As explained in [1], we should not assume the reason why the writing has
failed, and even if the reason is that child has existed not the reason
why it have done so. So instead just say that writing has failed.

[1] https://public-inbox.org/git/20180604044408.GD14451@sigill.intra.peff.net/Signed-off-by: NMax Kirillov <max@max630.net>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

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

* bw/ref-prefix-for-configured-refspec:
  fetch: do not pass ref-prefixes for fetch by exact SHA1

When v2.18.0-rc0~10^2~1 (refspec: consolidate ref-prefix generation
logic, 2018-05-16) factored out the ref-prefix generation code for
reuse, it left out the 'if (!item->exact_sha1)' test in the original
ref-prefix generation code. As a result, fetches by SHA-1 generate
ref-prefixes as though the SHA-1 being fetched were an abbreviated ref
name:

 $ GIT_TRACE_PACKET=1 bin-wrappers/git -c protocol.version=2 \
	fetch origin 12039e00
[...]
 packet:        fetch> ref-prefix 12039e00
 packet:        fetch> ref-prefix refs/12039e00
 packet:        fetch> ref-prefix refs/tags/12039e00
 packet:        fetch> ref-prefix refs/heads/12039e00
 packet:        fetch> ref-prefix refs/remotes/12039e00
 packet:        fetch> ref-prefix refs/remotes/12039e00/HEAD
 packet:        fetch> 0000

If there is another ref name on the command line or the object being
fetched is already available locally, then that's mostly harmless.
But otherwise, we error out with

 fatal: no matching remote head

since the server did not send any refs we are interested in.  Filter
out the exact_sha1 refspecs to avoid this.

This patch adds a test to check this behavior that notices another
behavior difference between protocol v0 and v2 in the process.  Add a
NEEDSWORK comment to clear it up.
Signed-off-by: NJonathan Nieder <jrnieder@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

Quite a many tests assumed that newly created refs are made as
loose refs using the files backend, which have been updated to use
proper plumbing like rev-parse and update-ref, to avoid breakage
once we start using different ref backends.

* cc/tests-without-assuming-ref-files-backend:
  t990X: use '.git/objects' as 'deep inside .git' path
  t: make many tests depend less on the refs being files

Message fix.

* rd/init-typo:
  init: fix grammar in "templates not found" msg

Hotfixes.

* js/rebase-recreate-merge:
  sequencer: ensure labels that are object IDs are rewritten
  git-rebase--interactive: fix copy-paste mistake

Docfix.

* rd/tag-doc-lightweight:
  tag: clarify in the doc that a tag can refer to a non-commit object

Docfix.

* rd/doc-options-placeholder:
  Use proper syntax for replaceables in command docs

"git rev-parse Y..." etc. misbehaved when given endpoints were
not committishes.

* en/rev-parse-invalid-range:
  rev-parse: check lookup'ed commit references for NULL

"git p4" learned to "unshelve" shelved commit from P4.

* ld/p4-unshelve:
  git-p4: add unshelve command

Code simplification.

* nd/use-opt-int-set-f:
  Use OPT_SET_INT_F() for cmdline option specification

The import-tars script (in contrib/) has been taught to handle
tarballs with overly long paths that use PAX extended headers.

* pa/import-tars-long-names:
  import-tars: read overlong names from pax extended header

The list of commands with their various attributes were spread
across a few places in the build procedure, but it now is getting a
bit more consolidated to allow more automation.

* nd/command-list:
  completion: allow to customize the completable command list
  completion: add and use --list-cmds=alias
  completion: add and use --list-cmds=nohelpers
  Move declaration for alias.c to alias.h
  completion: reduce completable command list
  completion: let git provide the completable command list
  command-list.txt: documentation and guide line
  help: use command-list.txt for the source of guides
  help: add "-a --verbose" to list all commands with synopsis
  git: support --list-cmds=list-<category>
  completion: implement and use --list-cmds=main,others
  git --list-cmds: collect command list in a string_list
  git.c: convert --list-* to --list-cmds=*
  Remove common-cmds.h
  help: use command-list.h for common command list
  generate-cmds.sh: export all commands to command-list.h
  generate-cmds.sh: factor out synopsis extract code

In the 2.18 cycle, directory rename detection was merged, then reverted,
then reworked in such a way to fix another prominent bug in addition to
the original problem causing it to be reverted.  When the reworked series
was merged, we ended up with two nearly duplicate release notes.  Remove
the second copy, but preserve the information about the extra bug fix.
Signed-off-by: NElijah Newren <newren@gmail.com>
Signed-off-by: NJunio C Hamano <gitster@pobox.com>

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

Our HTTP client code used to advertise that we accept gzip encoding
from the other side; instead, just let cURL library to advertise
and negotiate the best one.

* bw/remote-curl-compressed-responses:
  remote-curl: accept compressed responses with protocol v2
  remote-curl: accept all encodings supported by curl

Leak plugging.

