This fixes "convert: add filter.<driver>.process option" (edcc8581) on
Windows.

Consider the case of a file that requires filtering and is present in
branch A but not in branch B. If A is the current HEAD and we checkout B
then the following happens:

1. ce_compare_data() opens the file
2.   index_fd() detects that the file requires to run a clean filter and
     calls index_stream_convert_blob()
4.     index_stream_convert_blob() calls convert_to_git_filter_fd()
5.       convert_to_git_filter_fd() calls apply_filter() which creates a
         new long running filter process (in case it is the first file
         of this kind to be filtered)
6.       The new filter process inherits all file handles. This is the
         default on Linux/OSX and is explicitly defined in the
         `CreateProcessW` call in `mingw.c` on Windows.
7. ce_compare_data() closes the file
8. Git unlinks the file as it is not present in B

The unlink operation does not work on Windows because the filter process
has still an open handle to the file. On Linux/OSX the unlink operation
succeeds but the file descriptors still leak into the child process.

Fix this problem by opening files in read-cache with the O_CLOEXEC flag
to ensure that the file descriptor does not remain open in a newly
spawned process similar to 05d1ed61 ("mingw: ensure temporary file
handles are not inherited by child processes", 2016-08-22).
Signed-off-by: NLars Schneider <larsxschneider@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

All processes that the Git main process spawns inherit the open file
descriptors of the main process. These leaked file descriptors can
cause problems.

Use the O_CLOEXEC flag similar to 05d1ed61 to fix the leaked file
descriptors.
Signed-off-by: NLars Schneider <larsxschneider@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This function is meant to be used when reading from files in the
object store, and the original objective was to avoid smudging atime
of loose object files too often, hence its name.  Because we'll be
extending its role in the next commit to also arrange the file
descriptors they return auto-closed in the child processes, rename
it to lose "noatime" part that is too specific.
Signed-off-by: NLars Schneider <larsxschneider@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Codepaths involved in interacting alternate object store have
been cleaned up.

* jk/alt-odb-cleanup:
  alternates: use fspathcmp to detect duplicates
  sha1_file: always allow relative paths to alternates
  count-objects: report alternates via verbose mode
  fill_sha1_file: write into a strbuf
  alternates: store scratch buffer as strbuf
  fill_sha1_file: write "boring" characters
  alternates: use a separate scratch space
  alternates: encapsulate alt->base munging
  alternates: provide helper for allocating alternate
  alternates: provide helper for adding to alternates list
  link_alt_odb_entry: refactor string handling
  link_alt_odb_entry: handle normalize_path errors
  t5613: clarify "too deep" recursion tests
  t5613: do not chdir in main process
  t5613: whitespace/style cleanups
  t5613: use test_must_fail
  t5613: drop test_valid_repo function
  t5613: drop reachable_via function

Documentation for "git commit" was updated to clarify that "commit
-p <paths>" adds to the current contents of the index to come up
with what to commit.

* nd/commit-p-doc:
  git-commit.txt: clarify --patch mode with pathspec

"git clone" of a local repository can be done at the filesystem
level, but the codepath did not check errors while copying and
adjusting the file that lists alternate object stores.

* jk/clone-copy-alternates-fix:
  clone: detect errors in normalize_path_copy

http.emptyauth configuration is a way to allow an empty username to
pass when attempting to authenticate using mechanisms like
Kerberos.  We took an unspecified (NULL) username and sent ":"
(i.e. no username, no password) to CURLOPT_USERPWD, but did not do
the same when the username is explicitly set to an empty string.

* dt/http-empty-auth:
  http: http.emptyauth should allow empty (not just NULL) usernames

* maint:
  Start preparing for 2.10.2

Signed-off-by: NJunio C Hamano <gitster@pobox.com>

A low-level function verify_packfile() was meant to show errors
that were detected without dying itself, but under some conditions
it didn't and died instead, which has been fixed.

* jk/verify-packfile-gently:
  verify_packfile: check pack validity before accessing data

"git worktree", even though it used the default_abbrev setting that
ought to be affected by core.abbrev configuration variable, ignored
the variable setting.  The command has been taught to read the
default set of configuration variables to correct this.

* jc/worktree-config:
  worktree: honor configuration variables

Codepaths that read from an on-disk loose object were too loose in
validating what they are reading is a proper object file and
sometimes read past the data they read from the disk, which has
been corrected.  H/t to Gustavo Grieco for reporting.

