Merge branch 'main' into http-use-eauth

6 files changed, 916 insertions, 32 deletions

diff --git a/docs/changelog.md b/docs/changelog.md
index af80d48cf..1bbbe607d 100644
--- a/docs/changelog.md
+++ b/docs/changelog.md
@@ -1,5 +1,448 @@
-v0.28 + 1
----------
+vNext
+-----
+
+### Changes or improvements
+
+* Branch and tag name validation functions have been introduced:
+  `git_branch_name_is_valid` will check if a branch name is valid,
+  and `git_tag_name_is_valid` will check if a tag name is valid.
+
+* Some remote and reference validity checking functions have been
+  introduced with error reporting semantics.  `git_remote_name_is_valid`
+  replaces `git_remote_is_valid_name`.  `git_reference_name_is_valid`
+  replaces `git_reference_is_valid_name`.  Tthe former functions are
+  deprecated.
+
+### Breaking CMake configuration changes
+
+The `MVSC_CRTDBG` configuration option is now `WIN32_LEAKCHECK`.
+
+v1.1
+----
+
+This is release v1.1, "Fernweh".
+
+### Changes or improvements
+
+* Our bundled PCRE dependency has been updated to 8.44.
+
+* The `refs/remotes/origin/HEAD` file will be created at clone time to
+  point to the origin's default branch.
+
+* libgit2 now uses the `__atomic_` intrinsics instead of `__sync_`
+  intrinsics on supported gcc and clang versions.
+
+* The `init.defaultBranch` setting is now respected and `master` is
+  no longer the hardcoded as the default branch name.
+
+* Patch files that do not contain an `index` line can now be parsed.
+
+* Configuration files with multi-line values can now contain quotes
+  split across multiple lines.
+
+* Windows clients now attempt to use TLS1.3 when available.
+
+* Servers that request an upgrade to a newer HTTP version are
+  silently ignored instead of erroneously failing.
+
+* Users can pass `NULL` to the options argument to
+  `git_describe_commit`.
+
+* Clones and fetches of very large packfiles now succeeds on 32-bit
+  platforms.
+
+* Custom reference database backends can now handle the repository's
+  `HEAD` correctly.
+
+* Repositories with a large number of packfiles no longer exhaust the
+  number of file descriptors.
+
+* The test framework now supports TAP output when the `-t` flag is
+  specified.
+
+* The test framework can now specify an exact match to a test
+  function using a trailing `$`.
+
+* All checkout types support `GIT_CHECKOUT_DISABLE_PATHSPEC_MATCH`.
+
+* `git_blame` now can ignore whitespace changes using the option
+  `GIT_BLAME_IGNORE_WHITESPACE`.
+
+* Several new examples have been created, including an examples for
+  commit, add and push.
+
+* Mode changes during rename are now supported in patch application.
+
+* `git_checkout_head` now correctly removes untracked files in a
+  subdirectory when the `FORCE | REMOVE_UNTRACKED` options are specified.
+
+v1.0.1
+------
+
+This is a bugfix release with the following changes:
+
+- Calculating information about renamed files during merges is more
+  efficient because dissimilarity about files is now being cached and
+  no longer needs to be recomputed.
+  
+- The `git_worktree_prune_init_options` has been correctly restored for
+  backward compatibility.  In v1.0 it was incorrectly deprecated with a
+  typo.
+
+- The optional ntlmclient dependency now supports NetBSD.
+
+- A bug where attempting to stash on a bare repository may have failed
+  has been fixed.
+
+- Configuration files that are unreadable due to permissions are now
+  silently ignored, and treated as if they do not exist.  This matches
+  git's behavior; previously this case would have been an error.
+
+- v4 index files are now correctly written; previously we would read
+  them correctly but would not write the prefix-compression accurately,
+  causing corruption.
+
+- A bug where the smart HTTP transport could not read large data packets
+  has been fixed.  Previously, fetching from servers like Gerrit, that
+  sent large data packets, would error.
+
+v1.0
+----
+
+This is release v1.0 "Luftschloss", which is the first stabe release of
+libgit2. The API will stay compatible across all releases of the same major
+version. This release includes bugfixes only and supersedes v0.99, which will
+stop being maintained. Both v0.27 and v0.28 stay supported in accordance with
+our release policy.
+
+### Changes or improvements
+
+- CMake was converted to make use of the GNUInstallDirs module for both our
+  pkgconfig and install targets in favor of our custom build options
+  `BIN_INSTALL_DIR`, `LIB_INSTALL_DIR` and `INCLUDE_INSTALL_DIR`. Instead, you
+  can now use CMakes standard variables `CMAKE_INSTALL_BINDIR`,
+  `CMAKE_INSTALL_LIBDIR` and `CMAKE_INSTALL_INCLUDEDIR`.
+
+- Some CMake build options accepted either a specific value or a boolean value
+  to disable the option altogether or use automatic detection. We only accepted
+  "ON" or "OFF", but none of the other values CMake recognizes as boolean. This
+  was aligned with CMake's understanding of booleans.
+
+- The installed pkgconfig file contained incorrect values for both `libdir` and
+  `includedir` variables.
+
+- If using pcre2 for regular expressions, then we incorrectly added "pcre2"
+  instead of "pcre2-8" to our pkgconfig dependencies, which was corrected.
+
+- Fixed building the bundled ntlmclient dependency on FreeBSD, OpenBSD and
+  SunOS.
+
+- When writing symlinks on Windows, we incorrectly handled relative symlink
+  targets, which was corrected.
+
+- When using the HTTP protocol via macOS' SecureTransport implementation, reads
+  could stall at the end of the session and only continue after a timeout of 60
+  seconds was reached.
+
+- The filesystem-based reference callback didn't corectly initialize the backend
+  version.
+
+- A segmentation fault was fixed when calling `git_blame_buffer()` for files
+  that were modified and added to the index.
+
+- A backwards-incompatible change was introduced when we moved some structures
