Code

Merge branch 'rs/diff-no-minimal' into maint
authorJunio C Hamano <gitster@pobox.com>
Mon, 21 Jun 2010 12:38:50 +0000 (05:38 -0700)
committerJunio C Hamano <gitster@pobox.com>
Mon, 21 Jun 2010 12:38:50 +0000 (05:38 -0700)
* rs/diff-no-minimal:
  git diff too slow for a file

1  2 
builtin/blame.c
builtin/merge-file.c
builtin/merge-tree.c
builtin/rerere.c
combine-diff.c
diff.c
merge-file.c

diff --cc builtin/blame.c
index 4dd4c3f494042c91e2fa481314d2d8d6d61d9c43,0000000000000000000000000000000000000000..8506286dd271d4e92369d81ec2cce9240a559d64
mode 100644,000000..100644
--- /dev/null
@@@ -1,2477 -1,0 +1,2477 @@@
- static int xdl_opts = XDF_NEED_MINIMAL;
 +/*
 + * Blame
 + *
 + * Copyright (c) 2006, Junio C Hamano
 + */
 +
 +#include "cache.h"
 +#include "builtin.h"
 +#include "blob.h"
 +#include "commit.h"
 +#include "tag.h"
 +#include "tree-walk.h"
 +#include "diff.h"
 +#include "diffcore.h"
 +#include "revision.h"
 +#include "quote.h"
 +#include "xdiff-interface.h"
 +#include "cache-tree.h"
 +#include "string-list.h"
 +#include "mailmap.h"
 +#include "parse-options.h"
 +#include "utf8.h"
 +
 +static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
 +
 +static const char *blame_opt_usage[] = {
 +      blame_usage,
 +      "",
 +      "[rev-opts] are documented in git-rev-list(1)",
 +      NULL
 +};
 +
 +static int longest_file;
 +static int longest_author;
 +static int max_orig_digits;
 +static int max_digits;
 +static int max_score_digits;
 +static int show_root;
 +static int reverse;
 +static int blank_boundary;
 +static int incremental;
++static int xdl_opts;
 +
 +static enum date_mode blame_date_mode = DATE_ISO8601;
 +static size_t blame_date_width;
 +
 +static struct string_list mailmap;
 +
 +#ifndef DEBUG
 +#define DEBUG 0
 +#endif
 +
 +/* stats */
 +static int num_read_blob;
 +static int num_get_patch;
 +static int num_commits;
 +
 +#define PICKAXE_BLAME_MOVE            01
 +#define PICKAXE_BLAME_COPY            02
 +#define PICKAXE_BLAME_COPY_HARDER     04
 +#define PICKAXE_BLAME_COPY_HARDEST    010
 +
 +/*
 + * blame for a blame_entry with score lower than these thresholds
 + * is not passed to the parent using move/copy logic.
 + */
 +static unsigned blame_move_score;
 +static unsigned blame_copy_score;
 +#define BLAME_DEFAULT_MOVE_SCORE      20
 +#define BLAME_DEFAULT_COPY_SCORE      40
 +
 +/* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
 +#define METAINFO_SHOWN                (1u<<12)
 +#define MORE_THAN_ONE_PATH    (1u<<13)
 +
 +/*
 + * One blob in a commit that is being suspected
 + */
 +struct origin {
 +      int refcnt;
 +      struct origin *previous;
 +      struct commit *commit;
 +      mmfile_t file;
 +      unsigned char blob_sha1[20];
 +      char path[FLEX_ARRAY];
 +};
 +
 +/*
 + * Given an origin, prepare mmfile_t structure to be used by the
 + * diff machinery
 + */
 +static void fill_origin_blob(struct origin *o, mmfile_t *file)
 +{
 +      if (!o->file.ptr) {
 +              enum object_type type;
 +              num_read_blob++;
 +              file->ptr = read_sha1_file(o->blob_sha1, &type,
 +                                         (unsigned long *)(&(file->size)));
 +              if (!file->ptr)
 +                      die("Cannot read blob %s for path %s",
 +                          sha1_to_hex(o->blob_sha1),
 +                          o->path);
 +              o->file = *file;
 +      }
 +      else
 +              *file = o->file;
 +}
 +
 +/*
 + * Origin is refcounted and usually we keep the blob contents to be
 + * reused.
 + */
 +static inline struct origin *origin_incref(struct origin *o)
 +{
 +      if (o)
 +              o->refcnt++;
 +      return o;
 +}
 +
 +static void origin_decref(struct origin *o)
 +{
 +      if (o && --o->refcnt <= 0) {
 +              if (o->previous)
 +                      origin_decref(o->previous);
 +              free(o->file.ptr);
 +              free(o);
 +      }
 +}
 +
 +static void drop_origin_blob(struct origin *o)
 +{
 +      if (o->file.ptr) {
 +              free(o->file.ptr);
 +              o->file.ptr = NULL;
 +      }
 +}
 +
 +/*
 + * Each group of lines is described by a blame_entry; it can be split
 + * as we pass blame to the parents.  They form a linked list in the
 + * scoreboard structure, sorted by the target line number.
 + */
 +struct blame_entry {
 +      struct blame_entry *prev;
 +      struct blame_entry *next;
 +
 +      /* the first line of this group in the final image;
 +       * internally all line numbers are 0 based.
 +       */
 +      int lno;
 +
 +      /* how many lines this group has */
 +      int num_lines;
 +
 +      /* the commit that introduced this group into the final image */
 +      struct origin *suspect;
 +
 +      /* true if the suspect is truly guilty; false while we have not
 +       * checked if the group came from one of its parents.
 +       */
 +      char guilty;
 +
 +      /* true if the entry has been scanned for copies in the current parent
 +       */
 +      char scanned;
 +
 +      /* the line number of the first line of this group in the
 +       * suspect's file; internally all line numbers are 0 based.
 +       */
 +      int s_lno;
 +
 +      /* how significant this entry is -- cached to avoid
 +       * scanning the lines over and over.
 +       */
 +      unsigned score;
 +};
 +
 +/*
 + * The current state of the blame assignment.
 + */
 +struct scoreboard {
 +      /* the final commit (i.e. where we started digging from) */
 +      struct commit *final;
 +      struct rev_info *revs;
 +      const char *path;
 +
 +      /*
 +       * The contents in the final image.
 +       * Used by many functions to obtain contents of the nth line,
 +       * indexed with scoreboard.lineno[blame_entry.lno].
 +       */
 +      const char *final_buf;
 +      unsigned long final_buf_size;
 +
 +      /* linked list of blames */
 +      struct blame_entry *ent;
 +
 +      /* look-up a line in the final buffer */
 +      int num_lines;
 +      int *lineno;
 +};
 +
 +static inline int same_suspect(struct origin *a, struct origin *b)
 +{
 +      if (a == b)
 +              return 1;
 +      if (a->commit != b->commit)
 +              return 0;
 +      return !strcmp(a->path, b->path);
 +}
 +
 +static void sanity_check_refcnt(struct scoreboard *);
 +
 +/*
 + * If two blame entries that are next to each other came from
 + * contiguous lines in the same origin (i.e. <commit, path> pair),
 + * merge them together.
 + */
 +static void coalesce(struct scoreboard *sb)
 +{
 +      struct blame_entry *ent, *next;
 +
 +      for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 +              if (same_suspect(ent->suspect, next->suspect) &&
 +                  ent->guilty == next->guilty &&
 +                  ent->s_lno + ent->num_lines == next->s_lno) {
 +                      ent->num_lines += next->num_lines;
 +                      ent->next = next->next;
 +                      if (ent->next)
 +                              ent->next->prev = ent;
 +                      origin_decref(next->suspect);
 +                      free(next);
 +                      ent->score = 0;
 +                      next = ent; /* again */
 +              }
 +      }
 +
 +      if (DEBUG) /* sanity */
 +              sanity_check_refcnt(sb);
 +}
 +
 +/*
 + * Given a commit and a path in it, create a new origin structure.
 + * The callers that add blame to the scoreboard should use
 + * get_origin() to obtain shared, refcounted copy instead of calling
 + * this function directly.