* jc/verify-loose-object-header:
  unpack_sha1_header(): detect malformed object header
  streaming: make sure to notice corrupt object

The original command line syntax for "git merge", which was "git
merge <msg> HEAD <parent>...", has been deprecated for quite some
time, and "git gui" was the last in-tree user of the syntax.  This
is finally fixed, so that we can move forward with the deprecation.

* rs/git-gui-use-modern-git-merge-syntax:
  git-gui: stop using deprecated merge syntax

An author name, that spelled a backslash-quoted double quote in the
human readable part "My \"double quoted\" name", was not unquoted
correctly while applying a patch from a piece of e-mail.

* kd/mailinfo-quoted-string:
  mailinfo: unescape quoted-pair in header fields
  t5100-mailinfo: replace common path prefix with variable

Doc update to clarify what "log -3 --reverse" does.

* pb/rev-list-reverse-with-count:
  rev-list-options: clarify the usage of --reverse

Almost everybody uses DEFAULT_ABBREV to refer to the default
setting for the abbreviation, but "git blame" peeked into
underlying variable bypassing the macro for no good reason.

* jc/blame-abbrev:
  blame: use DEFAULT_ABBREV macro

The "graph" API used in "git log --graph" miscounted the number of
output columns consumed so far when drawing a padding line, which
has been fixed; this did not affect any existing code as nobody
tried to write anything after the padding on such a line, though.

* jk/graph-padding-fix:
  graph: fix extra spaces in graph_padding_line

The code that parses the format parameter of for-each-ref command
has seen a micro-optimization.

* sg/ref-filter-parse-optim:
  ref-filter: strip format option after a field name only once while parsing

Code cleanup.

* rs/copy-array:
  use COPY_ARRAY
  add COPY_ARRAY

* dt/mailinfo:
  add David Turner's Two Sigma address

Signed-off-by: NJunio C Hamano <gitster@pobox.com>

It is a common mistake to say "git blame --reverse OLD path",
expecting that the command line is dwimmed as if asking how lines
in path in an old revision OLD have survived up to the current
commit.

* jc/blame-reverse:
  blame: dwim "blame --reverse OLD" as "blame --reverse OLD.."
  blame: improve diagnosis for "--reverse NEW"

The existing "git fetch --depth=<n>" option was hard to use
correctly when making the history of an existing shallow clone
deeper.  A new option, "--deepen=<n>", has been added to make this
easier to use.  "git clone" also learned "--shallow-since=<date>"
and "--shallow-exclude=<tag>" options to make it easier to specify
"I am interested only in the recent N months worth of history" and
"Give me only the history since that version".

* nd/shallow-deepen: (27 commits)
  fetch, upload-pack: --deepen=N extends shallow boundary by N commits
  upload-pack: add get_reachable_list()
  upload-pack: split check_unreachable() in two, prep for get_reachable_list()
  t5500, t5539: tests for shallow depth excluding a ref
  clone: define shallow clone boundary with --shallow-exclude
  fetch: define shallow boundary with --shallow-exclude
  upload-pack: support define shallow boundary by excluding revisions
  refs: add expand_ref()
  t5500, t5539: tests for shallow depth since a specific date
  clone: define shallow clone boundary based on time with --shallow-since
  fetch: define shallow boundary with --shallow-since
  upload-pack: add deepen-since to cut shallow repos based on time
  shallow.c: implement a generic shallow boundary finder based on rev-list
  fetch-pack: use a separate flag for fetch in deepening mode
  fetch-pack.c: mark strings for translating
  fetch-pack: use a common function for verbose printing
  fetch-pack: use skip_prefix() instead of starts_with()
  upload-pack: move rev-list code out of check_non_tip()
  upload-pack: make check_non_tip() clean things up on error
  upload-pack: tighten number parsing at "deepen" lines
  ...

The command-line completion script (in contrib/) learned to
complete "git cmd ^mas<HT>" to complete the negative end of
reference to "git cmd ^master".

* cp/completion-negative-refs:
  completion: support excluding refs

The ./configure script generated from configure.ac was taught how
to detect support of SSL by libcurl better.

* dp/autoconf-curl-ssl:
  ./configure.ac: detect SSL in libcurl using curl-config

When we started cURL to talk to imap server when a new enough
version of cURL library is available, we forgot to explicitly add
imap(s):// before the destination.  To some folks, that didn't work
and the library tried to make HTTP(s) requests instead.

