c9e2e197217828db25fe2bbd5f72e93d395529c7
   1Git 2.18 Release Notes
   2======================
   3
   4Updates since v2.17
   5-------------------
   6
   7UI, Workflows & Features
   8
   9 * Rename detection logic in "diff" family that is used in "merge" has
  10   learned to guess when all of x/a, x/b and x/c have moved to z/a,
  11   z/b and z/c, it is likely that x/d added in the meantime would also
  12   want to move to z/d by taking the hint that the entire directory
  13   'x' moved to 'z'.  A bug causing dirty files involved in a rename
  14   to be overwritten during merge has also been fixed as part of this
  15   work.
  16
  17 * "git filter-branch" learned to use a different exit code to allow
  18   the callers to tell the case where there was no new commits to
  19   rewrite from other error cases.
  20
  21 * When built with more recent cURL, GIT_SSL_VERSION can now specify
  22   "tlsv1.3" as its value.
  23
  24 * "git gui" learned that "~/.ssh/id_ecdsa.pub" and
  25   "~/.ssh/id_ed25519.pub" are also possible SSH key files.
  26   (merge 2e2f0288ef bb/git-gui-ssh-key-files later to maint).
  27
  28 * "git gui" performs commit upon CTRL/CMD+ENTER but the
  29   CTRL/CMD+KP_ENTER (i.e. enter key on the numpad) did not have the
  30   same key binding.  It now does.
  31   (merge 28a1d94a06 bp/git-gui-bind-kp-enter later to maint).
  32
  33 * "git gui" has been taught to work with old versions of tk (like
  34   8.5.7) that do not support "ttk::style theme use" as a way to query
  35   the current theme.
  36   (merge 4891961105 cb/git-gui-ttk-style later to maint).
  37
  38 * "git rebase" has learned to honor "--signoff" option when using
  39   backends other than "am" (but not "--preserve-merges").
  40
  41 * "git branch --list" during an interrupted "rebase -i" now lets
  42   users distinguish the case where a detached HEAD is being rebased
  43   and a normal branch is being rebased.
  44
  45 * "git mergetools" learned talking to guiffy.
  46
  47 * The scripts in contrib/emacs/ have outlived their usefulness and
  48   have been replaced with a stub that errors out and tells the user
  49   there are replacements.
  50
  51 * The new "checkout-encoding" attribute can ask Git to convert the
  52   contents to the specified encoding when checking out to the working
  53   tree (and the other way around when checking in).
  54
  55 * The "git config" command uses separate options e.g. "--int",
  56   "--bool", etc. to specify what type the caller wants the value to
  57   be interpreted as.  A new "--type=<typename>" option has been
  58   introduced, which would make it cleaner to define new types.
  59
  60 * "git config --get" learned the "--default" option, to help the
  61   calling script.  Building on top of the above changes, the
  62   "git config" learns "--type=color" type.  Taken together, you can
  63   do things like "git config --get foo.color --default blue" and get
  64   the ANSI color sequence for the color given to foo.color variable,
  65   or "blue" if the variable does not exist.
  66
  67 * "git ls-remote" learned an option to allow sorting its output based
  68   on the refnames being shown.
  69
  70 * The command line completion (in contrib/) has been taught that "git
  71   stash save" has been deprecated ("git stash push" is the preferred
  72   spelling in the new world) and does not offer it as a possible
  73   completion candidate when "git stash push" can be.
  74
  75 * "git gc --prune=nonsense" spent long time repacking and then
  76   silently failed when underlying "git prune --expire=nonsense"
  77   failed to parse its command line.  This has been corrected.
  78
  79 * Error messages from "git push" can be painted for more visibility.
  80
  81 * "git http-fetch" (deprecated) had an optional and experimental
  82   "feature" to fetch only commits and/or trees, which nobody used.
  83   This has been removed.
  84
  85 * The functionality of "$GIT_DIR/info/grafts" has been superseded by
  86   the "refs/replace/" mechanism for some time now, but the internal
  87   code had support for it in many places, which has been cleaned up
  88   in order to drop support of the "grafts" mechanism.
  89
  90 * "git worktree add" learned to check out an existing branch.
  91
  92 * "git --no-pager cmd" did not have short-and-sweet single letter
  93   option. Now it does as "-P".
