Test command 'time curl http://localhost/80GB -so /dev/null' on a Debian
Linux.

Before (middle performing run out 9):

 real    0m28.078s
 user    0m11.240s
 sys     0m12.876s

After (middle performing run out 9)

 real    0m26.356s (93.9%)
 user    0m5.324s  (47.4%)
 sys     0m8.368s  (65.0%)

Also, doing SFTP over a 200 millsecond latency link is now about 6 times
faster.

Closes #1446

The data->req.uploadbuf struct member served no good purpose, instead we
use ->state.uploadbuffer directly. It makes it clearer in the code which
buffer that's being used.

Removed the 'SingleRequest *' argument from the readwrite_upload() proto
as it can be derived from the Curl_easy struct. Also made the code in
the readwrite_upload() function use the 'k->' shortcut to all references
to struct fields in 'data->req', which previously was made with a mix of
both.

- Change prepends to appends because user's LDFLAGS and CPPFLAGS should
  always come first so they're searched before ours.

Bug: https://github.com/curl/curl/issues/1420
Reported-by: Helmut K. C. Tessarek

Follow-up to 11903732, which fixed the
warning in the HAVE_GETIFADDRS block, but not in the
HAVE_IOCTL_SIOCGIFADDR block.

$< is only allowed in implicit rules in some non-GNU makes (e.g. BSD,
AIX) so avoid use elsewhere by referencing the dependent curl.1 file
directly instead. This is somewhat tricky because the file is supplied
in the packaged tar ball (but not in git) but must still be able to be
rebuilt when its dependencies change. The right thing must happen in
both tar ball and git source trees, as well as in both in-tree and
out-of-tree builds.

This commit fixes compilation failure caused by
cbae73e1.

-Og, introduced in GCC 4.8, optimizes for debugging experience.
-Ofast, introduced in GCC 4.7, builds on -O3 and enables further
optimizations breaking strict standards compliance.
When specified in CFLAGS, these were always overridden by -O0 or -O2.
Fix this by adding them to flags_opt_all.

Ref: https://gcc.gnu.org/onlinedocs/gcc-4.8.0/gcc/Optimize-Options.html
Ref: https://github.com/curl/curl/pull/1404#issuecomment-296401570
Closes https://github.com/curl/curl/pull/1440

The check for if -ldl is needed to build with (a statically built)
openssl was broken. This repairs the check, and adds a check for
-lpthread as well since OpenSSL 1.1.0+ does in fact require -lpthread so
only adding -ldl for a static openssl build is no longer enough.

Reported-by: Jay Satiro
Ref: #1426
Closes #1427

Pointed-it-by: Kevin Ji
URL: https://github.com/curl/curl/commit/cbae73e1dd95946597ea74ccb580c30f78e3fa73#commitcomment-21872622

- Track when the cached encrypted data contains only a partial record
  that can't be decrypted without more data (SEC_E_INCOMPLETE_MESSAGE).

- Change Curl_schannel_data_pending to return false in such a case.

Other SSL libraries have pending data functions that behave similarly.

Ref: https://github.com/curl/curl/pull/1387

Closes https://github.com/curl/curl/pull/1392

`if(nfds || extra_nfds) {` is followed by `malloc(nfds * ...)`.

If `extra_fs` could be non-zero when `nfds` was zero, then we have
`malloc(0)` which is allowed to return `NULL`. But, malloc returning
NULL can be confusing. In this code, the next line would treat the NULL
as an allocation failure.

It turns out, if `nfds` is zero then `extra_nfds` must also be zero.
The final value of `nfds` includes `extra_nfds`.  So the test for
`extra_nfds` is redundant.  It can only confuse the reader.

Closes #1439

With -Og, GCC complains:

easy.c:628:7: error: ‘mcode’ may be used uninitialized in this function [-Werror=maybe-uninitialized]

../lib/strcase.h:35:29: error: ‘tok_buf’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
vauth/digest.c:208:9: note: ‘tok_buf’ was declared here

../lib/strcase.h:35:29: error: ‘tok_buf’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
vauth/digest.c:566:15: note: ‘tok_buf’ was declared here

Fix this by initializing the variables.

This reduces the binary size and fixes a compile warning.

The 'list element' struct now has to be within the data that is being
added to the list. Removes 16.6% (tiny) mallocs from a simple HTTP
transfer. (96 => 80)

Also removed return codes since the llist functions can't fail now.

Test 1300 updated accordingly.

Closes #1435

All the callbacks passed to curl_easy_setopt are defined as function
pointers. The possibility to pass both functions and function pointers
was handled for the callbacks that typecheck-gcc.h defined as
compatible, but not for the public callback types themselves.