+  from "git2/credentials.h" into "git2/sys/credentials.h". This was fixed in the
+  case where you do not use hard deprecation.
+
+- Improved error handling in various places.
+
+
+v0.99
+-----
+
+This is v0.99 "Torschlusspanik".  This will be the last minor release
+before libgit2 v1.0.  We expect to only respond to bugs in this release,
+to stabilize it for next major release.
+
+It contains significant refactorings, but is expected to be API-compatible
+with v0.28.0.
+
+### Changes or improvements
+
+* When fetching from an anonymous remote using a URL with authentication
+  information provided in the URL (eg `https://foo:bar@example.com/repo`),
+  we would erroneously include the literal URL in the FETCH_HEAD file.
+  We now remove that to match git's behavior.
+
+* Some credential structures, enums and values have been renamed:
+  `git_cred` is now `git_credential`.  `git_credtype_t` is now
+  `git_credential_t`.  Functions and types beginning with
+  `git_cred_` now begin with `git_credential`, and constants beginning
+  with `GIT_CREDTYPE` now begin with `GIT_CREDENTIAL`.  The former names
+  are deprecated.
+
+* Several function signatures have been changed to return an `int` to
+  indicate error conditions.  We encourage you to check them for errors
+  in the standard way. 
+
+  * `git_attr_cache_flush`
+  * `git_error_set_str`
+  * `git_index_name_clear`
+  * `git_index_reuc_clear`
+  * `git_libgit2_version`
+  * `git_mempack_reset`
+  * `git_oid_cpy`
+  * `git_oid_fmt`
+  * `git_oid_fromraw`
+  * `git_oid_nfmt`
+  * `git_oid_pathfmt`
+  * `git_remote_stop`
+  * `git_remote_disconnect`
+  * `git_repository__cleanup`
+  * `git_repository_set_config`
+  * `git_repository_set_index`
+  * `git_repository_set_odb`
+  * `git_repository_set_refdb`
+  * `git_revwalk_reset`
+  * `git_revwalk_simplify_first_parent`
+  * `git_revwalk_sorting`
+  * `git_treebuilder_clear`
+  * `git_treebuilder_filter`
+
+* The NTLM and Negotiate authentication mechanisms are now supported when
+  talking to git implementations hosted on Apache or nginx servers.
+
+* The `HEAD` symbolic reference can no longer be deleted.
+
+* `git_merge_driver_source_repo` no longer returns a `const git_repository *`,
+  it now returns a non-`const` `git_repository *`.
+
+* Relative symbolic links are now supported on Windows when `core.symlinks`
+  is enabled.
+
+* Servers that provide query parameters with a redirect are now supported.
+
+* `git_submodule_sync` will now resolve relative URLs.
+
+* When creating git endpoint URLs, double-slashes are no longer used when
+  the given git URL has a trailing slash.
+
+* On Windows, a `DllMain` function is no longer included and thread-local
+  storage has moved to fiber-local storage in order to prevent race
+  conditions during shutdown.
+
+* The tracing mechanism (`GIT_TRACE`) is now enabled by default and does
+  not need to be explicitly enabled in CMake.
+
+* The size of Git objects is now represented by `git_object_size_t`
+  instead of `off_t`.
+
+* Binary patches without data can now be parsed.
+
+* A configuration snapshot can now be created from another configuration
+  snapshot, not just a "true" configuration object.
+
+* The `git_commit_with_signature` API will now ensure that referenced
+  objects exist in the object database.
+
+* Stash messages containing newlines will now be replaced with spaces;
+  they will no longer be (erroneously) written to the repository.
+
+* `git_commit_create_with_signature` now verifies the commit information
+  to ensure that it points to a valid tree and valid parents.
+
+* `git_apply` has an option `GIT_APPLY_CHECK` that will only do a dry-run.
+  The index and working directory will remain unmodified, and application
+  will report if it would have worked.
+
+* Patches produced by Mercurial (those that lack some git extended headers)
+  can now be parsed and applied.
+
+* Reference locks are obeyed correctly on POSIX platforms, instead of
+  being removed.
+
+* Patches with empty new files can now be read and applied.
+
+* `git_apply_to_tree` can now correctly apply patches that add new files.
+
+* The program data configuration on Windows (`C:\ProgramData\Git\config`)
+  must be owned by an administrator, a system account or the current user
+  to be read.
+
+* `git_blob_filtered_content` is now deprecated in favor of `git_blob_filter`.
+
+* Configuration files can now be included conditionally using the
+  `onbranch` conditional.
+
+* Checkout can now properly create and remove symbolic links to directories
+  on Windows.
+
+* Stash no longer recomputes trees when committing a worktree, for
+  improved performance.
+
+* Repository templates can now include a `HEAD` file to default the
+  initial default branch.
+
+* Some configuration structures, enums and values have been renamed:
+  `git_cvar_map` is now `git_configmap`, `git_cvar_t` is now
+  `git_configmap_t`, `GIT_CVAR_FALSE` is now `GIT_CONFIGMAP_FALSE`,
+  `GIT_CVAR_TRUE` is now `GIT_CONFIGMAP_TRUE`, `GIT_CVAR_INT32` is now
+  `GIT_CONFIGMAP_INT32`, and `GIT_CVAR_STRING` is now `GIT_CONFIGMAP_STRING`.
+  The former names are deprecated.
+
+* Repositories can now be created at the root of a Windows drive.
+
+* Configuration lookups are now more efficiently cached.
+
+* `git_commit_create_with_signature` now supports a `NULL` signature,
+  which will create a commit without adding a signature.
+
+* When a repository lacks an `info` "common directory", we will no
+  longer erroneously return `GIT_ENOTFOUND` for all attribute lookups.
+
+* Several attribute macros have been renamed: `GIT_ATTR_TRUE` is now
+  `GIT_ATTR_IS_TRUE`, `GIT_ATTR_FALSE` is now `GIT_ATTR_IS_FALSE`,
+  `GIT_ATTR_UNSPECIFIED` is now `GIT_ATTR_IS_UNSPECIFIED`.  The 
+  attribute enum `git_attr_t` is now `git_attr_value_t` and its
+  values have been renamed: `GIT_ATTR_UNSPECIFIED_T` is now