 + */
 +static struct origin *make_origin(struct commit *commit, const char *path)
 +{
 +      struct origin *o;
 +      o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
 +      o->commit = commit;
 +      o->refcnt = 1;
 +      strcpy(o->path, path);
 +      return o;
 +}
 +
 +/*
 + * Locate an existing origin or create a new one.
 + */
 +static struct origin *get_origin(struct scoreboard *sb,
 +                               struct commit *commit,
 +                               const char *path)
 +{
 +      struct blame_entry *e;
 +
 +      for (e = sb->ent; e; e = e->next) {
 +              if (e->suspect->commit == commit &&
 +                  !strcmp(e->suspect->path, path))
 +                      return origin_incref(e->suspect);
 +      }
 +      return make_origin(commit, path);
 +}
 +
 +/*
 + * Fill the blob_sha1 field of an origin if it hasn't, so that later
 + * call to fill_origin_blob() can use it to locate the data.  blob_sha1
 + * for an origin is also used to pass the blame for the entire file to
 + * the parent to detect the case where a child's blob is identical to
 + * that of its parent's.
 + */
 +static int fill_blob_sha1(struct origin *origin)
 +{
 +      unsigned mode;
 +
 +      if (!is_null_sha1(origin->blob_sha1))
 +              return 0;
 +      if (get_tree_entry(origin->commit->object.sha1,
 +                         origin->path,
 +                         origin->blob_sha1, &mode))
 +              goto error_out;
 +      if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
 +              goto error_out;
 +      return 0;
 + error_out:
 +      hashclr(origin->blob_sha1);
 +      return -1;
 +}
 +
 +/*
 + * We have an origin -- check if the same path exists in the
 + * parent and return an origin structure to represent it.
 + */
 +static struct origin *find_origin(struct scoreboard *sb,
 +                                struct commit *parent,
 +                                struct origin *origin)
 +{
 +      struct origin *porigin = NULL;
 +      struct diff_options diff_opts;
 +      const char *paths[2];
 +
 +      if (parent->util) {
 +              /*
 +               * Each commit object can cache one origin in that
 +               * commit.  This is a freestanding copy of origin and
 +               * not refcounted.
 +               */
 +              struct origin *cached = parent->util;
 +              if (!strcmp(cached->path, origin->path)) {
 +                      /*
 +                       * The same path between origin and its parent
 +                       * without renaming -- the most common case.
 +                       */
 +                      porigin = get_origin(sb, parent, cached->path);
 +
 +                      /*
 +                       * If the origin was newly created (i.e. get_origin
 +                       * would call make_origin if none is found in the
 +                       * scoreboard), it does not know the blob_sha1,
 +                       * so copy it.  Otherwise porigin was in the
 +                       * scoreboard and already knows blob_sha1.
 +                       */
 +                      if (porigin->refcnt == 1)
 +                              hashcpy(porigin->blob_sha1, cached->blob_sha1);
 +                      return porigin;
 +              }
 +              /* otherwise it was not very useful; free it */
 +              free(parent->util);
 +              parent->util = NULL;
 +      }
 +
 +      /* See if the origin->path is different between parent
 +       * and origin first.  Most of the time they are the
 +       * same and diff-tree is fairly efficient about this.
 +       */
 +      diff_setup(&diff_opts);
 +      DIFF_OPT_SET(&diff_opts, RECURSIVE);
 +      diff_opts.detect_rename = 0;
 +      diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 +      paths[0] = origin->path;
 +      paths[1] = NULL;
 +
 +      diff_tree_setup_paths(paths, &diff_opts);
 +      if (diff_setup_done(&diff_opts) < 0)
 +              die("diff-setup");
 +
 +      if (is_null_sha1(origin->commit->object.sha1))
 +              do_diff_cache(parent->tree->object.sha1, &diff_opts);
 +      else
 +              diff_tree_sha1(parent->tree->object.sha1,
 +                             origin->commit->tree->object.sha1,
 +                             "", &diff_opts);
 +      diffcore_std(&diff_opts);
 +
 +      if (!diff_queued_diff.nr) {
 +              /* The path is the same as parent */
 +              porigin = get_origin(sb, parent, origin->path);
 +              hashcpy(porigin->blob_sha1, origin->blob_sha1);
 +      } else {
 +              /*
 +               * Since origin->path is a pathspec, if the parent
 +               * commit had it as a directory, we will see a whole
 +               * bunch of deletion of files in the directory that we
 +               * do not care about.
 +               */
 +              int i;
 +              struct diff_filepair *p = NULL;
 +              for (i = 0; i < diff_queued_diff.nr; i++) {
 +                      const char *name;
 +                      p = diff_queued_diff.queue[i];
 +                      name = p->one->path ? p->one->path : p->two->path;
 +                      if (!strcmp(name, origin->path))
 +                              break;
 +              }
 +              if (!p)
 +                      die("internal error in blame::find_origin");
 +              switch (p->status) {
 +              default:
 +                      die("internal error in blame::find_origin (%c)",
 +                          p->status);
 +              case 'M':
 +                      porigin = get_origin(sb, parent, origin->path);
 +                      hashcpy(porigin->blob_sha1, p->one->sha1);
 +                      break;
 +              case 'A':
 +              case 'T':
 +                      /* Did not exist in parent, or type changed */
 +                      break;
 +              }
 +      }
 +      diff_flush(&diff_opts);
 +      diff_tree_release_paths(&diff_opts);
 +      if (porigin) {
 +              /*
 +               * Create a freestanding copy that is not part of
 +               * the refcounted origin found in the scoreboard, and
 +               * cache it in the commit.
 +               */
 +              struct origin *cached;
 +
 +              cached = make_origin(porigin->commit, porigin->path);
 +              hashcpy(cached->blob_sha1, porigin->blob_sha1);
 +              parent->util = cached;
 +      }
 +      return porigin;
 +}
 +
 +/*
 + * We have an origin -- find the path that corresponds to it in its
 + * parent and return an origin structure to represent it.
 + */
 +static struct origin *find_rename(struct scoreboard *sb,
 +                                struct commit *parent,
 +                                struct origin *origin)
 +{
 +      struct origin *porigin = NULL;
 +      struct diff_options diff_opts;
 +      int i;
 +      const char *paths[2];
 +
 +      diff_setup(&diff_opts);
 +      DIFF_OPT_SET(&diff_opts, RECURSIVE);
 +      diff_opts.detect_rename = DIFF_DETECT_RENAME;
 +      diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 +      diff_opts.single_follow = origin->path;
 +      paths[0] = NULL;
 +      diff_tree_setup_paths(paths, &diff_opts);
 +      if (diff_setup_done(&diff_opts) < 0)
 +              die("diff-setup");
 +
 +      if (is_null_sha1(origin->commit->object.sha1))
 +              do_diff_cache(parent->tree->object.sha1, &diff_opts);
 +      else
 +              diff_tree_sha1(parent->tree->object.sha1,
 +                             origin->commit->tree->object.sha1,
 +                             "", &diff_opts);
 +      diffcore_std(&diff_opts);
 +
 +      for (i = 0; i < diff_queued_diff.nr; i++) {
 +              struct diff_filepair *p = diff_queued_diff.queue[i];
 +              if ((p->status == 'R' || p->status == 'C') &&
 +                  !strcmp(p->two->path, origin->path)) {
 +                      porigin = get_origin(sb, parent, p->one->path);
 +                      hashcpy(porigin->blob_sha1, p->one->sha1);
 +                      break;
 +              }
 +      }
 +      diff_flush(&diff_opts);
 +      diff_tree_release_paths(&diff_opts);
 +      return porigin;
 +}
 +
 +/*
 + * Link in a new blame entry to the scoreboard.  Entries that cover the
 + * same line range have been removed from the scoreboard previously.