This makes all compatible callback types defined in typecheck-gcc.h
function pointers too and checks all functions uniformly with
_curl_callback_compatible, which handles both functions and function
pointers.

A symptom of the problem was a warning in tool_operate.c with
--disable-libcurl-option and without --enable-debug as that file
passes the callback functions to curl_easy_setopt directly.

Fixes https://github.com/curl/curl/issues/1403
Closes https://github.com/curl/curl/pull/1404

In that case, use libcurl's internal MD4 routine. This fixes tests 1013
and 1014 which were failing due to configure assuming NTLM and SMB were
always available whenever mbed TLS was in use (which is now true).

... when failing to get the server certificate.

This reverts commit 5b4cbcf1.

Since it broke out-of-tree builds from tarballs. See discussion in #1432

... since apparently "BSD make" doesn't support it.

Reported-by: Thomas Klausner
Fixes #1432

This fixes 3 warnings issued by MinGW:
1. PR_ImportTCPSocket actually has a paramter of type PROsfd instead of
PRInt32, which is 64 bits on Windows. Fixed this by including the
corresponding header file instead of redeclaring the function, which is
supported even though it is in the private include folder. [1]
2. In 64-bit mode, size_t is 64 bits while CK_ULONG is 32 bits, so an explicit
narrowing cast is needed.
3. Curl_timeleft returns time_t instead of long since commit
21aa32d3.

[1] https://developer.mozilla.org/en-US/docs/Mozilla/Projects/NSPR/Reference/PR_ImportTCPSocket

Closes https://github.com/curl/curl/pull/1393

Move the sessionid flag to ssl_primary_config so that ssl and proxy_ssl
will each have their own sessionid flag.

Regression since HTTPS-Proxy support was added in cb4e2be7. Prior to that
this issue had been fixed in 247d890d, CVE-2016-5419.

Bug: https://github.com/curl/curl/issues/1341


Reported-by:  <lijian996@users.noreply.github.com>

The new incarnation of this bug is called CVE-2017-7468 and is documented
here: https://curl.haxx.se/docs/adv_20170419.html

X.509 certificates carry public keys, not private keys. Fields
corresponding to the private half of the key will always be NULL.

Closes #1425

ERR_error_string with NULL parameter is not thread-safe. The library
writes the string into some static buffer. Two threads doing this at
once may clobber each other and run into problems. Switch to
ERR_error_string_n which avoids this problem and is explicitly
bounds-checked.

Also clean up some remnants of OpenSSL 0.9.5 around here. A number of
comments (fixed buffer size, explaining that ERR_error_string_n was
added in a particular version) date to when ossl_strerror tried to
support pre-ERR_error_string_n OpenSSLs.

Closes #1424

Rather than making assumptions about the values, use a switch-case.

Closes #1424

A few random typos, and minor whitespace cleanups, found in comments
while reading code.

Closes #1423

This makes test 1135 pass with CRLF checkouts.

Ref: https://github.com/curl/curl/pull/1344#issuecomment-289243166
Closes https://github.com/curl/curl/pull/1422

Ignore everything after the version numbers in LIBCURL_VERSION and
LIBCURL_VERSION_NUM to ged rid of the extra CR character.
This makes tests 1022 and 1023 pass on Linux with a CRLF checkout.

Ref: https://github.com/curl/curl/pull/1344#issuecomment-289243166
Closes https://github.com/curl/curl/pull/1422

Bash on Linux errors out on CR characters.
This makes tests 1221 and 1222 pass on Linux with a CRLF checkout.

Ref: https://github.com/curl/curl/pull/1344#issuecomment-289243166
Closes https://github.com/curl/curl/pull/1422

MinGW-w64 complains:
warning: conversion to 'long int' from 'time_t {aka long long int}' may
alter its value [-Wconversion]
Fix this by using the correct type.

Add missing newhandle free call in push_promise().

Closes #1416

Add missing our_ssl_sessionid free call in mbed_connect_step3().

Closes #1417

This adds -Werror-implicit-function-declaration for GCC 2.95+ so that
these errors are visible at the point where they occur instead of only
at link time.
Implicit function declarations are illegal in C99 and C++ anyway, and
the same warning has been turned into an error for ICC in commit
3072c5b8.

Ref: https://gcc.gnu.org/onlinedocs/gcc-2.95.2/gcc_2.html#SEC8
Ref: https://curl.haxx.se/mail/lib-2017-04/0001.html
Closes https://github.com/curl/curl/pull/1419

Ref: https://github.com/curl/curl/issues/1408
Closes https://github.com/curl/curl/pull/1412

Follow-up to aa573c3c

Ref: https://github.com/curl/curl/pull/1406