+  `GIT_ATTR_VALUE_UNSPECIFIED`, `GIT_ATTR_TRUE_T` is now
+  `GIT_ATTR_VALUE_TRUE`, `GIT_ATTR_FALSE_T` is now `GIT_ATTR_VALUE_FALSE`,
+  and `GIT_ATTR_VALUE_T` is now `GIT_ATTR_VALUE_STRING`.  The
+  former names are deprecated.
+
+* `git_object__size` is now `git_object_size`.  The former name is
+  deprecated.
+
+* `git_tag_create_frombuffer` is now `git_tag_create_from_buffer`.  The
+  former name is deprecated.
+
+* Several blob creation functions have been renamed:
+  `git_blob_create_frombuffer` is now named `git_blob_create_from_buffer`,
+  `git_blob_create_fromdisk` is now named `git_blob_create_from_disk`,
+  `git_blob_create_fromworkdir` is now named `git_blob_create_from_workdir`,
+  `git_blob_create_fromstream` is now named `git_blob_create_from_stream`,
+  and `git_blob_create_fromstream_commit` is now named
+  `git_blob_create_from_stream_commit`.  The former names are deprecated.
+
+* The function `git_oid_iszero` is now named `git_oid_is_zero`.  The
+  former name is deprecated.
+
+* Pattern matching is now done using `wildmatch` instead of `fnmatch`
+  for compatibility with git.
+
+* The option initialization functions suffixed by `init_options` are now
+  suffixed with `options_init`.  (For example, `git_checkout_init_options`
+  is now `git_checkout_options_init`.)  The former names are deprecated.
+
+* NTLM2 authentication is now supported on non-Windows platforms.
+
+* The `git_cred_sign_callback` callback is now named `git_cred_sign_cb`.
+  The `git_cred_ssh_interactive_callback` callback is now named
+  `git_cred_ssh_interactive_cb`.
+
+* Ignore files now:
+
+  * honor escaped trailing whitespace.
+  * do not incorrectly negate sibling paths of a negated pattern.
+  * honor rules that stop ignoring files after a wildcard
+
+* Attribute files now:
+
+  * honor leading and trailing whitespace.
+  * treat paths beginning with `\` as absolute only on Windows.
+  * properly handle escaped characters.
+  * stop reading macros defined in subdirectories
+
+* The C locale is now correctly used when parsing regular expressions.
+
+* The system PCRE2 or PCRE regular expression libraries are now used
+  when `regcomp_l` is not available on the system.  If none of these
+  are available on the system, an included version of PCRE is used.
+
+* Wildcards in reference specifications are now supported beyond simply
+  a bare wildcard (`*`) for compatibility with git.
+
+* When `git_ignore_path_is_ignored` is provided a path with a trailing
+  slash (eg, `dir/`), it will now treat it as a directory for the
+  purposes of ignore matching.
+
+* Patches that add or remove a file with a space in the path can now
+  be correctly parsed.
+
+* The `git_remote_completion_type` type is now `git_remote_completion_t`.
+  The former name is deprecated.
+
+* The `git_odb_backend_malloc` is now `git_odb_backend_data_alloc`.  The
+  former name is deprecated.
+
+* The `git_transfer_progress_cb` callback is now `git_indexer_progress_cb`
+  and the `git_transfer_progress` structure is now `git_indexer_progress`.
+  The former names are deprecated.
+
+* The example projects are now contained in a single `lg2` executable
+  for ease of use.
+
+* libgit2 now correctly handles more URLs, such as
+  `http://example.com:/repo.git` (colon but no port),
+  `http://example.com` (no path),
+  and `http://example.com:8080/` (path is /, nonstandard port).
+
+* A carefully constructed commit object with a very large number
+  of parents may lead to potential out-of-bounds writes or
+  potential denial of service.
+
+* The ProgramData configuration file is always read for compatibility
+  with Git for Windows and Portable Git installations.  The ProgramData
+  location is not necessarily writable only by administrators, so we
+  now ensure that the configuration file is owned by the administrator
+  or the current user.
+
+### API additions
+
+* The SSH host key now supports SHA-256 when `GIT_CERT_SSH_SHA256` is set.
+
+* The diff format option `GIT_DIFF_FORMAT_PATCH_ID` can now be used to
+  emit an output like `git patch-id`.
+
+* The `git_apply_options_init` function will initialize a
+  `git_apply_options` structure.
+
+* The remote callbacks structure adds a `git_url_resolve_cb` callback
+  that is invoked when connecting to a server, so that applications
+  may edit or replace the URL before connection.
+
+* The information about the original `HEAD` in a rebase operation is
+  available with `git_rebase_orig_head_name`.  Its ID is available with
+  `git_rebase_orig_head_id`.  The `onto` reference name is available with
+  `git_rebase_onto_name` and its ID is available with `git_rebase_onto_id`.
+
+* ODB backends can now free backend data when an error occurs during its
+  backend data creation using `git_odb_backend_data_free`.
+
+* Options may be specified to `git_repository_foreach_head` to control
+  its behavior: `GIT_REPOSITORY_FOREACH_HEAD_SKIP_REPO` will not skip
+  the main repository's HEAD reference, while
+  `GIT_REPOSITORY_FOREACH_HEAD_SKIP_WORKTREES` will now skip the
+  worktree HEAD references.
+
+* The `GIT_OPT_DISABLE_PACK_KEEP_FILE_CHECKS` option can be specified to
+  `git_libgit2_opts()` to avoid looking for `.keep` files that correspond
+  to packfiles.  This setting can improve performance when packfiles are
+  stored on high-latency filesystems like network filesystems.
+
+* Blobs can now be filtered with `git_blob_filter`, which allows for
+  options to be set with `git_blob_filter_options`, including
+  `GIT_FILTER_NO_SYSTEM_ATTRIBUTES` to disable filtering with system-level
+  attributes in `/etc/gitattributes` and `GIT_ATTR_CHECK_INCLUDE_HEAD` to
+  enable filtering with `.gitattributes` files in the HEAD revision.
+
+### API removals
+
+* The unused `git_headlist_cb` function declaration was removed.
+
+* The unused `git_time_monotonic` API is removed.
+
+* The erroneously exported `inttypes.h` header was removed.
 