* ma/unpack-trees-free-msgs:
  unpack_trees_options: free messages when done
  argv-array: return the pushed string from argv_push*()
  merge-recursive: provide pair of `unpack_trees_{start,finish}()`
  merge: setup `opts` later in `checkout_fast_forward()`

Many tests hardcode the raw object names, which would change once
we migrate away from SHA-1.  While some of them must test against
exact object names, most of them do not have to use hardcoded
constants in the test.  The latter kind of tests have been updated
to test the moral equivalent of the original without hardcoding the
actual object names.

* bc/hash-independent-tests: (28 commits)
  t5300: abstract away SHA-1-specific constants
  t4208: abstract away SHA-1-specific constants
  t4045: abstract away SHA-1-specific constants
  t4042: abstract away SHA-1-specific constants
  t4205: sort log output in a hash-independent way
  t/lib-diff-alternative: abstract away SHA-1-specific constants
  t4030: abstract away SHA-1-specific constants
  t4029: abstract away SHA-1-specific constants
  t4029: fix test indentation
  t4022: abstract away SHA-1-specific constants
  t4020: abstract away SHA-1-specific constants
  t4014: abstract away SHA-1-specific constants
  t4008: abstract away SHA-1-specific constants
  t4007: abstract away SHA-1-specific constants
  t3905: abstract away SHA-1-specific constants
  t3702: abstract away SHA-1-specific constants
  t3103: abstract away SHA-1-specific constants
  t2203: abstract away SHA-1-specific constants
  t: skip pack tests if not using SHA-1
  t4044: skip test if not using SHA-1
  ...

We used to call regfree() after regcomp() failed in some codepaths,
which have been corrected.

* ma/regex-no-regfree-after-comp-fail:
  regex: do not call `regfree()` if compilation fails

Leak plugging.

* ma/config-store-data-clear:
  config: let `config_store_data_clear()` handle `key`
  config: let `config_store_data_clear()` handle `value_regex`
  config: free resources of `struct config_store_data`

Avoid unchecked snprintf() to make future code auditing easier.

* jk/snprintf-truncation:
  fmt_with_err: add a comment that truncation is OK
  shorten_unambiguous_ref: use xsnprintf
  fsmonitor: use internal argv_array of struct child_process
  log_write_email_headers: use strbufs
  http: use strbufs instead of fixed buffers

Error codepath fix.

* jk/config-blob-sans-repo:
  config: die when --blob is used outside a repository

By code restructuring of submodule merge in merge-recursive,
informational messages from the codepath are now given using the
same mechanism as other output, and honor the merge.verbosity
configuration.  The code also learned to give a few new messages
when a submodule three-way merge resolves cleanly when one side
records a descendant of the commit chosen by the other side.

* sb/submodule-merge-in-merge-recursive:
  merge-recursive: give notice when submodule commit gets fast-forwarded
  merge-recursive: i18n submodule merge output and respect verbosity
  submodule.c: move submodule merging to merge-recursive.c

Error codepath fix.

* js/empty-config-section-fix:
  config: a user-provided invalid section is not a BUG

"git fetch $there $refspec" that talks over protocol v2 can take
advantage of server-side ref filtering; the code has been extended
so that this mechanism triggers also when fetching with configured
refspec.

* bw/ref-prefix-for-configured-refspec: (38 commits)
  fetch: generate ref-prefixes when using a configured refspec
  refspec: consolidate ref-prefix generation logic
  submodule: convert push_unpushed_submodules to take a struct refspec
  remote: convert check_push_refs to take a struct refspec
  remote: convert match_push_refs to take a struct refspec
  http-push: store refspecs in a struct refspec
  transport: remove transport_verify_remote_names
  send-pack: store refspecs in a struct refspec
  transport: convert transport_push to take a struct refspec
  push: convert to use struct refspec
  push: check for errors earlier
  remote: convert match_explicit_refs to take a struct refspec
  remote: convert get_ref_match to take a struct refspec
  remote: convert query_refspecs to take a struct refspec
  remote: convert apply_refspecs to take a struct refspec
  remote: convert get_stale_heads to take a struct refspec
  fetch: convert prune_refs to take a struct refspec
  fetch: convert get_ref_map to take a struct refspec
  fetch: convert do_fetch to take a struct refspec
  refspec: remove the deprecated functions
  ...

Error behaviour of "git grep" when it cannot read the index was
inconsistent with other commands that uses the index, which has
been corrected to error out early.

* sb/grep-die-on-unreadable-index:
  grep: handle corrupt index files early

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

"git submodule update" attempts two different kinds of "git fetch"
against the upstream repository to grab a commit bound at the
submodule's path, but it incorrectly gave up if the first kind
(i.e. a normal fetch) failed, making the second "last resort" one
(i.e. fetching an exact commit object by object name) ineffective.
This has been corrected.

* sb/submodule-update-try-harder:
  git-submodule.sh: try harder to fetch a submodule

Update credential-netrc helper (in contrib/) to allow customizing
the GPG used to decrypt the encrypted .netrc file.