 + */
 +static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
 +{
 +      struct blame_entry *ent, *prev = NULL;
 +
 +      origin_incref(e->suspect);
 +
 +      for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
 +              prev = ent;
 +
 +      /* prev, if not NULL, is the last one that is below e */
 +      e->prev = prev;
 +      if (prev) {
 +              e->next = prev->next;
 +              prev->next = e;
 +      }
 +      else {
 +              e->next = sb->ent;
 +              sb->ent = e;
 +      }
 +      if (e->next)
 +              e->next->prev = e;
 +}
 +
 +/*
 + * src typically is on-stack; we want to copy the information in it to
 + * a malloced blame_entry that is already on the linked list of the
 + * scoreboard.  The origin of dst loses a refcnt while the origin of src
 + * gains one.
 + */
 +static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
 +{
 +      struct blame_entry *p, *n;
 +
 +      p = dst->prev;
 +      n = dst->next;
 +      origin_incref(src->suspect);
 +      origin_decref(dst->suspect);
 +      memcpy(dst, src, sizeof(*src));
 +      dst->prev = p;
 +      dst->next = n;
 +      dst->score = 0;
 +}
 +
 +static const char *nth_line(struct scoreboard *sb, int lno)
 +{
 +      return sb->final_buf + sb->lineno[lno];
 +}
 +
 +/*
 + * It is known that lines between tlno to same came from parent, and e
 + * has an overlap with that range.  it also is known that parent's
 + * line plno corresponds to e's line tlno.
 + *
 + *                <---- e ----->
 + *                   <------>
 + *                   <------------>
 + *             <------------>
 + *             <------------------>
 + *
 + * Split e into potentially three parts; before this chunk, the chunk
 + * to be blamed for the parent, and after that portion.
 + */
 +static void split_overlap(struct blame_entry *split,
 +                        struct blame_entry *e,
 +                        int tlno, int plno, int same,
 +                        struct origin *parent)
 +{
 +      int chunk_end_lno;
 +      memset(split, 0, sizeof(struct blame_entry [3]));
 +
 +      if (e->s_lno < tlno) {
 +              /* there is a pre-chunk part not blamed on parent */
 +              split[0].suspect = origin_incref(e->suspect);
 +              split[0].lno = e->lno;
 +              split[0].s_lno = e->s_lno;
 +              split[0].num_lines = tlno - e->s_lno;
 +              split[1].lno = e->lno + tlno - e->s_lno;
 +              split[1].s_lno = plno;
 +      }
 +      else {
 +              split[1].lno = e->lno;
 +              split[1].s_lno = plno + (e->s_lno - tlno);
 +      }
 +
 +      if (same < e->s_lno + e->num_lines) {
 +              /* there is a post-chunk part not blamed on parent */
 +              split[2].suspect = origin_incref(e->suspect);
 +              split[2].lno = e->lno + (same - e->s_lno);
 +              split[2].s_lno = e->s_lno + (same - e->s_lno);
 +              split[2].num_lines = e->s_lno + e->num_lines - same;
 +              chunk_end_lno = split[2].lno;
 +      }
 +      else
 +              chunk_end_lno = e->lno + e->num_lines;
 +      split[1].num_lines = chunk_end_lno - split[1].lno;
 +
 +      /*
 +       * if it turns out there is nothing to blame the parent for,
 +       * forget about the splitting.  !split[1].suspect signals this.
 +       */
 +      if (split[1].num_lines < 1)
 +              return;
 +      split[1].suspect = origin_incref(parent);
 +}
 +
 +/*
 + * split_overlap() divided an existing blame e into up to three parts
 + * in split.  Adjust the linked list of blames in the scoreboard to
 + * reflect the split.
 + */
 +static void split_blame(struct scoreboard *sb,
 +                      struct blame_entry *split,
 +                      struct blame_entry *e)
 +{
 +      struct blame_entry *new_entry;
 +
 +      if (split[0].suspect && split[2].suspect) {
 +              /* The first part (reuse storage for the existing entry e) */
 +              dup_entry(e, &split[0]);
 +
 +              /* The last part -- me */
 +              new_entry = xmalloc(sizeof(*new_entry));
 +              memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 +              add_blame_entry(sb, new_entry);
 +
 +              /* ... and the middle part -- parent */
 +              new_entry = xmalloc(sizeof(*new_entry));
 +              memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 +              add_blame_entry(sb, new_entry);
 +      }
 +      else if (!split[0].suspect && !split[2].suspect)
 +              /*
 +               * The parent covers the entire area; reuse storage for
 +               * e and replace it with the parent.
 +               */
 +              dup_entry(e, &split[1]);
 +      else if (split[0].suspect) {
 +              /* me and then parent */
 +              dup_entry(e, &split[0]);
 +
 +              new_entry = xmalloc(sizeof(*new_entry));
 +              memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 +              add_blame_entry(sb, new_entry);
 +      }
 +      else {
 +              /* parent and then me */
 +              dup_entry(e, &split[1]);
 +
 +              new_entry = xmalloc(sizeof(*new_entry));
 +              memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 +              add_blame_entry(sb, new_entry);
 +      }
 +
 +      if (DEBUG) { /* sanity */
 +              struct blame_entry *ent;
 +              int lno = sb->ent->lno, corrupt = 0;
 +
 +              for (ent = sb->ent; ent; ent = ent->next) {
 +                      if (lno != ent->lno)
 +                              corrupt = 1;
 +                      if (ent->s_lno < 0)
 +                              corrupt = 1;
 +                      lno += ent->num_lines;
 +              }
 +              if (corrupt) {
 +                      lno = sb->ent->lno;
 +                      for (ent = sb->ent; ent; ent = ent->next) {
 +                              printf("L %8d l %8d n %8d\n",
 +                                     lno, ent->lno, ent->num_lines);
 +                              lno = ent->lno + ent->num_lines;
 +                      }
 +                      die("oops");
 +              }
 +      }
 +}
 +
 +/*
 + * After splitting the blame, the origins used by the
 + * on-stack blame_entry should lose one refcnt each.
 + */
 +static void decref_split(struct blame_entry *split)
 +{
 +      int i;
 +
 +      for (i = 0; i < 3; i++)
 +              origin_decref(split[i].suspect);
 +}
 +
 +/*
 + * Helper for blame_chunk().  blame_entry e is known to overlap with
 + * the patch hunk; split it and pass blame to the parent.
 + */
 +static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
 +                        int tlno, int plno, int same,
 +                        struct origin *parent)
 +{
 +      struct blame_entry split[3];
 +
 +      split_overlap(split, e, tlno, plno, same, parent);
 +      if (split[1].suspect)
 +              split_blame(sb, split, e);
 +      decref_split(split);
 +}
 +
 +/*
 + * Find the line number of the last line the target is suspected for.
 + */
 +static int find_last_in_target(struct scoreboard *sb, struct origin *target)
 +{
 +      struct blame_entry *e;
 +      int last_in_target = -1;
 +
 +      for (e = sb->ent; e; e = e->next) {
 +              if (e->guilty || !same_suspect(e->suspect, target))
 +                      continue;
 +              if (last_in_target < e->s_lno + e->num_lines)
 +                      last_in_target = e->s_lno + e->num_lines;
 +      }
 +      return last_in_target;
 +}
 +
 +/*
 + * Process one hunk from the patch between the current suspect for
 + * blame_entry e and its parent.  Find and split the overlap, and
 + * pass blame to the overlapping part to the parent.
 + */
 +static void blame_chunk(struct scoreboard *sb,
 +                      int tlno, int plno, int same,
 +                      struct origin *target, struct origin *parent)
 +{
 +      struct blame_entry *e;
 +
 +      for (e = sb->ent; e; e = e->next) {
 +              if (e->guilty || !same_suspect(e->suspect, target))
 +                      continue;
 +              if (same <= e->s_lno)
 +                      continue;
 +              if (tlno < e->s_lno + e->num_lines)
 +                      blame_overlap(sb, e, tlno, plno, same, parent);
 +      }
 +}
 +
 +struct blame_chunk_cb_data {
 +      struct scoreboard *sb;
 +      struct origin *target;
 +      struct origin *parent;
 +      long plno;
 +      long tlno;
 +};
 +
 +static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
 +{
 +      struct blame_chunk_cb_data *d = data;
 +      blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
 +      d->plno = p_next;
 +      d->tlno = t_next;
 +}
 +
 +/*
 + * We are looking at the origin 'target' and aiming to pass blame
 + * for the lines it is suspected to its parent.  Run diff to find
 + * which lines came from parent and pass blame for them.