* ak/curl-imap-send-explicit-scheme:
  imap-send: Tell cURL to use imap:// or imaps://

"git pack-objects" in a repository with many packfiles used to
spend a lot of time looking for/at objects in them; the accesses to
the packfiles are now optimized by checking the most-recently-used
packfile first.

* jk/pack-objects-optim-mru:
  pack-objects: use mru list when iterating over packs
  pack-objects: break delta cycles before delta-search phase
  sha1_file: make packed_object_info public
  provide an initializer for "struct object_info"

We call "qsort(array, nelem, sizeof(array[0]), fn)", and most of
the time third parameter is redundant.  A new QSORT() macro lets us
omit it.

* rs/qsort:
  show-branch: use QSORT
  use QSORT, part 2
  coccicheck: use --all-includes by default
  remove unnecessary check before QSORT
  use QSORT
  add QSORT

On a case-insensitive filesystem, we should realize that
"a/objects" and "A/objects" are the same path. We already
use fspathcmp() to check against the main object directory,
but until recently we couldn't use it for comparing against
other alternates (because their paths were not
NUL-terminated strings). But now we can, so let's do so.

Note that we also need to adjust count-objects to load the
config, so that it can see the setting of core.ignorecase
(this is required by the test, but is also a general bugfix
for users of count-objects).
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

We recursively expand alternates repositories, so that if A
borrows from B which borrows from C, A can see all objects.

For the root object database, we allow relative paths, so A
can point to B as "../B/objects". However, we currently do
not allow relative paths when recursing, so B must use an
absolute path to reach C.

That is an ancient protection from c2f493a4 (Transitively
read alternatives, 2006-05-07) that tries to avoid adding
the same alternate through two different paths. Since
5bdf0a84 (sha1_file: normalize alt_odb path before comparing
and storing, 2011-09-07), we use a normalized absolute path
for each alt_odb entry.

This means that in most cases the protection is no longer
necessary; we will detect the duplicate no matter how we got
there (but see below).  And it's a good idea to get rid of
it, as it creates an unnecessary complication when setting
up recursive alternates (B has to know that A is going to
borrow from it and make sure to use an absolute path).

Note that our normalization doesn't actually look at the
filesystem, so it can still be fooled by crossing symbolic
links. But that's also true of absolute paths, so it's not a
good reason to disallow only relative paths (it's
potentially a reason to switch to real_path(), but that's a
separate and non-trivial change).

We adjust the test script here to demonstrate that this now
works, and add new tests to show that the normalization does
indeed suppress duplicates.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

There's no way to get the list of alternates that git
computes internally; our tests only infer it based on which
objects are available. In addition to testing, knowing this
list may be helpful for somebody debugging their alternates
setup.

Let's add it to the "count-objects -v" output. We could give
it a separate flag, but there's not really any need.
"count-objects -v" is already a debugging catch-all for the
object database, its output is easily extensible to new data
items, and printing the alternates is not expensive (we
already had to find them to count the objects).
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

It's currently the responsibility of the caller to give
fill_sha1_file() enough bytes to write into, leading them to
manually compute the required lengths. Instead, let's just
write into a strbuf so that it's impossible to get this
wrong.

The alt_odb caller already has a strbuf, so this makes
things strictly simpler. The other caller, sha1_file_name(),
uses a static PATH_MAX buffer and dies when it would
overflow. We can convert this to a static strbuf, which
means our allocation cost is amortized (and as a bonus, we
no longer have to worry about PATH_MAX being too short for
normal use).

This does introduce some small overhead in fill_sha1_file(),
as each strbuf_addchar() will check whether it needs to
grow. However, between the optimization in fec501da
(strbuf_addch: avoid calling strbuf_grow, 2015-04-16) and
the fact that this is not generally called in a tight loop
(after all, the next step is typically to access the file!)
this probably doesn't matter. And even if it did, the right
place to micro-optimize is inside fill_sha1_file(), by
calling a single strbuf_grow() there.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

We pre-size the scratch buffer to hold a loose object
filename of the form "xx/yyyy...", which leads to allocation
code that is hard to verify. We have to use some magic
numbers during the initial allocation, and then writers must
blindly assume that the buffer is big enough. Using a strbuf
makes it more clear that we cannot overflow.