  94   (merge 7213c28818 js/no-pager-shorthand later to maint).
  95
  96 * "git rebase" learned "--rebase-merges" to transplant the whole
  97   topology of commit graph elsewhere.
  98
  99 * "git status" learned to pay attention to UI related diff
 100   configuration variables such as diff.renames.
 101
 102 * The command line completion mechanism (in contrib/) learned to load
 103   custom completion file for "git $command" where $command is a
 104   custom "git-$command" that the end user has on the $PATH when using
 105   newer version of bash.
 106
 107 * "git send-email" can sometimes offer confirmation dialog "Send this
 108   email?" with choices 'Yes', 'No', 'Quit', and 'All'.  A new action
 109   'Edit' has been added to this dialog's choice.
 110
 111 * With merge.renames configuration set to false, the recursive merge
 112   strategy can be told not to spend cycles trying to find renamed
 113   paths and merge them accordingly.
 114
 115 * "git status" learned to honor a new status.renames configuration to
 116   skip rename detection, which could be useful for those who want to
 117   do so without disabling the default rename detection done by the
 118   "git diff" command.
 119
 120 * Command line completion (in contrib/) learned to complete pathnames
 121   for various commands better.
 122
 123 * "git blame" learns to unhighlight uninteresting metadata from the
 124   originating commit on lines that are the same as the previous one,
 125   and also paint lines in different colors depending on the age of
 126   the commit.
 127
 128 * Transfer protocol v2 learned to support the partial clone.
 129
 130 * When a short hexadecimal string is used to name an object but there
 131   are multiple objects that share the string as the prefix of their
 132   names, the code lists these ambiguous candidates in a help message.
 133   These object names are now sorted according to their types for
 134   easier eyeballing.
 135
 136 * "git fetch $there $refspec" that talks over protocol v2 can take
 137   advantage of server-side ref filtering; the code has been extended
 138   so that this mechanism triggers also when fetching with configured
 139   refspec.
 140
 141 * Our HTTP client code used to advertise that we accept gzip encoding
 142   from the other side; instead, just let cURL library to advertise
 143   and negotiate the best one.
 144
 145
 146Performance, Internal Implementation, Development Support etc.
 147
 148 * A "git fetch" from a repository with insane number of refs into a
 149   repository that is already up-to-date still wasted too many cycles
 150   making many lstat(2) calls to see if these objects at the tips
 151   exist as loose objects locally.  These lstat(2) calls are optimized
 152   away by enumerating all loose objects beforehand.
 153   It is unknown if the new strategy negatively affects existing use
 154   cases, fetching into a repository with many loose objects from a
 155   repository with small number of refs.
 156
 157 * Git can be built to use either v1 or v2 of the PCRE library, and so
 158   far, the build-time configuration USE_LIBPCRE=YesPlease instructed
 159   the build procedure to use v1, but now it means v2.  USE_LIBPCRE1
 160   and USE_LIBPCRE2 can be used to explicitly choose which version to
 161   use, as before.
 162
 163 * The build procedure learned to optionally use symbolic links
 164   (instead of hardlinks and copies) to install "git-foo" for built-in
 165   commands, whose binaries are all identical.
 166
 167 * Conversion from uchar[20] to struct object_id continues.
 168
 169 * The way "git worktree prune" worked internally has been simplified,
 170   by assuming how "git worktree move" moves an existing worktree to a
 171   different place.
 172
 173 * Code clean-up for the "repository" abstraction.
 174   (merge 00a3da2a13 nd/remove-ignore-env-field later to maint).
 175
 176 * Code to find the length to uniquely abbreviate object names based
 177   on packfile content, which is a relatively recent addtion, has been
 178   optimized to use the same fan-out table.
 179
 180 * The mechanism to use parse-options API to automate the command line
 181   completion continues to get extended and polished.
 182
 183 * Copies of old scripted Porcelain commands in contrib/examples/ have
 184   been removed.
 185
 186 * Some tests that rely on the exact hardcoded values of object names
 187   have been updated in preparation for hash function migration.
 188
 189 * Perf-test update.
 190
 191 * Test helper update.
 192
 193 * The effort continues to refactor the internal global data structure
 194   to make it possible to open multiple repositories, work with and
 195   then close them,
 196
 197 * Small test-helper programs have been consolidated into a single
 198   binary.
