When linking against bundled libraries, we include their header
directories by using "-isystem". The reason for that is that we
want to handle our vendored library headers specially, most
importantly to ignore warnings generated by including them. By
using "-isystem", though, we screw up the order of searched
include directories by moving those bundled dependencies towards
the end of the lookup order. Like this, chances are high that any
other specified include directory contains a file that collides
with the actual desired include file.

Fix this by not treating the bundled dependencies' include
directories as system includes. This will move them to the front
of the lookup order and thus cause them to override
system-provided headers. While this may cause the compiler to
generate additional warnings when processing bundled headers,
this is a tradeoff we should make regardless to fix builds on
systems hitting this issue.

(cherry picked from commit ee3d71fb)

We assume that if we are on Win32, Amiga OS, Solaris or SunOS,
that the regcomp(3P) function cannot be provided by the system.
Thus we will in these cases always include our own, bundled regex
sources to make a regcomp implementation available. This test is
obviously very fragile, and we have seen it fail on MSYS2/MinGW
systems, which do in fact provide the regcomp symbol. The effect
is that during compilation, we will use the "regex.h" header
provided by MinGW, but use symbols provided by ourselves. This
in fact may cause subtle memory layout issues, as the structure
made available via MinGW doesn't match what our bundled code
expects.

There's one more problem with our regex detection: on the listed
platforms, we will incorrectly include the bundled regex code
even in case where the system provides regcomp_l(3), but it will
never be used for anything.

Fix the issue by improving our regcomp detection code. Instead of
relying on a fragile listing of platforms, we can just use
`CHECK_FUNCTION_EXISTS` instead. This will not in fact avoid the
header-ordering problem. But we can assume that as soon as a
system-provided "regex.h" header is provided, that
`CHECK_FUNCTION_EXISTS` will now correctly find the desired
symbol and thus not include our bundled regex code.

(cherry picked from commit 13cb9f7a)

git_array_alloc can return NULL if no memory is available, causing
a segmentation fault in memset. This adds GIT_ERROR_CHECK_ALLOC
similar to how other parts of the code base deal with the return
value of git_array_alloc.

Stop traversing the filesystem at the Windows directory root.  We were
calculating the filesystem root for the given directory to create, and
walking up the filesystem hierarchy.  We intended to stop when the
traversal path length is equal to the root path length (ie, stopping at
the root, since no path may be shorter than the root path).

However, on Windows, the root path may be specified in two different
ways, as either `Z:` or `Z:\`, where `Z:` is the current drive letter.
`git_path_dirname_r` returns the path _without_ a trailing slash, even
for the Windows root.  As a result, during traversal, we need to test
that the traversal path is _less than or equal to_ the root path length
to determine if we've hit the root to ensure that we stop when our
traversal path is `Z:` and our calculated root path was `Z:\`.

The function `git_ignore_path_is_ignored` is there to test the
ignore status of paths that need not necessarily exist inside of
a repository. This has the implication that for a given path, we
cannot always decide whether it references a directory or a file,
and we need to distinguish those cases because ignore rules may
treat those differently. E.g. given the following gitignore file:

    *
    !/**/

we'd only want to unignore directories, while keeping files
ignored. But still, calling `git_ignore_path_is_ignored("dir/")`
will say that this directory is ignored because it treats "dir/"
as a file path.

As said, the `is_ignored` function cannot always decide whether
the given path is a file or directory, and thus it may produce
wrong results in some cases. While this is unfixable in the
general case, we can do better when we are being passed a path
name with a trailing path separator (e.g. "dir/") and always
treat them as directories.

Add a test that verifies that we are able to parse patches which add a
new file that has spaces in its path.

The diff header format is a strange beast in that it is inherently
unparseable in an unambiguous way. While parsing

    a/file.txt b/file.txt

is obvious and trivially doable, parsing a diff header of

    a/file b/file ab.txt b/file b/file ab.txt

is not (but in fact valid and created by git.git).

Due to that, we have relaxed our diff header parser in commit 80226b5f
(patch_parse: allow parsing ambiguous patch headers, 2017-09-22), so
that we started to bail out when seeing diff headers with spaces in
their file names. Instead, we try to use the "---" and "+++" lines,
which are unambiguous.

In some cases, though, we neither have a useable file name from the
header nor from the "---" or "+++" lines. This is the case when we have
a deletion or addition of a file with spaces: the header is unparseable
and the other lines will simply show "/dev/null". This trips our parsing
logic when we try to extract the prefix (the "a/" part) that is being
used in the path line, where we unconditionally try to dereference a
NULL pointer in such a scenario.