 # Security Fixes
 
@@ -85,22 +528,62 @@ v0.28 + 1
   "CollisionDetection". If you were using `SHA1_BACKEND` previously, you'll
   need to check the value you've used, or switch to the autodetection.
 
-### Changes or improvements
-
-* libgit2 now correctly handles more URLs, such as
-  `http://example.com:/repo.git` (colon but no port),
-  `http://example.com` (no path),
-  and `http://example.com:8080/` (path is /, nonstandard port).
-
-* A carefully constructed commit object with a very large number
-  of parents may lead to potential out-of-bounds writes or
-  potential denial of service.
-
-* The ProgramData configuration file is always read for compatibility
-  with Git for Windows and Portable Git installations.  The ProgramData
-  location is not necessarily writable only by administrators, so we
-  now ensure that the configuration file is owned by the administrator
-  or the current user.
+### Authors
+
+The following individuals provided changes that were included in this
+release:
+
+* Aaron Patterson
+* Alberto Fanjul
+* Anders Borum
+* Augie Fackler
+* Augustin Fabre
+* Ayush Shridhar
+* brian m. carlson
+* buddyspike
+* Carlos Martín Nieto
+* cheese1
+* Dan Skorupski
+* Daniel Cohen Gindi
+* Dave Lee
+* David Brooks
+* David Turner
+* Denis Laxalde
+* Dhruva Krishnamurthy
+* Dominik Ritter
+* Drew DeVault
+* Edward Thomson
+* Eric Huss
+* Erik Aigner
+* Etienne Samson
+* Gregory Herrero
+* Heiko Voigt
+* Ian Hattendorf
+* Jacques Germishuys
+* Janardhan Pulivarthi
+* Jason Haslam
+* Johannes Schindelin
+* Jordan Wallet
+* Josh Bleecher Snyder
+* kas
+* kdj0c
+* Laurence McGlashan
+* lhchavez
+* Lukas Berk
+* Max Kostyukevich
+* Patrick Steinhardt
+* pcpthm
+* Remy Suen
+* Robert Coup
+* romkatv
+* Scott Furry
+* Sebastian Henke
+* Stefan Widgren
+* Steve King Jr
+* Sven Strickroth
+* Tobias Nießen
+* Tyler Ang-Wanek
+* Tyler Wanek
 