Unfortunately, we do still need some magic numbers to grow
our strbuf before calling fill_sha1_path(), but the strbuf
growth is much closer to the point of use. This makes it
easier to see that it's correct, and opens the possibility
of pushing it even further down if fill_sha1_path() learns
to work on strbufs.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

This function forms a sha1 as "xx/yyyy...", but skips over
the slot for the slash rather than writing it, leaving it to
the caller to do so. It also does not bother to put in a
trailing NUL, even though every caller would want it (we're
forming a path which by definition is not a directory, so
the only thing to do with it is feed it to a system call).

Let's make the lives of our callers easier by just writing
out the internal "/" and the NUL.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The alternate_object_database struct uses a single buffer
both for storing the path to the alternate, and as a scratch
buffer for forming object names. This is efficient (since
otherwise we'd end up storing the path twice), but it makes
life hard for callers who just want to know the path to the
alternate. They have to remember to stop reading after
"alt->name - alt->base" bytes, and to subtract one for the
trailing '/'.

It would be much simpler if they could simply access a
NUL-terminated path string. We could encapsulate this in a
function which puts a NUL in the scratch buffer and returns
the string, but that opens up questions about the lifetime
of the result. The first time another caller uses the
alternate, the scratch buffer may get other data tacked onto
it.

Let's instead just store the root path separately from the
scratch buffer. There aren't enough alternates being stored
for the duplicated data to matter for performance, and this
keeps things simple and safe for the callers.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The alternate_object_database struct holds a path to the
alternate objects, but we also use that buffer as scratch
space for forming loose object filenames. Let's pull that
logic into a helper function so that we can more easily
modify it.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Allocating a struct alternate_object_database is tricky, as
we must over-allocate the buffer to provide scratch space,
and then put in particular '/' and NUL markers.

Let's encapsulate this in a function so that the complexity
doesn't leak into callers (and so that we can modify it
later).
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The submodule code wants to temporarily add an alternate
object store to our in-memory alt_odb list, but does it
manually. Let's provide a helper so it can reuse the code in
link_alt_odb_entry().

While we're adding our new add_to_alternates_memory(), let's
document add_to_alternates_file(), as the two are related.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

The string handling in link_alt_odb_entry() is mostly an
artifact of the original version, which took the path as a
ptr/len combo, and did not have a NUL-terminated string
until we created one in the alternate_object_database
struct.  But since 5bdf0a84 (sha1_file: normalize alt_odb
path before comparing and storing, 2011-09-07), the first
thing we do is put the path into a strbuf, which gives us
some easy opportunities for cleanup.

In particular:

  - we call strlen(pathbuf.buf), which is silly; we can look
    at pathbuf.len.

  - even though we have a strbuf, we don't maintain its
    "len" field when chomping extra slashes from the
    end, and instead keep a separate "pfxlen" variable. We
    can fix this and then drop "pfxlen" entirely.

  - we don't check whether the path is usable until after we
    allocate the new struct, making extra cleanup work for
    ourselves. Since we have a NUL-terminated string, we can
    bump the "is it usable" checks higher in the function.
    While we're at it, we can move that logic to its own
    helper, which makes the flow of link_alt_odb_entry()
    easier to follow.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

When we add a new alternate to the list, we try to normalize
out any redundant "..", etc. However, we do not look at the
return value of normalize_path_copy(), and will happily
continue with a path that could not be normalized. Worse,
the normalizing process is done in-place, so we are left
with whatever half-finished working state the normalizing
function was in.

Fortunately, this cannot cause us to read past the end of
our buffer, as that working state will always leave the
NUL from the original path in place. And we do tend to
notice problems when we check is_directory() on the path.
But you can see the nonsense that we feed to is_directory
with an entry like:

  this/../../is/../../way/../../too/../../deep/../../to/../../resolve

in your objects/info/alternates, which yields:

  error: object directory
  /to/e/deep/too/way//ects/this/../../is/../../way/../../too/../../deep/../../to/../../resolve
  does not exist; check .git/objects/info/alternates.

We can easily fix this just by checking the return value.
But that makes it hard to generate a good error message,
since we're normalizing in-place and our input value has
been overwritten by cruft.

Instead, let's provide a strbuf helper that does an in-place
normalize, but restores the original contents on error. This
uses a second buffer under the hood, which is slightly less
efficient, but this is not a performance-critical code path.

The strbuf helper can also properly set the "len" parameter
of the strbuf before returning. Just doing:

  normalize_path_copy(buf.buf, buf.buf);

will shorten the string, but leave buf.len at the original
length. That may be confusing to later code which uses the
strbuf.
Signed-off-by: NJeff King <peff@peff.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>