 + */
 +static int pass_blame_to_parent(struct scoreboard *sb,
 +                              struct origin *target,
 +                              struct origin *parent)
 +{
 +      int last_in_target;
 +      mmfile_t file_p, file_o;
 +      struct blame_chunk_cb_data d = { sb, target, parent, 0, 0 };
 +      xpparam_t xpp;
 +      xdemitconf_t xecfg;
 +
 +      last_in_target = find_last_in_target(sb, target);
 +      if (last_in_target < 0)
 +              return 1; /* nothing remains for this target */
 +
 +      fill_origin_blob(parent, &file_p);
 +      fill_origin_blob(target, &file_o);
 +      num_get_patch++;
 +
 +      memset(&xpp, 0, sizeof(xpp));
 +      xpp.flags = xdl_opts;
 +      memset(&xecfg, 0, sizeof(xecfg));
 +      xecfg.ctxlen = 0;
 +      xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
 +      /* The rest (i.e. anything after tlno) are the same as the parent */
 +      blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
 +
 +      return 0;
 +}
 +
 +/*
 + * The lines in blame_entry after splitting blames many times can become
 + * very small and trivial, and at some point it becomes pointless to
 + * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
 + * ordinary C program, and it is not worth to say it was copied from
 + * totally unrelated file in the parent.
 + *
 + * Compute how trivial the lines in the blame_entry are.
 + */
 +static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
 +{
 +      unsigned score;
 +      const char *cp, *ep;
 +
 +      if (e->score)
 +              return e->score;
 +
 +      score = 1;
 +      cp = nth_line(sb, e->lno);
 +      ep = nth_line(sb, e->lno + e->num_lines);
 +      while (cp < ep) {
 +              unsigned ch = *((unsigned char *)cp);
 +              if (isalnum(ch))
 +                      score++;
 +              cp++;
 +      }
 +      e->score = score;
 +      return score;
 +}
 +
 +/*
 + * best_so_far[] and this[] are both a split of an existing blame_entry
 + * that passes blame to the parent.  Maintain best_so_far the best split
 + * so far, by comparing this and best_so_far and copying this into
 + * bst_so_far as needed.
 + */
 +static void copy_split_if_better(struct scoreboard *sb,
 +                               struct blame_entry *best_so_far,
 +                               struct blame_entry *this)
 +{
 +      int i;
 +
 +      if (!this[1].suspect)
 +              return;
 +      if (best_so_far[1].suspect) {
 +              if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
 +                      return;
 +      }
 +
 +      for (i = 0; i < 3; i++)
 +              origin_incref(this[i].suspect);
 +      decref_split(best_so_far);
 +      memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
 +}
 +
 +/*
 + * We are looking at a part of the final image represented by
 + * ent (tlno and same are offset by ent->s_lno).
 + * tlno is where we are looking at in the final image.
 + * up to (but not including) same match preimage.
 + * plno is where we are looking at in the preimage.
 + *
 + * <-------------- final image ---------------------->
 + *       <------ent------>
 + *         ^tlno ^same
 + *    <---------preimage----->
 + *         ^plno
 + *
 + * All line numbers are 0-based.
 + */
 +static void handle_split(struct scoreboard *sb,
 +                       struct blame_entry *ent,
 +                       int tlno, int plno, int same,
 +                       struct origin *parent,
 +                       struct blame_entry *split)
 +{
 +      if (ent->num_lines <= tlno)
 +              return;
 +      if (tlno < same) {
 +              struct blame_entry this[3];
 +              tlno += ent->s_lno;
 +              same += ent->s_lno;
 +              split_overlap(this, ent, tlno, plno, same, parent);
 +              copy_split_if_better(sb, split, this);
 +              decref_split(this);
 +      }
 +}
 +
 +struct handle_split_cb_data {
 +      struct scoreboard *sb;
 +      struct blame_entry *ent;
 +      struct origin *parent;
 +      struct blame_entry *split;
 +      long plno;
 +      long tlno;
 +};
 +
 +static void handle_split_cb(void *data, long same, long p_next, long t_next)
 +{
 +      struct handle_split_cb_data *d = data;
 +      handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
 +      d->plno = p_next;
 +      d->tlno = t_next;
 +}
 +
 +/*
 + * Find the lines from parent that are the same as ent so that
 + * we can pass blames to it.  file_p has the blob contents for
 + * the parent.
 + */
 +static void find_copy_in_blob(struct scoreboard *sb,
 +                            struct blame_entry *ent,
 +                            struct origin *parent,
 +                            struct blame_entry *split,
 +                            mmfile_t *file_p)
 +{
 +      const char *cp;
 +      int cnt;
 +      mmfile_t file_o;
 +      struct handle_split_cb_data d = { sb, ent, parent, split, 0, 0 };
 +      xpparam_t xpp;
 +      xdemitconf_t xecfg;
 +
 +      /*
 +       * Prepare mmfile that contains only the lines in ent.
 +       */
 +      cp = nth_line(sb, ent->lno);
 +      file_o.ptr = (char *) cp;
 +      cnt = ent->num_lines;
 +
 +      while (cnt && cp < sb->final_buf + sb->final_buf_size) {
 +              if (*cp++ == '\n')
 +                      cnt--;
 +      }
 +      file_o.size = cp - file_o.ptr;
 +
 +      /*
 +       * file_o is a part of final image we are annotating.
 +       * file_p partially may match that image.
 +       */
 +      memset(&xpp, 0, sizeof(xpp));
 +      xpp.flags = xdl_opts;
 +      memset(&xecfg, 0, sizeof(xecfg));
 +      xecfg.ctxlen = 1;
 +      memset(split, 0, sizeof(struct blame_entry [3]));
 +      xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
 +      /* remainder, if any, all match the preimage */
 +      handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
 +}
 +
 +/*
 + * See if lines currently target is suspected for can be attributed to
 + * parent.
 + */
 +static int find_move_in_parent(struct scoreboard *sb,
 +                             struct origin *target,
 +                             struct origin *parent)
 +{
 +      int last_in_target, made_progress;
 +      struct blame_entry *e, split[3];
 +      mmfile_t file_p;
 +
 +      last_in_target = find_last_in_target(sb, target);
 +      if (last_in_target < 0)
 +              return 1; /* nothing remains for this target */
 +
 +      fill_origin_blob(parent, &file_p);
 +      if (!file_p.ptr)
 +              return 0;
 +
 +      made_progress = 1;
 +      while (made_progress) {
 +              made_progress = 0;
 +              for (e = sb->ent; e; e = e->next) {
 +                      if (e->guilty || !same_suspect(e->suspect, target) ||
 +                          ent_score(sb, e) < blame_move_score)
 +                              continue;
 +                      find_copy_in_blob(sb, e, parent, split, &file_p);
 +                      if (split[1].suspect &&
 +                          blame_move_score < ent_score(sb, &split[1])) {
 +                              split_blame(sb, split, e);
 +                              made_progress = 1;
 +                      }
 +                      decref_split(split);
 +              }
 +      }
 +      return 0;
 +}
 +
 +struct blame_list {
 +      struct blame_entry *ent;
 +      struct blame_entry split[3];
 +};
 +
 +/*
 + * Count the number of entries the target is suspected for,
 + * and prepare a list of entry and the best split.
 + */
 +static struct blame_list *setup_blame_list(struct scoreboard *sb,
 +                                         struct origin *target,
 +                                         int min_score,
 +                                         int *num_ents_p)
 +{
 +      struct blame_entry *e;
 +      int num_ents, i;
 +      struct blame_list *blame_list = NULL;
 +
 +      for (e = sb->ent, num_ents = 0; e; e = e->next)
 +              if (!e->scanned && !e->guilty &&
 +                  same_suspect(e->suspect, target) &&
 +                  min_score < ent_score(sb, e))
 +                      num_ents++;
 +      if (num_ents) {
 +              blame_list = xcalloc(num_ents, sizeof(struct blame_list));
 +              for (e = sb->ent, i = 0; e; e = e->next)
 +                      if (!e->scanned && !e->guilty &&
 +                          same_suspect(e->suspect, target) &&
 +                          min_score < ent_score(sb, e))
 +                              blame_list[i++].ent = e;
 +      }
 +      *num_ents_p = num_ents;
 +      return blame_list;
 +}
 +
 +/*
 + * Reset the scanned status on all entries.