 v0.28
 -----
diff --git a/docs/coding-style.md b/docs/coding-style.md
new file mode 100644
index 000000000..b8b94d69c
--- /dev/null
+++ b/docs/coding-style.md
@@ -0,0 +1,364 @@
+# libgit2 Coding Style
+
+This documentation describes the preferred coding style for the libgit2 project.
+While not all parts of our code base conform to this coding style, the outlined
+rules are what we aim for.
+
+Note that in no case do we accept changes that convert huge parts of the code
+base to use our coding style. Instead, it is encouraged to modernize small parts
+of code you're going to modify anyway for a given change you want to introduce.
+A good rule to follow is the Boy Scout Rule: "Leave the campground cleaner than
+you found it."
+
+## C Coding Style
+
+The following sections define the coding style for all code files and headers.
+
+### Indentation and Alignment
+
+Code is indented by tabs, where a tab is 8 spaces. Each opening scope increases
+the indentation level.
+
+```c
+int foobar(int void)
+{
+	if (condition)
+		doit();
+	/* Body */
+}
+```
+
+Switch statements have their `case`s aligned with the `switch` keyword. Case
+bodies are indented by an additional level. Case bodies should not open their
+own scope to declare variables.
+
+```c
+switch (c) {
+case 'a':
+case 'b':
+	return 0;
+default:
+	return -1;
+}
+```
+
+Multi-line conditions should be aligned with the opening brace of the current
+statement:
+
+```c
+if (one_very_long_condition(c) &&
+    another_very_long_condition(c))
+	doit();
+```
+
+### Spaces
+
+There must be no space between the function and its arguments, arguments must be
+separated by a space:
+
+```c
+int doit(int first_arg, int second_arg);
+doit(1, 2);
+```
+
+For any binary or ternary operators, the arguments and separator must be
+separated by a space:
+
+```c
+1 + 2;
+x ? x : NULL;
+```
+
+Unary operators do not have a space between them and the argument they refer to:
+
+```c
+*c
+&c
+```
+
+The `sizeof` operator always must not have a space and must use braces around
+the type:
+
+```
+sizeof(int)
+```
+
+There must be a space after the keywords `if`, `switch`, `case`, `do` and
+`while`.
+
+### Braces
+
+Functions must have their opening brace on the following line:
+
+```c
+void foobar(void)
+{
+	doit();
+}
+```
+
+For conditions, braces should be placed on the same line as the condition:
+
+```c
+if (condition(c)) {
+	doit();
+	dothat();
+}
+
+while (true) {
+	doit();
+}
+```
+
+In case a condition's body has a single line, only, it's allowed to omit braces,
+except if any of its `else if` or `else` branches has more than one line:
+
+```c
+if (condition(c))
+	doit();
+
+if (condition(c))
+	doit();
+else if (other_condition(c))
+	doit();
+
+/* This example must use braces as the `else if` requires them. */
+if (condition(c)) {
+	doit();
+} else if (other_condition(c)) {
+	doit();
+	dothat();
+} else {
+	abort();
+}
+```
+
+### Comments
+
+Comments must use C-style `/* */` comments. C++-style `// `comments are not
+allowed in our codebase. This is a strict requirement as libgit2 tries to be
+compliant with the ISO C90 standard, which only allows C-style comments.
+
+Single-line comments may have their opening and closing tag on the same line:
+
+```c
+/* This is a short comment. */
+```
+
+For multi-line comments, the opening and closing tag should be empty:
+
+```c
+/*
+ * This is a rather long and potentially really unwiedly but informative
+ * multiline comment that helps quite a lot.
+ */
+```
+
+Public functions must have documentation that explain their usage, internal
+functions should have a comment. We use Docurium to generate documentation
+derived from these comments, which uses syntax similar to Doxygen. The first
+line should be a short summary of what the function does. More in-depth
+explanation should be separated from that first line by an empty line.
+Parameters and return values should be documented via `@return` and `@param`
+tags:
+
+```c
+/*
+ * Froznicate the string.
+ *
+ * Froznicate the string by foobaring its internal structure into a more obvious
+ * translation. Note that the returned string is a newly allocated string that
+ * shall be `free`d by the caller.
+ *
+ * @param s String to froznicate
+ * @return A newly allocated string or `NULL` in case an error occurred.
+ */
+char *froznicate(const char *s);
+```
+
+### Variables
+
+Variables must be declared at the beginning of their scope. This is a strict
+requirement as libgit2 tries to be compliant with the ISO C90 standard, which
+forbids mixed declarations and code:
+
+```c
+void foobar(void)
+{
+	char *c = NULL;
+	int a, b;
+
+	a = 0;
+	b = 1;
+
+	return c;
+}
+```
+
+### Naming
+
+Variables must have all-lowercase names. In case a variable name has multiple