* lm/credential-netrc:
  git-credential-netrc: accept gpg option
  git-credential-netrc: adapt to test framework for git

When a short hexadecimal string is used to name an object but there
are multiple objects that share the string as the prefix of their
names, the code lists these ambiguous candidates in a help message.
These object names are now sorted according to their types for
easier eyeballing.

* ab/get-short-oid:
  get_short_oid: sort ambiguous objects by type, then SHA-1
  sha1-name.c: move around the collect_ambiguous() function
  git-p4: change "commitish" typo to "committish"
  sha1-array.h: align function arguments
  sha1-name.c: remove stray newline

Transfer protocol v2 learned to support the partial clone.

* jt/partial-clone-proto-v2:
  {fetch,upload}-pack: support filter in protocol v2
  upload-pack: read config when serving protocol v2
  upload-pack: fix error message typo

Conversion from uchar[20] to struct object_id continues.

* bc/object-id: (42 commits)
  merge-one-file: compute empty blob object ID
  add--interactive: compute the empty tree value
  Update shell scripts to compute empty tree object ID
  sha1_file: only expose empty object constants through git_hash_algo
  dir: use the_hash_algo for empty blob object ID
  sequencer: use the_hash_algo for empty tree object ID
  cache-tree: use is_empty_tree_oid
  sha1_file: convert cached object code to struct object_id
  builtin/reset: convert use of EMPTY_TREE_SHA1_BIN
  builtin/receive-pack: convert one use of EMPTY_TREE_SHA1_HEX
  wt-status: convert two uses of EMPTY_TREE_SHA1_HEX
  submodule: convert several uses of EMPTY_TREE_SHA1_HEX
  sequencer: convert one use of EMPTY_TREE_SHA1_HEX
  merge: convert empty tree constant to the_hash_algo
  builtin/merge: switch tree functions to use object_id
  builtin/am: convert uses of EMPTY_TREE_SHA1_BIN to the_hash_algo
  sha1-file: add functions for hex empty tree and blob OIDs
  builtin/receive-pack: avoid hard-coded constants for push certs
  diff: specify abbreviation size in terms of the_hash_algo
  upload-pack: replace use of several hard-coded constants
  ...

"git blame" learns to unhighlight uninteresting metadata from the
originating commit on lines that are the same as the previous one,
and also paint lines in different colors depending on the age of
the commit.

* sb/blame-color:
  builtin/blame: add new coloring scheme config
  builtin/blame: highlight recently changed lines
  builtin/blame: dim uninteresting metadata lines

"git submodule update" and "git submodule add" supported the
"--reference" option to borrow objects from a neighbouring local
repository like "git clone" does, but lacked the more recent
invention "--dissociate".  Also "git submodule add" has been taught
to take the "--progress" option.

* cf/submodule-progress-dissociate:
  submodule: add --dissociate option to add/update commands
  submodule: add --progress option to add command
  submodule: clean up substitutions in script

Command line completion (in contrib/) learned to complete pathnames
for various commands better.

* sg/complete-paths:
  t9902-completion: exercise __git_complete_index_file() directly
  completion: don't return with error from __gitcomp_file_direct()
  completion: fill COMPREPLY directly when completing paths
  completion: improve handling quoted paths in 'git ls-files's output
  completion: remove repeated dirnames with 'awk' during path completion
  t9902-completion: ignore COMPREPLY element order in some tests
  completion: use 'awk' to strip trailing path components
  completion: let 'ls-files' and 'diff-index' filter matching paths
  completion: improve handling quoted paths on the command line
  completion: support completing non-ASCII pathnames
  completion: simplify prefix path component handling during path completion
  completion: move __git_complete_index_file() next to its helpers
  t9902-completion: add tests demonstrating issues with quoted pathnames

Developer support.  Use newer GCC on one of the builds done at
TravisCI.org to get more warnings and errors diagnosed.

* nd/travis-gcc-8:
  travis-ci: run gcc-8 on linux-gcc jobs

Memory optimization.

* nd/pack-struct-commit:
  commit.h: rearrange 'index' to shrink struct commit

"git update-ref A B" is supposed to ensure that ref A does not yet
exist when B is a NULL OID, but this check was not done correctly
for pseudo-refs outside refs/ hierarchy, e.g. MERGE_HEAD.

* ma/create-pseudoref-with-null-old-oid:
  refs: handle zero oid for pseudorefs
  t1400: add tests around adding/deleting pseudorefs
  refs.c: refer to "object ID", not "sha1", in error messages

Code clean-up to turn history traversal more robust in a
semi-corrupt repository.

* jk/unavailable-can-be-missing:
  mark_parents_uninteresting(): avoid most allocation
  mark_parents_uninteresting(): replace list with stack
  mark_parents_uninteresting(): drop missing object check
  mark_tree_contents_uninteresting(): drop missing object check