 + */
 +static void reset_scanned_flag(struct scoreboard *sb)
 +{
 +      struct blame_entry *e;
 +      for (e = sb->ent; e; e = e->next)
 +              e->scanned = 0;
 +}
 +
 +/*
 + * For lines target is suspected for, see if we can find code movement
 + * across file boundary from the parent commit.  porigin is the path
 + * in the parent we already tried.
 + */
 +static int find_copy_in_parent(struct scoreboard *sb,
 +                             struct origin *target,
 +                             struct commit *parent,
 +                             struct origin *porigin,
 +                             int opt)
 +{
 +      struct diff_options diff_opts;
 +      const char *paths[1];
 +      int i, j;
 +      int retval;
 +      struct blame_list *blame_list;
 +      int num_ents;
 +
 +      blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
 +      if (!blame_list)
 +              return 1; /* nothing remains for this target */
 +
 +      diff_setup(&diff_opts);
 +      DIFF_OPT_SET(&diff_opts, RECURSIVE);
 +      diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 +
 +      paths[0] = NULL;
 +      diff_tree_setup_paths(paths, &diff_opts);
 +      if (diff_setup_done(&diff_opts) < 0)
 +              die("diff-setup");
 +
 +      /* Try "find copies harder" on new path if requested;
 +       * we do not want to use diffcore_rename() actually to
 +       * match things up; find_copies_harder is set only to
 +       * force diff_tree_sha1() to feed all filepairs to diff_queue,
 +       * and this code needs to be after diff_setup_done(), which
 +       * usually makes find-copies-harder imply copy detection.
 +       */
 +      if ((opt & PICKAXE_BLAME_COPY_HARDEST)
 +          || ((opt & PICKAXE_BLAME_COPY_HARDER)
 +              && (!porigin || strcmp(target->path, porigin->path))))
 +              DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
 +
 +      if (is_null_sha1(target->commit->object.sha1))
 +              do_diff_cache(parent->tree->object.sha1, &diff_opts);
 +      else
 +              diff_tree_sha1(parent->tree->object.sha1,
 +                             target->commit->tree->object.sha1,
 +                             "", &diff_opts);
 +
 +      if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
 +              diffcore_std(&diff_opts);
 +
 +      retval = 0;
 +      while (1) {
 +              int made_progress = 0;
 +
 +              for (i = 0; i < diff_queued_diff.nr; i++) {
 +                      struct diff_filepair *p = diff_queued_diff.queue[i];
 +                      struct origin *norigin;
 +                      mmfile_t file_p;
 +                      struct blame_entry this[3];
 +
 +                      if (!DIFF_FILE_VALID(p->one))
 +                              continue; /* does not exist in parent */
 +                      if (S_ISGITLINK(p->one->mode))
 +                              continue; /* ignore git links */
 +                      if (porigin && !strcmp(p->one->path, porigin->path))
 +                              /* find_move already dealt with this path */
 +                              continue;
 +
 +                      norigin = get_origin(sb, parent, p->one->path);
 +                      hashcpy(norigin->blob_sha1, p->one->sha1);
 +                      fill_origin_blob(norigin, &file_p);
 +                      if (!file_p.ptr)
 +                              continue;
 +
 +                      for (j = 0; j < num_ents; j++) {
 +                              find_copy_in_blob(sb, blame_list[j].ent,
 +                                                norigin, this, &file_p);
 +                              copy_split_if_better(sb, blame_list[j].split,
 +                                                   this);
 +                              decref_split(this);
 +                      }
 +                      origin_decref(norigin);
 +              }
 +
 +              for (j = 0; j < num_ents; j++) {
 +                      struct blame_entry *split = blame_list[j].split;
 +                      if (split[1].suspect &&
 +                          blame_copy_score < ent_score(sb, &split[1])) {
 +                              split_blame(sb, split, blame_list[j].ent);
 +                              made_progress = 1;
 +                      }
 +                      else
 +                              blame_list[j].ent->scanned = 1;
 +                      decref_split(split);
 +              }
 +              free(blame_list);
 +
 +              if (!made_progress)
 +                      break;
 +              blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
 +              if (!blame_list) {
 +                      retval = 1;
 +                      break;
 +              }
 +      }
 +      reset_scanned_flag(sb);
 +      diff_flush(&diff_opts);
 +      diff_tree_release_paths(&diff_opts);
 +      return retval;
 +}
 +
 +/*
 + * The blobs of origin and porigin exactly match, so everything
 + * origin is suspected for can be blamed on the parent.
 + */
 +static void pass_whole_blame(struct scoreboard *sb,
 +                           struct origin *origin, struct origin *porigin)
 +{
 +      struct blame_entry *e;
 +
 +      if (!porigin->file.ptr && origin->file.ptr) {
 +              /* Steal its file */
 +              porigin->file = origin->file;
 +              origin->file.ptr = NULL;
 +      }
 +      for (e = sb->ent; e; e = e->next) {
 +              if (!same_suspect(e->suspect, origin))
 +                      continue;
 +              origin_incref(porigin);
 +              origin_decref(e->suspect);
 +              e->suspect = porigin;
 +      }
 +}
 +
 +/*
 + * We pass blame from the current commit to its parents.  We keep saying
 + * "parent" (and "porigin"), but what we mean is to find scapegoat to
 + * exonerate ourselves.
 + */
 +static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
 +{
 +      if (!reverse)
 +              return commit->parents;
 +      return lookup_decoration(&revs->children, &commit->object);
 +}
 +
 +static int num_scapegoats(struct rev_info *revs, struct commit *commit)
 +{
 +      int cnt;
 +      struct commit_list *l = first_scapegoat(revs, commit);
 +      for (cnt = 0; l; l = l->next)
 +              cnt++;
 +      return cnt;
 +}
 +
 +#define MAXSG 16
 +
 +static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
 +{
 +      struct rev_info *revs = sb->revs;
 +      int i, pass, num_sg;
 +      struct commit *commit = origin->commit;
 +      struct commit_list *sg;
 +      struct origin *sg_buf[MAXSG];
 +      struct origin *porigin, **sg_origin = sg_buf;
 +
 +      num_sg = num_scapegoats(revs, commit);
 +      if (!num_sg)
 +              goto finish;
 +      else if (num_sg < ARRAY_SIZE(sg_buf))
 +              memset(sg_buf, 0, sizeof(sg_buf));
 +      else
 +              sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
 +
 +      /*
 +       * The first pass looks for unrenamed path to optimize for
 +       * common cases, then we look for renames in the second pass.
 +       */
 +      for (pass = 0; pass < 2; pass++) {
 +              struct origin *(*find)(struct scoreboard *,
 +                                     struct commit *, struct origin *);
 +              find = pass ? find_rename : find_origin;
 +
 +              for (i = 0, sg = first_scapegoat(revs, commit);
 +                   i < num_sg && sg;
 +                   sg = sg->next, i++) {
 +                      struct commit *p = sg->item;
 +                      int j, same;
 +
 +                      if (sg_origin[i])
 +                              continue;
 +                      if (parse_commit(p))
 +                              continue;
 +                      porigin = find(sb, p, origin);
 +                      if (!porigin)
 +                              continue;
 +                      if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
 +                              pass_whole_blame(sb, origin, porigin);
 +                              origin_decref(porigin);
 +                              goto finish;
 +                      }
 +                      for (j = same = 0; j < i; j++)
 +                              if (sg_origin[j] &&
 +                                  !hashcmp(sg_origin[j]->blob_sha1,
 +                                           porigin->blob_sha1)) {
 +                                      same = 1;
 +                                      break;
 +                              }
 +                      if (!same)
 +                              sg_origin[i] = porigin;
 +                      else
 +                              origin_decref(porigin);
 +              }
 +      }
 +
 +      num_commits++;
 +      for (i = 0, sg = first_scapegoat(revs, commit);
 +           i < num_sg && sg;
 +           sg = sg->next, i++) {
 +              struct origin *porigin = sg_origin[i];
 +              if (!porigin)
 +                      continue;
 +              if (!origin->previous) {
 +                      origin_incref(porigin);
 +                      origin->previous = porigin;
 +              }
 +              if (pass_blame_to_parent(sb, origin, porigin))
 +                      goto finish;
 +      }
 +
 +      /*
 +       * Optionally find moves in parents' files.