+words, words should be separated by an underscore `_` character. While
+recommended to use descriptive naming, common variable names like `i` for
+indices are allowed.
+
+All public functions must have a `git` prefix as well as a prefix indicating
+their respective subsystem. E.g. a function that opens a repository should be
+called `git_repository_open()`. Functions that are not public but declared in
+an internal header file for use by other subsystems should follow the same
+naming pattern. File-local static functions must not have a `git` prefix, but
+should have a prefix indicating their respective subsystem.
+
+All structures declared in the libgit2 project must have a `typedef`, we do not
+use `struct type` variables. Type names follow the same schema as functions.
+
+### Error Handling
+
+The libgit2 project mostly uses error codes to indicate errors. Error codes are
+always of type `int`, where `0` indicates success and a negative error code
+indicates an error case. In some cases, positive error codes may be used to
+indicate special cases. Returned values that are not an error code should be
+returned via an out parameter. Out parameters must always come first in the list
+of arguments.
+
+```c
+int doit(const char **out, int arg)
+{
+	if (!arg)
+		return -1;
+	*out = "Got an argument";
+	return 0;
+}
+```
+
+To avoid repetitive and fragile error handling in case a function has resources
+that need to be free'd, we use `goto out`s:
+
+```c
+int doit(char **out, int arg)
+{
+	int error = 0;
+	char *c;
+
+	c = malloc(strlen("Got an argument") + 1);
+	if (!c) {
+		error = -1;
+		goto out;
+	}
+
+	if (!arg) {
+		error = -1;
+		goto out;
+	}
+
+	strcpy(c, "Got an argument")
+	*out = c;
+
+out:
+	if (error)
+		free(c);
+	return error;
+}
+```
+
+When calling functions that return an error code, you should assign the error
+code to an `error` variable and, in case an error case is indicated and no
+custom error handling is required, return that error code:
+
+```c
+int foobar(void)
+{
+	int error;
+
+	if ((error = doit()) < 0)
+		return error;
+
+	return 0;
+}
+```
+
+When doing multiple function calls where all of the functions return an error
+code, it's common practice to chain these calls together:
+
+```c
+int doit(void)
+{
+	int error;
+
+	if ((error = dothis()) < 0 ||
+	    (error = dothat()) < 0)
+		return error;
+
+	return 0;
+}
+```
+
+## CMake Coding Style
+
+The following section defines the coding style for our CMake build system.
+
+### Indentation
+
+Code is indented by tabs, where a tab is 8 spaces. Each opening scope increases
+the indentation level.
+
+```cmake
+if(CONDITION)
+	doit()
+endif()
+```
+
+### Spaces
+
+There must be no space between keywords and their opening brace. While this is
+the same as in our C codebase for function calls, this also applies to
+conditional keywords. This is done to avoid the awkward-looking `else ()`
+statement.
+
+```cmake
+if(CONDITION)
+	doit()
+else()
+	dothat()
+endif()
+```
+
+### Case
+
+While CMake is completely case-insensitive when it comes to function calls, we
+want to agree on a common coding style for this. To reduce the danger of
+repetitive strain injuries, all function calls should be lower-case (NB: this is
+not currently the case yet, but introduced as a new coding style by this
+document).
+
+Variables are written all-uppercase. In contrast to functions, variables are
+case-sensitive in CMake. As CMake itself uses upper-case variables in all
+places, we should follow suit and do the same.
+
+Control flow keywords must be all lowercase. In contrast to that, test keywords
+must be all uppercase:
+
+```cmake
+if(NOT CONDITION)
+        doit()
+elseif(FOO AND BAR)
+        dothat()
+endif()
+```
+
+### Targets
+
+CMake code should not use functions that modify the global scope but prefer
+their targeted equivalents, instead. E.g. instead of using
+`include_directories()`, you must use `target_include_directories()`. An
+exception to this rule is setting up global compiler flags like warnings or
+flags required to set up the build type.
+
+### Dependencies
+
+Dependencies should not be discovered or set up in the main "CMakeLists.txt"
+module. Instead, they should either have their own module in our top-level
+"cmake/" directory or have a "CMakeLists.txt" in their respective "deps/"
+directory in case it is a vendored library. All dependencies should expose
+interface library targets that can be linked against with
+`target_link_libraries()`.
diff --git a/docs/contributing.md b/docs/contributing.md
index 21e42b1d9..f3e559c9d 100644
--- a/docs/contributing.md
+++ b/docs/contributing.md
@@ -18,7 +18,7 @@ The bundled dependencies in the `deps` directories are governed
 by the following licenses:
 