 199
 200 * API clean-up around ref-filter code.
 201
 202 * Shell completion (in contrib) that gives list of paths have been
 203   optimized somewhat.
 204
 205 * The index file is updated to record the fsmonitor section after a
 206   full scan was made, to avoid wasting the effort that has already
 207   spent.
 208
 209 * Performance measuring framework in t/perf learned to help bisecting
 210   performance regressions.
 211
 212 * Some multi-word source filenames are being renamed to separate
 213   words with dashes instead of underscores.
 214
 215 * An reusable "memory pool" implementation has been extracted from
 216   fast-import.c, which in turn has become the first user of the
 217   mem-pool API.
 218
 219 * A build-time option has been added to allow Git to be told to refer
 220   to its associated files relative to the main binary, in the same
 221   way that has been possible on Windows for quite some time, for
 222   Linux, BSDs and Darwin.
 223
 224 * Precompute and store information necessary for ancestry traversal
 225   in a separate file to optimize graph walking.
 226
 227 * The effort to pass the repository in-core structure throughout the
 228   API continues.  This round deals with the code that implements the
 229   refs/replace/ mechanism.
 230
 231 * The build procedure "make DEVELOPER=YesPlease" learned to enable a
 232   bit more warning options depending on the compiler used to help
 233   developers more.  There also is "make DEVOPTS=tokens" knob
 234   available now, for those who want to help fixing warnings we
 235   usually ignore, for example.
 236
 237 * A new version of the transport protocol is being worked on.
 238
 239 * The code to interface to GPG has been restructured somewhat to make
 240   it cleaner to integrate with other types of signature systems later.
 241
 242 * The code has been taught to use the duplicated information stored
 243   in the commit-graph file to learn the tree object name for a commit
 244   to avoid opening and parsing the commit object when it makes sense
 245   to do so.
 246
 247 * "git gc" in a large repository takes a lot of time as it considers
 248   to repack all objects into one pack by default.  The command has
 249   been taught to pretend as if the largest existing packfile is
 250   marked with ".keep" so that it is left untouched while objects in
 251   other packs and loose ones are repacked.
 252
 253 * The transport protocol v2 is getting updated further.
 254
 255 * The codepath around object-info API has been taught to take the
 256   repository object (which in turn tells the API which object store
 257   the objects are to be located).
 258
 259 * Rename detection logic in "diff" family that is used in "merge" has
 260   learned to guess when all of x/a, x/b and x/c have moved to z/a,
 261   z/b and z/c, it is likely that x/d added in the meantime would also
 262   want to move to z/d by taking the hint that the entire directory
 263   'x' moved to 'z'.  A bug causing dirty files involved in a rename
 264   to be overwritten during merge has also been fixed as part of this
 265   work.  Incidentally, this also avoids updating a file in the
 266   working tree after a (non-trivial) merge whose result matches what
 267   our side originally had.
 268
 269 * "git pack-objects" needs to allocate tons of "struct object_entry"
 270   while doing its work, and shrinking its size helps the performance
 271   quite a bit.
 272
 273 * The implementation of "git rebase -i --root" has been updated to use
 274   the sequencer machinery more.
 275
 276 * Developer support update, by using BUG() macro instead of die() to
 277   mark codepaths that should not happen more clearly.
 278
 279 * Developer support.  Use newer GCC on one of the builds done at
 280   TravisCI.org to get more warnings and errors diagnosed.
 281
 282 * Conversion from uchar[20] to struct object_id continues.
 283
 284 * By code restructuring of submodule merge in merge-recursive,
 285   informational messages from the codepath are now given using the
 286   same mechanism as other output, and honor the merge.verbosity
 287   configuration.  The code also learned to give a few new messages
 288   when a submodule three-way merge resolves cleanly when one side
 289   records a descendant of the commit chosen by the other side.
 290
 291 * Avoid unchecked snprintf() to make future code auditing easier.
 292   (merge ac4896f007 jk/snprintf-truncation later to maint).
 293
 294 * Many tests hardcode the raw object names, which would change once
 295   we migrate away from SHA-1.  While some of them must test against
 296   exact object names, most of them do not have to use hardcoded
 297   constants in the test.  The latter kind of tests have been updated
 298   to test the moral equivalent of the original without hardcoding the
 299   actual object names.