 +       */
 +      if (opt & PICKAXE_BLAME_MOVE)
 +              for (i = 0, sg = first_scapegoat(revs, commit);
 +                   i < num_sg && sg;
 +                   sg = sg->next, i++) {
 +                      struct origin *porigin = sg_origin[i];
 +                      if (!porigin)
 +                              continue;
 +                      if (find_move_in_parent(sb, origin, porigin))
 +                              goto finish;
 +              }
 +
 +      /*
 +       * Optionally find copies from parents' files.
 +       */
 +      if (opt & PICKAXE_BLAME_COPY)
 +              for (i = 0, sg = first_scapegoat(revs, commit);
 +                   i < num_sg && sg;
 +                   sg = sg->next, i++) {
 +                      struct origin *porigin = sg_origin[i];
 +                      if (find_copy_in_parent(sb, origin, sg->item,
 +                                              porigin, opt))
 +                              goto finish;
 +              }
 +
 + finish:
 +      for (i = 0; i < num_sg; i++) {
 +              if (sg_origin[i]) {
 +                      drop_origin_blob(sg_origin[i]);
 +                      origin_decref(sg_origin[i]);
 +              }
 +      }
 +      drop_origin_blob(origin);
 +      if (sg_buf != sg_origin)
 +              free(sg_origin);
 +}
 +
 +/*
 + * Information on commits, used for output.
 + */
 +struct commit_info
 +{
 +      const char *author;
 +      const char *author_mail;
 +      unsigned long author_time;
 +      const char *author_tz;
 +
 +      /* filled only when asked for details */
 +      const char *committer;
 +      const char *committer_mail;
 +      unsigned long committer_time;
 +      const char *committer_tz;
 +
 +      const char *summary;
 +};
 +
 +/*
 + * Parse author/committer line in the commit object buffer
 + */
 +static void get_ac_line(const char *inbuf, const char *what,
 +                      int person_len, char *person,
 +                      int mail_len, char *mail,
 +                      unsigned long *time, const char **tz)
 +{
 +      int len, tzlen, maillen;
 +      char *tmp, *endp, *timepos, *mailpos;
 +
 +      tmp = strstr(inbuf, what);
 +      if (!tmp)
 +              goto error_out;
 +      tmp += strlen(what);
 +      endp = strchr(tmp, '\n');
 +      if (!endp)
 +              len = strlen(tmp);
 +      else
 +              len = endp - tmp;
 +      if (person_len <= len) {
 +      error_out:
 +              /* Ugh */
 +              *tz = "(unknown)";
 +              strcpy(person, *tz);
 +              strcpy(mail, *tz);
 +              *time = 0;
 +              return;
 +      }
 +      memcpy(person, tmp, len);
 +
 +      tmp = person;
 +      tmp += len;
 +      *tmp = 0;
 +      while (person < tmp && *tmp != ' ')
 +              tmp--;
 +      if (tmp <= person)
 +              goto error_out;
 +      *tz = tmp+1;
 +      tzlen = (person+len)-(tmp+1);
 +
 +      *tmp = 0;
 +      while (person < tmp && *tmp != ' ')
 +              tmp--;
 +      if (tmp <= person)
 +              goto error_out;
 +      *time = strtoul(tmp, NULL, 10);
 +      timepos = tmp;
 +
 +      *tmp = 0;
 +      while (person < tmp && *tmp != ' ')
 +              tmp--;
 +      if (tmp <= person)
 +              return;
 +      mailpos = tmp + 1;
 +      *tmp = 0;
 +      maillen = timepos - tmp;
 +      memcpy(mail, mailpos, maillen);
 +
 +      if (!mailmap.nr)
 +              return;
 +
 +      /*
 +       * mailmap expansion may make the name longer.
 +       * make room by pushing stuff down.
 +       */
 +      tmp = person + person_len - (tzlen + 1);
 +      memmove(tmp, *tz, tzlen);
 +      tmp[tzlen] = 0;
 +      *tz = tmp;
 +
 +      /*
 +       * Now, convert both name and e-mail using mailmap
 +       */
 +      if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
 +              /* Add a trailing '>' to email, since map_user returns plain emails
 +                 Note: It already has '<', since we replace from mail+1 */
 +              mailpos = memchr(mail, '\0', mail_len);
 +              if (mailpos && mailpos-mail < mail_len - 1) {
 +                      *mailpos = '>';
 +                      *(mailpos+1) = '\0';
 +              }
 +      }
 +}
 +
 +static void get_commit_info(struct commit *commit,
 +                          struct commit_info *ret,
 +                          int detailed)
 +{
 +      int len;
 +      char *tmp, *endp, *reencoded, *message;
 +      static char author_name[1024];
 +      static char author_mail[1024];
 +      static char committer_name[1024];
 +      static char committer_mail[1024];
 +      static char summary_buf[1024];
 +
 +      /*
 +       * We've operated without save_commit_buffer, so
 +       * we now need to populate them for output.
 +       */
 +      if (!commit->buffer) {
 +              enum object_type type;
 +              unsigned long size;
 +              commit->buffer =
 +                      read_sha1_file(commit->object.sha1, &type, &size);
 +              if (!commit->buffer)
 +                      die("Cannot read commit %s",
 +                          sha1_to_hex(commit->object.sha1));
 +      }
 +      reencoded = reencode_commit_message(commit, NULL);
 +      message   = reencoded ? reencoded : commit->buffer;
 +      ret->author = author_name;
 +      ret->author_mail = author_mail;
 +      get_ac_line(message, "\nauthor ",
 +                  sizeof(author_name), author_name,
 +                  sizeof(author_mail), author_mail,
 +                  &ret->author_time, &ret->author_tz);
 +
 +      if (!detailed) {
 +              free(reencoded);
 +              return;
 +      }
 +
 +      ret->committer = committer_name;
 +      ret->committer_mail = committer_mail;
 +      get_ac_line(message, "\ncommitter ",
 +                  sizeof(committer_name), committer_name,
 +                  sizeof(committer_mail), committer_mail,
 +                  &ret->committer_time, &ret->committer_tz);
 +
 +      ret->summary = summary_buf;
 +      tmp = strstr(message, "\n\n");
 +      if (!tmp) {
 +      error_out:
 +              sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
 +              free(reencoded);
 +              return;
 +      }
 +      tmp += 2;
 +      endp = strchr(tmp, '\n');
 +      if (!endp)
 +              endp = tmp + strlen(tmp);
 +      len = endp - tmp;
 +      if (len >= sizeof(summary_buf) || len == 0)
 +              goto error_out;
 +      memcpy(summary_buf, tmp, len);
 +      summary_buf[len] = 0;
 +      free(reencoded);
 +}
 +
 +/*
 + * To allow LF and other nonportable characters in pathnames,
 + * they are c-style quoted as needed.
 + */
 +static void write_filename_info(const char *path)
 +{
 +      printf("filename ");
 +      write_name_quoted(path, stdout, '\n');
 +}
 +
 +/*
 + * Porcelain/Incremental format wants to show a lot of details per
 + * commit.  Instead of repeating this every line, emit it only once,
 + * the first time each commit appears in the output.