 - http-parser is licensed under [MIT license](../deps/http-parser/COPYING)
-- regex is governed by [LGPL v2.1+ license](../deps/regex/COPYING)
+- pcre is governed by [BSD license](../deps/pcre/LICENCE)
 - winhttp is governed by [LGPL v2.1+](../deps/winhttp/COPYING.LGPL) and [GPL v2 with linking exception](../deps/winhttp/COPYING.GPL)
 - zlib is governed by [zlib license](../deps/zlib/COPYING)
 
@@ -35,7 +35,7 @@ easily accessible permanent record of the conversation.
 
 ## Libgit2 Versions
 
-The `master` branch is the main branch where development happens.
+The `main` branch is the main branch where development happens.
 Releases are tagged
 (e.g. [v0.21.0](https://github.com/libgit2/libgit2/releases/tag/v0.21.0) )
 and when a critical bug fix needs to be backported, it will be done on a
@@ -51,7 +51,7 @@ commit SHA
 Using [`git describe`](http://git-scm.com/docs/git-describe) is a
 great way to tell us what version you're working with.
 
-If you're not running against the latest `master` branch version,
+If you're not running against the latest `main` branch version,
 please compile and test against that to avoid re-reporting an issue that's
 already been fixed.
 
@@ -68,11 +68,11 @@ flow](https://guides.github.com/introduction/flow/index.html), where
 contributors fork the [libgit2 repository](https://github.com/libgit2/libgit2),
 make their changes on branch, and submit a
 [Pull Request](https://help.github.com/articles/using-pull-requests)
-(a.k.a. "PR").  Pull requests should usually be targeted at the `master`
+(a.k.a. "PR").  Pull requests should usually be targeted at the `main`
 branch.
 
 Life will be a lot easier for you (and us) if you follow this pattern
-(i.e. fork, named branch, submit PR).  If you use your fork's `master`
+(i.e. fork, named branch, submit PR).  If you use your fork's `main`
 branch directly, things can get messy.
 
 Please include a nice description of your changes when you submit your PR;
diff --git a/docs/fuzzing.md b/docs/fuzzing.md
index cd825766b..25b32cb04 100644
--- a/docs/fuzzing.md
+++ b/docs/fuzzing.md
@@ -3,7 +3,7 @@
 libgit2 is currently using [libFuzzer](https://libfuzzer.info) to perform
 automated fuzz testing. libFuzzer only works with clang.
 