We can fix this by simply not trying to parse the prefix in cases where
we have no useable path name. That'd leave the parsed patch without
either `old_prefix` or `new_prefix` populated. But in fact such cases
are already handled by users of the patch object, which simply opt to
use the default prefixes in that case.

Matching on the prefix of a negated pattern was triggering false
negatives on siblings of that pattern. e.g.

Given the .gitignore:
dir/*
!dir/sub1/sub2/**

The path `dir/a.text` would not be ignored.

When a directory's contents are ignored, and then a glob negation is made to a nested subdir, other subdirectories are now unignored

This change fixes a mostly theoretical memory leak in got_odb_new()
that can only manifest if git_cache_init() fails due to running out of
memory or not being able to acquire its lock.

This change frees a copy of a cached object in odb_otype_fast().

Release v0.28.1

Deprecation: export the deprecated functions properly

Enable hard deprecation in our builds to ensure that we do not call
deprecated functions internally.

Add a CMake option to enable hard deprecation; the resultant library
will _not_ include any deprecated functions.  This may be useful for
internal CI builds that create libraries that are not shared with
end-users to ensure that we do not use deprecated bits internally.

Although the error functions were deprecated, we did not properly mark
them as deprecated.  We need to include the `deprecated.h` file in order
to ensure that the functions get their export attributes.

Similarly, do not define `GIT_DEPRECATE_HARD` within the library, or
those functions will also not get their export attributes.  Define that
only on the tests and examples.

ci: skip ssh tests on macOS nightly

Like 811c1c0f, disable the SSH tests on
macOS until we can resolve the newly introduced infrastructure issues.

CI build fixups

SSH tests on macOS have begun failing for an unknown reason after an
infrastructure upgrade to macOS 10.13.6.  Disable those tests
temporarily, until we can resolve it.

Subtle changes in the host OS can have impacts in the CI system that
may be hard to debug.  We previously showed the results of `uname` which
can be difficult to interpret.  Provide more information where
available.

The URL was incorrect for the nightly badge image; it was erroneously
showing the master branch continuous integration build badge.

Include a build badge for `maint/v0.28` builds.

v0.28 rc1

Docs

The mbedtls library uses a callback mechanism to allow downstream users
to plug in their own receive and send functions. We implement `bio_read`
and `bio_write` functions, which simply wrap the `git_stream_read` and
`git_stream_write` functions, respectively.

The problem arises due to the return value of the callback functions:
mbedtls expects us to return an `int` containing the actual number of
bytes that were read or written. But this is in fact completely
misdesigned, as callers are allowed to pass in a buffer with length
`SIZE_MAX`. We thus may be unable to represent the number of bytes
written via the return value.

Fix this by only ever reading or writing at most `INT_MAX` bytes.

The mbedtls stream implementation makes use of some global variables
which are not marked as `static`, even though they're only used in this
compilation unit. Fix this and remove a duplicate declaration.

Our `openssl_write` function calls `SSL_write` by passing in both `data`
and `len` arguments directly. Thing is, our `len` parameter is of type
`size_t` and theirs is of type `int`. We thus need to clamp our length
to be at most `INT_MAX`.

Now that the function `git_stream__write_full` exists and callers of
`git_stream_write` have been adjusted, we can lift logic for short
writes out of the stream implementations. Instead, this is now handled
either by `git_stream__write_full` or by callers of `git_stream_write`
directly.

Similar to the write(3) function, implementations of `git_stream_write`
do not guarantee that all bytes are written. Instead, they return the
number of bytes that actually have been written, which may be smaller
than the total number of bytes. Furthermore, due to an interface design
issue, we cannot ever write more than `SSIZE_MAX` bytes at once, as
otherwise we cannot represent the number of bytes written to the caller.

Unfortunately, no caller of `git_stream_write` ever checks the return
value, except to verify that no error occurred. Due to this, they are
susceptible to the case where only partial data has been written.

Fix this by introducing a new function `git_stream__write_full`. In
contrast to `git_stream_write`, it will always return either success or
failure, without returning the number of bytes written. Thus, it is able
to write all `SIZE_MAX` bytes and loop around `git_stream_write` until
all data has been written. Adjust all callers except the BIO callbacks
in our mbedtls and OpenSSL streams, which already do the right thing and
require the amount of bytes written.

The callback functions that implement the `git_stream` structure are
only used inside of their respective implementation files, but they are
not marked as `static`. Fix this.