 300
 301
 302Also contains various documentation updates and code clean-ups.
 303
 304
 305Fixes since v2.17
 306-----------------
 307
 308 * "git shortlog cruft" aborted with a BUG message when run outside a
 309   Git repository.  The command has been taught to complain about
 310   extra and unwanted arguments on its command line instead in such a
 311   case.
 312   (merge 4aa0161e83 ma/shortlog-revparse later to maint).
 313
 314 * "git stash push -u -- <pathspec>" gave an unnecessary and confusing
 315   error message when there was no tracked files that match the
 316   <pathspec>, which has been fixed.
 317   (merge 353278687e tg/stash-untracked-with-pathspec-fix later to maint).
 318
 319 * "git tag --contains no-such-commit" gave a full list of options
 320   after giving an error message.
 321   (merge 3bb0923f06 ps/contains-id-error-message later to maint).
 322
 323 * "diff-highlight" filter (in contrib/) learned to undertand "git log
 324   --graph" output better.
 325   (merge 4551fbba14 jk/diff-highlight-graph-fix later to maint).
 326
 327 * when refs that do not point at committish are given, "git
 328   filter-branch" gave a misleading error messages.  This has been
 329   corrected.
 330   (merge f78ab355e7 yk/filter-branch-non-committish-refs later to maint).
 331
 332 * "git submodule status" misbehaved on a submodule that has been
 333   removed from the working tree.
 334   (merge 74b6bda32f rs/status-with-removed-submodule later to maint).
 335
 336 * When credential helper exits very quickly without reading its
 337   input, it used to cause Git to die with SIGPIPE, which has been
 338   fixed.
 339   (merge a0d51e8d0e eb/cred-helper-ignore-sigpipe later to maint).
 340
 341 * "git rebase --keep-empty" still removed an empty commit if the
 342   other side contained an empty commit (due to the "does an
 343   equivalent patch exist already?" check), which has been corrected.
 344   (merge 3d946165e1 pw/rebase-keep-empty-fixes later to maint).
 345
 346 * Some codepaths, including the refs API, get and keep relative
 347   paths, that go out of sync when the process does chdir(2).  The
 348   chdir-notify API is introduced to let these codepaths adjust these
 349   cached paths to the new current directory.
 350   (merge fb9c2d2703 jk/relative-directory-fix later to maint).
 351
 352 * "cd sub/dir && git commit ../path" ought to record the changes to
 353   the file "sub/path", but this regressed long time ago.
 354   (merge 86238e07ef bw/commit-partial-from-subdirectory-fix later to maint).
 355
 356 * Recent introduction of "--log-destination" option to "git daemon"
 357   did not work well when the daemon was run under "--inetd" mode.
 358   (merge e67d906d73 lw/daemon-log-destination later to maint).
 359
 360 * Small fix to the autoconf build procedure.
 361   (merge 249482daf0 es/fread-reads-dir-autoconf-fix later to maint).
 362
 363 * Fix an unexploitable (because the oversized contents are not under
 364   attacker's control) buffer overflow.
 365   (merge d8579accfa bp/fsmonitor-bufsize-fix later to maint).
 366
 367 * Recent simplification of build procedure forgot a bit of tweak to
 368   the build procedure of contrib/mw-to-git/
 369   (merge d8698987f3 ab/simplify-perl-makefile later to maint).
 370
 371 * Moving a submodule that itself has submodule in it with "git mv"
 372   forgot to make necessary adjustment to the nested sub-submodules;
 373   now the codepath learned to recurse into the submodules.
 374
 375 * "git config --unset a.b", when "a.b" is the last variable in an
 376   otherwise empty section "a", left an empty section "a" behind, and
 377   worse yet, a subsequent "git config a.c value" did not reuse that
 378   empty shell and instead created a new one.  These have been
 379   (partially) corrected.
 380   (merge c71d8bb38a js/empty-config-section-fix later to maint).
 381
 382 * "git worktree remove" learned that "-f" is a shorthand for
 383   "--force" option, just like for "git worktree add".
 384   (merge d228eea514 sb/worktree-remove-opt-force later to maint).