-## Prerequisites** for building fuzz targets:
+## Prerequisites for building fuzz targets:
 
 1. All the prerequisites for [building libgit2](https://github.com/libgit2/libgit2).
 2. A recent version of clang. 6.0 is preferred. [pre-build Debian/Ubuntu
@@ -27,14 +27,15 @@ automated fuzz testing. libFuzzer only works with clang.
 
 ## Run the fuzz targets
 
-1. `ASAN_SYMBOLIZER_PATH=/usr/bin/llvm-symbolize-6.0
+1. `ASAN_SYMBOLIZER_PATH=/usr/bin/llvm-symbolize
    LSAN_OPTIONS=allocator_may_return_null=1
-   ASAN_OPTIONS=allocator_may_return_null=1 ./build/fuzz/fuzz_packfile_raw
-   fuzz/corpora/fuzz_packfile_raw/`
+   ASAN_OPTIONS=allocator_may_return_null=1 ./build/fuzzers/packfile_fuzzer
+   fuzzers/corpora/packfile/`
 
 The `LSAN_OPTIONS` and `ASAN_OPTIONS` are there to allow `malloc(3)` to return
-`NULL`. The `LLVM_PROFILE_FILE` is there to override the path where libFuzzer
-will write the coverage report.
+`NULL`, which is expected if a huge chunk of memory is allocated. The
+`LLVM_PROFILE_FILE` environment string can also be added to override the path
+where libFuzzer will write the coverage report.
 
 ## Get coverage
 
diff --git a/docs/release.md b/docs/release.md
index 22b35ede7..3200c4449 100644
--- a/docs/release.md
+++ b/docs/release.md
@@ -1,6 +1,6 @@
 # Releasing the library
 
-We have three kinds of releases: "full" releases, maintenance releases and security releases. Full ones release the state of the `master` branch whereas maintenance releases provide bugfixes building on top of the currently released series. Security releases are also for the current series but only contain security fixes on top of the previous release.
+We have three kinds of releases: "full" releases, maintenance releases and security releases. Full ones release the state of the `main` branch whereas maintenance releases provide bugfixes building on top of the currently released series. Security releases are also for the current series but only contain security fixes on top of the previous release.
 
 ## Full release
 
@@ -40,7 +40,7 @@ followed by the three sections in the changelog. For release candidates we can a
 
 During the freeze, and certainly after the first release candidate, any bindings the core team work with should be updated in order to discover any issues that might come up with the multitude of approaches to memory management, embedding or linking.
 
-Create a branch `maint/v0.X` at the current state of `master` after you've created the tag. This will be used for maintenance releases and lets our dependents track the latest state of the series.
+Create a branch `maint/v0.X` at the current state of `main` after you've created the tag. This will be used for maintenance releases and lets our dependents track the latest state of the series.
 
 ## Maintenance release
 
@@ -76,7 +76,7 @@ We use docurium to generate our documentation. It is a tool written in ruby whic
 
     gem install docurium
 
-and run it against our description file with the tip of master checked out.
+and run it against our description file with the tip of `main` checked out.
 
     cm doc api.docurium
 
diff --git a/docs/win32-longpaths.md b/docs/win32-longpaths.md
new file mode 100644
index 000000000..a18152fc9
--- /dev/null
+++ b/docs/win32-longpaths.md
@@ -0,0 +1,36 @@
+core.longpaths support
+======================
+
+Historically, Windows has limited absolute path lengths to `MAX_PATH`
+(260) characters.
+
+Unfortunately, 260 characters is a punishing small maximum.  This is
+especially true for developers where dependencies may have dependencies
+in a folder, each dependency themselves having dependencies in a
+sub-folder, ad (seemingly) infinitum.
+
+So although the Windows APIs _by default_ honor this 260 character
+maximum, you can get around this by using separate APIs.  Git honors a
+`core.longpaths` configuration option that allows some paths on Windows
+to exceed these 260 character limits.
+
+And because they've gone and done it, that means that libgit2 has to
+honor this value, too.
+
+Since `core.longpaths` is a _configuration option_ that means that we
+need to be able to resolve a configuration - including in _the repository
+itself_ in order to know whether long paths should be supported.
+
+Therefore, in libgit2, `core.longpaths` affects paths in working
+directories _only_.  Paths to the repository, and to items inside the
+`.git` folder, must be no longer than 260 characters.
+
+This definition is required to avoid a paradoxical setting: if you
+had a repository in a folder that was 280 characters long, how would
+you know whether `core.longpaths` support should be enabled?  Even if
+`core.longpaths` was set to true in a system configuration file, the
+repository itself may set `core.longpaths` to false in _its_ configuration
+file, which you could only read if `core.longpaths` were set to true.
+
+Thus, `core.longpaths` must _only_ apply to working directory items,
+and cannot apply to the `.git` folder or its contents.