087.html

to consider ECONNREFUSED as a transient error.

Closes #1064

Now I've managed to negotiate TLS 1.3 with https://enabled.tls13.com/ when
using boringssl.

- Fix GnuTLS code for CURL_SSLVERSION_TLSv1_2 that broke when the
TLS 1.3 support was added in 6ad3add6.

- Homogenize across code for all backends the error message when TLS 1.3
is not available to "<backend>: TLS 1.3 is not yet supported".

- Return an error when a user-specified ssl version is unrecognized.

---

Prior to this change our code for some of the backends used the
'default' label in the switch statement (ie ver unrecognized) for
ssl.version and treated it the same as CURL_SSLVERSION_DEFAULT.

Bug: https://curl.haxx.se/mail/lib-2016-11/0048.html
Reported-by: Kamil Dudka

If SPNEGO fails, cleanup the negotiate handle right away.

Fixes #1115

Signed-off-by: Isaac Boukris <iboukris@gmail.com>
Reported-by: ashman-p