 385
 386 * The completion script (in contrib/) learned to clear cached list of
 387   command line options upon dot-sourcing it again in a more efficient
 388   way.
 389   (merge 94408dc71c sg/completion-clear-cached later to maint).
 390
 391 * "git svn" had a minor thinko/typo which has been fixed.
 392   (merge 51db271587 ab/git-svn-get-record-typofix later to maint).
 393
 394 * During a "rebase -i" session, the code could give older timestamp
 395   to commits created by later "pick" than an earlier "reword", which
 396   has been corrected.
 397   (merge 12f7babd6b js/ident-date-fix later to maint).
 398
 399 * "git submodule status" did not check the symbolic revision name it
 400   computed for the submodule HEAD is not the NULL, and threw it at
 401   printf routines, which has been corrected.
 402   (merge 0b5e2ea7cf nd/submodule-status-fix later to maint).
 403
 404 * When fed input that already has In-Reply-To: and/or References:
 405   headers and told to add the same information, "git send-email"
 406   added these headers separately, instead of appending to an existing
 407   one, which is a violation of the RFC.  This has been corrected.
 408   (merge 256be1d3f0 sa/send-email-dedup-some-headers later to maint).
 409
 410 * "git fast-export" had a regression in v2.15.0 era where it skipped
 411   some merge commits in certain cases, which has been corrected.
 412   (merge be011bbe00 ma/fast-export-skip-merge-fix later to maint).
 413
 414 * The code did not propagate the terminal width to subprocesses via
 415   COLUMNS environment variable, which it now does.  This caused
 416   trouble to "git column" helper subprocess when "git tag --column=row"
 417   tried to list the existing tags on a display with non-default width.
 418   (merge b5d5a567fb nd/term-columns later to maint).
 419
 420 * We learned that our source files with ".pl" and ".py" extensions
 421   are Perl and Python files respectively and changes to them are
 422   better viewed as such with appropriate diff drivers.
 423   (merge 7818b619e2 ab/perl-python-attrs later to maint).
 424
 425 * "git rebase -i" sometimes left intermediate "# This is a
 426   combination of N commits" message meant for the human consumption
 427   inside an editor in the final result in certain corner cases, which
 428   has been fixed.
 429   (merge 15ef69314d js/rebase-i-clean-msg-after-fixup-continue later to maint).
 430
 431 * A test to see if the filesystem normalizes UTF-8 filename has been
 432   updated to check what we need to know in a more direct way, i.e. a
 433   path created in NFC form can be accessed with NFD form (or vice
 434   versa) to cope with APFS as well as HFS.
 435   (merge 742ae10e35 tb/test-apfs-utf8-normalization later to maint).
 436
 437 * "git format-patch --cover --attach" created a broken MIME multipart
 438   message for the cover letter, which has been fixed by keeping the
 439   cover letter as plain text file.
 440   (merge 50cd54ef4e bc/format-patch-cover-no-attach later to maint).
 441
 442 * The split-index feature had a long-standing and dormant bug in
 443   certain use of the in-core merge machinery, which has been fixed.
 444   (merge 7db118303a en/unpack-trees-split-index-fix later to maint).
 445
 446 * Asciidoctor gives a reasonable imitation for AsciiDoc, but does not
 447   render illustration in a literal block correctly when indented with
 448   HT by default. The problem is fixed by forcing 8-space tabs.
 449   (merge 379805051d bc/asciidoctor-tab-width later to maint).
 450
 451 * Code clean-up to adjust to a more recent lockfile API convention that
 452   allows lockfile instances kept on the stack.
 453   (merge 0fa5a2ed8d ma/lockfile-cleanup later to maint).
 454
 455 * the_repository->index is not a allocated piece of memory but
 456   repo_clear() indiscriminately attempted to free(3) it, which has
 457   been corrected.
 458   (merge 74373b5f10 nd/repo-clear-keep-the-index later to maint).
 459
 460 * Code clean-up to avoid non-standard-conformant pointer arithmetic.
 461   (merge c112084af9 rs/no-null-ptr-arith-in-fast-export later to maint).
 462
 463 * Code clean-up to turn history traversal more robust in a
 464   semi-corrupt repository.
 465   (merge 8702b30fd7 jk/unavailable-can-be-missing later to maint).