 + */
 +static int emit_one_suspect_detail(struct origin *suspect)
 +{
 +      struct commit_info ci;
 +
 +      if (suspect->commit->object.flags & METAINFO_SHOWN)
 +              return 0;
 +
 +      suspect->commit->object.flags |= METAINFO_SHOWN;
 +      get_commit_info(suspect->commit, &ci, 1);
 +      printf("author %s\n", ci.author);
 +      printf("author-mail %s\n", ci.author_mail);
 +      printf("author-time %lu\n", ci.author_time);
 +      printf("author-tz %s\n", ci.author_tz);
 +      printf("committer %s\n", ci.committer);
 +      printf("committer-mail %s\n", ci.committer_mail);
 +      printf("committer-time %lu\n", ci.committer_time);
 +      printf("committer-tz %s\n", ci.committer_tz);
 +      printf("summary %s\n", ci.summary);
 +      if (suspect->commit->object.flags & UNINTERESTING)
 +              printf("boundary\n");
 +      if (suspect->previous) {
 +              struct origin *prev = suspect->previous;
 +              printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
 +              write_name_quoted(prev->path, stdout, '\n');
 +      }
 +      return 1;
 +}
 +
 +/*
 + * The blame_entry is found to be guilty for the range.  Mark it
 + * as such, and show it in incremental output.
 + */
 +static void found_guilty_entry(struct blame_entry *ent)
 +{
 +      if (ent->guilty)
 +              return;
 +      ent->guilty = 1;
 +      if (incremental) {
 +              struct origin *suspect = ent->suspect;
 +
 +              printf("%s %d %d %d\n",
 +                     sha1_to_hex(suspect->commit->object.sha1),
 +                     ent->s_lno + 1, ent->lno + 1, ent->num_lines);
 +              emit_one_suspect_detail(suspect);
 +              write_filename_info(suspect->path);
 +              maybe_flush_or_die(stdout, "stdout");
 +      }
 +}
 +
 +/*
 + * The main loop -- while the scoreboard has lines whose true origin
 + * is still unknown, pick one blame_entry, and allow its current
 + * suspect to pass blames to its parents.
 + */
 +static void assign_blame(struct scoreboard *sb, int opt)
 +{
 +      struct rev_info *revs = sb->revs;
 +
 +      while (1) {
 +              struct blame_entry *ent;
 +              struct commit *commit;
 +              struct origin *suspect = NULL;
 +
 +              /* find one suspect to break down */
 +              for (ent = sb->ent; !suspect && ent; ent = ent->next)
 +                      if (!ent->guilty)
 +                              suspect = ent->suspect;
 +              if (!suspect)
 +                      return; /* all done */
 +
 +              /*
 +               * We will use this suspect later in the loop,
 +               * so hold onto it in the meantime.
 +               */
 +              origin_incref(suspect);
 +              commit = suspect->commit;
 +              if (!commit->object.parsed)
 +                      parse_commit(commit);
 +              if (reverse ||
 +                  (!(commit->object.flags & UNINTERESTING) &&
 +                   !(revs->max_age != -1 && commit->date < revs->max_age)))
 +                      pass_blame(sb, suspect, opt);
 +              else {
 +                      commit->object.flags |= UNINTERESTING;
 +                      if (commit->object.parsed)
 +                              mark_parents_uninteresting(commit);
 +              }
 +              /* treat root commit as boundary */
 +              if (!commit->parents && !show_root)
 +                      commit->object.flags |= UNINTERESTING;
 +
 +              /* Take responsibility for the remaining entries */
 +              for (ent = sb->ent; ent; ent = ent->next)
 +                      if (same_suspect(ent->suspect, suspect))
 +                              found_guilty_entry(ent);
 +              origin_decref(suspect);
 +
 +              if (DEBUG) /* sanity */
 +                      sanity_check_refcnt(sb);
 +      }
 +}
 +
 +static const char *format_time(unsigned long time, const char *tz_str,
 +                             int show_raw_time)
 +{
 +      static char time_buf[128];
 +      const char *time_str;
 +      int time_len;
 +      int tz;
 +
 +      if (show_raw_time) {
 +              sprintf(time_buf, "%lu %s", time, tz_str);
 +      }
 +      else {
 +              tz = atoi(tz_str);
 +              time_str = show_date(time, tz, blame_date_mode);
 +              time_len = strlen(time_str);
 +              memcpy(time_buf, time_str, time_len);
 +              memset(time_buf + time_len, ' ', blame_date_width - time_len);
 +      }
 +      return time_buf;
 +}
 +
 +#define OUTPUT_ANNOTATE_COMPAT        001
 +#define OUTPUT_LONG_OBJECT_NAME       002
 +#define OUTPUT_RAW_TIMESTAMP  004
 +#define OUTPUT_PORCELAIN      010
 +#define OUTPUT_SHOW_NAME      020
 +#define OUTPUT_SHOW_NUMBER    040
 +#define OUTPUT_SHOW_SCORE      0100
 +#define OUTPUT_NO_AUTHOR       0200
 +
 +static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
 +{
 +      int cnt;
 +      const char *cp;
 +      struct origin *suspect = ent->suspect;
 +      char hex[41];
 +
 +      strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
 +      printf("%s%c%d %d %d\n",
 +             hex,
 +             ent->guilty ? ' ' : '*', /* purely for debugging */
 +             ent->s_lno + 1,
 +             ent->lno + 1,
 +             ent->num_lines);
 +      if (emit_one_suspect_detail(suspect) ||
 +          (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
 +              write_filename_info(suspect->path);
 +
 +      cp = nth_line(sb, ent->lno);
 +      for (cnt = 0; cnt < ent->num_lines; cnt++) {
 +              char ch;
 +              if (cnt)
 +                      printf("%s %d %d\n", hex,
 +                             ent->s_lno + 1 + cnt,
 +                             ent->lno + 1 + cnt);
 +              putchar('\t');
 +              do {
 +                      ch = *cp++;
 +                      putchar(ch);
 +              } while (ch != '\n' &&
 +                       cp < sb->final_buf + sb->final_buf_size);
 +      }
 +
 +      if (sb->final_buf_size && cp[-1] != '\n')
 +              putchar('\n');
 +}
 +
 +static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
 +{
 +      int cnt;
 +      const char *cp;
 +      struct origin *suspect = ent->suspect;
 +      struct commit_info ci;
 +      char hex[41];
 +      int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
 +
 +      get_commit_info(suspect->commit, &ci, 1);
 +      strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
 +
 +      cp = nth_line(sb, ent->lno);
 +      for (cnt = 0; cnt < ent->num_lines; cnt++) {
 +              char ch;
 +              int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
 +
 +              if (suspect->commit->object.flags & UNINTERESTING) {
 +                      if (blank_boundary)
 +                              memset(hex, ' ', length);
 +                      else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
 +                              length--;
 +                              putchar('^');
 +                      }
 +              }
 +
 +              printf("%.*s", length, hex);
 +              if (opt & OUTPUT_ANNOTATE_COMPAT)
 +                      printf("\t(%10s\t%10s\t%d)", ci.author,
 +                             format_time(ci.author_time, ci.author_tz,
 +                                         show_raw_time),
 +                             ent->lno + 1 + cnt);
 +              else {
 +                      if (opt & OUTPUT_SHOW_SCORE)
 +                              printf(" %*d %02d",
 +                                     max_score_digits, ent->score,
 +                                     ent->suspect->refcnt);
 +                      if (opt & OUTPUT_SHOW_NAME)
 +                              printf(" %-*.*s", longest_file, longest_file,
 +                                     suspect->path);
 +                      if (opt & OUTPUT_SHOW_NUMBER)
 +                              printf(" %*d", max_orig_digits,
 +                                     ent->s_lno + 1 + cnt);
 +
 +                      if (!(opt & OUTPUT_NO_AUTHOR)) {
 +                              int pad = longest_author - utf8_strwidth(ci.author);
 +                              printf(" (%s%*s %10s",
 +                                     ci.author, pad, "",
 +                                     format_time(ci.author_time,
 +                                                 ci.author_tz,
 +                                                 show_raw_time));
 +                      }
 +                      printf(" %*d) ",
 +                             max_digits, ent->lno + 1 + cnt);
 +              }
 +              do {
 +                      ch = *cp++;
 +                      putchar(ch);
 +              } while (ch != '\n' &&
 +                       cp < sb->final_buf + sb->final_buf_size);
 +      }
 +
 +      if (sb->final_buf_size && cp[-1] != '\n')
 +              putchar('\n');
 +}
 +
 +static void output(struct scoreboard *sb, int option)
 +{
 +      struct blame_entry *ent;
 +
 +      if (option & OUTPUT_PORCELAIN) {
 +              for (ent = sb->ent; ent; ent = ent->next) {
 +                      struct blame_entry *oth;
 +                      struct origin *suspect = ent->suspect;
 +                      struct commit *commit = suspect->commit;
 +                      if (commit->object.flags & MORE_THAN_ONE_PATH)
 +                              continue;
 +                      for (oth = ent->next; oth; oth = oth->next) {
 +                              if ((oth->suspect->commit != commit) ||
 +                                  !strcmp(oth->suspect->path, suspect->path))
 +                                      continue;
 +                              commit->object.flags |= MORE_THAN_ONE_PATH;
 +                              break;
 +                      }
 +              }
 +      }
 +
 +      for (ent = sb->ent; ent; ent = ent->next) {
 +              if (option & OUTPUT_PORCELAIN)
 +                      emit_porcelain(sb, ent);
 +              else {
 +                      emit_other(sb, ent, option);
 +              }
 +      }
 +}
 +
 +/*
 + * To allow quick access to the contents of nth line in the
 + * final image, prepare an index in the scoreboard.