 466
 467 * "git update-ref A B" is supposed to ensure that ref A does not yet
 468   exist when B is a NULL OID, but this check was not done correctly
 469   for pseudo-refs outside refs/ hierarchy, e.g. MERGE_HEAD.
 470
 471 * "git submodule update" and "git submodule add" supported the
 472   "--reference" option to borrow objects from a neighbouring local
 473   repository like "git clone" does, but lacked the more recent
 474   invention "--dissociate".  Also "git submodule add" has been taught
 475   to take the "--progress" option.
 476   (merge a0ef29341a cf/submodule-progress-dissociate later to maint).
 477
 478 * Update credential-netrc helper (in contrib/) to allow customizing
 479   the GPG used to decrypt the encrypted .netrc file.
 480   (merge 786ef50a23 lm/credential-netrc later to maint).
 481
 482 * "git submodule update" attempts two different kinds of "git fetch"
 483   against the upstream repository to grab a commit bound at the
 484   submodule's path, but it incorrectly gave up if the first kind
 485   (i.e. a normal fetch) failed, making the second "last resort" one
 486   (i.e. fetching an exact commit object by object name) ineffective.
 487   This has been corrected.
 488   (merge e30d833671 sb/submodule-update-try-harder later to maint).
 489
 490 * Error behaviour of "git grep" when it cannot read the index was
 491   inconsistent with other commands that uses the index, which has
 492   been corrected to error out early.
 493   (merge b2aa84c789 sb/grep-die-on-unreadable-index later to maint).
 494
 495 * We used to call regfree() after regcomp() failed in some codepaths,
 496   which have been corrected.
 497   (merge 17154b1576 ma/regex-no-regfree-after-comp-fail later to maint).
 498
 499 * Other minor doc, test and build updates and code cleanups.
 500   (merge 248f66ed8e nd/trace-with-env later to maint).
 501   (merge 14ced5562c ys/bisect-object-id-missing-conversion-fix later to maint).
 502   (merge 5988eb631a ab/doc-hash-brokenness later to maint).
 503   (merge a4d4e32a70 pk/test-avoid-pipe-hiding-exit-status later to maint).
 504   (merge 05e293c1ac jk/flockfile-stdio later to maint).
 505   (merge e9184b0789 jk/t5561-missing-curl later to maint).
 506   (merge b1801b85a3 nd/worktree-move later to maint).
 507   (merge bbd374dd20 ak/bisect-doc-typofix later to maint).
 508   (merge 4855f06fb3 mn/send-email-credential-doc later to maint).
 509   (merge 8523b1e355 en/doc-typoes later to maint).
 510   (merge 43b44ccfe7 js/t5404-path-fix later to maint).
 511   (merge decf711fc1 ps/test-chmtime-get later to maint).
 512   (merge 22d11a6e8e es/worktree-docs later to maint).
 513   (merge 92a5dbbc22 tg/use-git-contacts later to maint).
 514   (merge adc887221f tq/t1510 later to maint).
 515   (merge bed21a8ad6 sg/doc-gc-quote-mismatch-fix later to maint).
 516   (merge 73364e4f10 tz/doc-git-urls-reference later to maint).
 517   (merge cd1e606bad bc/mailmap-self later to maint).
 518   (merge f7997e3682 ao/config-api-doc later to maint).
 519   (merge ee930754d8 jk/apply-p-doc later to maint).
 520   (merge 011b648646 nd/pack-format-doc later to maint).
 521   (merge 87a6bb701a sg/t5310-jgit-bitmap-test later to maint).
 522   (merge f6b82970aa sg/t5516-fixes later to maint).
 523   (merge 4362da078e sg/t7005-spaces-in-filenames-cleanup later to maint).
 524   (merge 7d0ee47c11 js/test-unset-prereq later to maint).
 525   (merge 5356a3c354 ah/misc-doc-updates later to maint).
 526   (merge 92c4a7a129 nd/completion-aliasfiletype-typofix later to maint).
 527   (merge 58bd77b66a nd/pack-unreachable-objects-doc later to maint).
 528   (merge 4ed79d5203 sg/t6500-no-redirect-of-stdin later to maint).
 529   (merge 17b8a2d6cd jk/config-blob-sans-repo later to maint).