 + */
 +static int prepare_lines(struct scoreboard *sb)
 +{
 +      const char *buf = sb->final_buf;
 +      unsigned long len = sb->final_buf_size;
 +      int num = 0, incomplete = 0, bol = 1;
 +
 +      if (len && buf[len-1] != '\n')
 +              incomplete++; /* incomplete line at the end */
 +      while (len--) {
 +              if (bol) {
 +                      sb->lineno = xrealloc(sb->lineno,
 +                                            sizeof(int *) * (num + 1));
 +                      sb->lineno[num] = buf - sb->final_buf;
 +                      bol = 0;
 +              }
 +              if (*buf++ == '\n') {
 +                      num++;
 +                      bol = 1;
 +              }
 +      }
 +      sb->lineno = xrealloc(sb->lineno,
 +                            sizeof(int *) * (num + incomplete + 1));
 +      sb->lineno[num + incomplete] = buf - sb->final_buf;
 +      sb->num_lines = num + incomplete;
 +      return sb->num_lines;
 +}
 +
 +/*
 + * Add phony grafts for use with -S; this is primarily to
 + * support git's cvsserver that wants to give a linear history
 + * to its clients.
 + */
 +static int read_ancestry(const char *graft_file)
 +{
 +      FILE *fp = fopen(graft_file, "r");
 +      char buf[1024];
 +      if (!fp)
 +              return -1;
 +      while (fgets(buf, sizeof(buf), fp)) {
 +              /* The format is just "Commit Parent1 Parent2 ...\n" */
 +              int len = strlen(buf);
 +              struct commit_graft *graft = read_graft_line(buf, len);
 +              if (graft)
 +                      register_commit_graft(graft, 0);
 +      }
 +      fclose(fp);
 +      return 0;
 +}
 +
 +/*
 + * How many columns do we need to show line numbers in decimal?
 + */
 +static int lineno_width(int lines)
 +{
 +      int i, width;
 +
 +      for (width = 1, i = 10; i <= lines; width++)
 +              i *= 10;
 +      return width;
 +}
 +
 +/*
 + * How many columns do we need to show line numbers, authors,
 + * and filenames?
 + */
 +static void find_alignment(struct scoreboard *sb, int *option)
 +{
 +      int longest_src_lines = 0;
 +      int longest_dst_lines = 0;
 +      unsigned largest_score = 0;
 +      struct blame_entry *e;
 +
 +      for (e = sb->ent; e; e = e->next) {
 +              struct origin *suspect = e->suspect;
 +              struct commit_info ci;
 +              int num;
 +
 +              if (strcmp(suspect->path, sb->path))
 +                      *option |= OUTPUT_SHOW_NAME;
 +              num = strlen(suspect->path);
 +              if (longest_file < num)
 +                      longest_file = num;
 +              if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
 +                      suspect->commit->object.flags |= METAINFO_SHOWN;
 +                      get_commit_info(suspect->commit, &ci, 1);
 +                      num = utf8_strwidth(ci.author);
 +                      if (longest_author < num)
 +                              longest_author = num;
 +              }
 +              num = e->s_lno + e->num_lines;
 +              if (longest_src_lines < num)
 +                      longest_src_lines = num;
 +              num = e->lno + e->num_lines;
 +              if (longest_dst_lines < num)
 +                      longest_dst_lines = num;
 +              if (largest_score < ent_score(sb, e))
 +                      largest_score = ent_score(sb, e);
 +      }
 +      max_orig_digits = lineno_width(longest_src_lines);
 +      max_digits = lineno_width(longest_dst_lines);
 +      max_score_digits = lineno_width(largest_score);
 +}
 +
 +/*
 + * For debugging -- origin is refcounted, and this asserts that
 + * we do not underflow.
 + */
 +static void sanity_check_refcnt(struct scoreboard *sb)
 +{
 +      int baa = 0;
 +      struct blame_entry *ent;
 +
 +      for (ent = sb->ent; ent; ent = ent->next) {
 +              /* Nobody should have zero or negative refcnt */
 +              if (ent->suspect->refcnt <= 0) {
 +                      fprintf(stderr, "%s in %s has negative refcnt %d\n",
 +                              ent->suspect->path,
 +                              sha1_to_hex(ent->suspect->commit->object.sha1),
 +                              ent->suspect->refcnt);
 +                      baa = 1;
 +              }
 +      }
 +      if (baa) {
 +              int opt = 0160;
 +              find_alignment(sb, &opt);
 +              output(sb, opt);
 +              die("Baa %d!", baa);
 +      }
 +}
 +
 +/*
 + * Used for the command line parsing; check if the path exists
 + * in the working tree.
 + */
 +static int has_string_in_work_tree(const char *path)
 +{
 +      struct stat st;
 +      return !lstat(path, &st);
 +}
 +
 +static unsigned parse_score(const char *arg)
 +{
 +      char *end;
 +      unsigned long score = strtoul(arg, &end, 10);
 +      if (*end)
 +              return 0;
 +      return score;
 +}
 +
 +static const char *add_prefix(const char *prefix, const char *path)
 +{
 +      return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
 +}
 +
 +/*
 + * Parsing of (comma separated) one item in the -L option
 + */
 +static const char *parse_loc(const char *spec,
 +                           struct scoreboard *sb, long lno,
 +                           long begin, long *ret)
 +{
 +      char *term;
 +      const char *line;
 +      long num;
 +      int reg_error;
 +      regex_t regexp;
 +      regmatch_t match[1];
 +
 +      /* Allow "-L <something>,+20" to mean starting at <something>
 +       * for 20 lines, or "-L <something>,-5" for 5 lines ending at
 +       * <something>.
 +       */
 +      if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
 +              num = strtol(spec + 1, &term, 10);
 +              if (term != spec + 1) {
 +                      if (spec[0] == '-')
 +                              num = 0 - num;
 +                      if (0 < num)
 +                              *ret = begin + num - 2;
 +                      else if (!num)
 +                              *ret = begin;
 +                      else
 +                              *ret = begin + num;
 +                      return term;
 +              }
 +              return spec;
 +      }
 +      num = strtol(spec, &term, 10);
 +      if (term != spec) {
 +              *ret = num;
 +              return term;
 +      }
 +      if (spec[0] != '/')
 +              return spec;
 +
 +      /* it could be a regexp of form /.../ */
 +      for (term = (char *) spec + 1; *term && *term != '/'; term++) {
 +              if (*term == '\\')
 +                      term++;
 +      }
 +      if (*term != '/')
 +              return spec;
 +
 +      /* try [spec+1 .. term-1] as regexp */
 +      *term = 0;
 +      begin--; /* input is in human terms */
 +      line = nth_line(sb, begin);
 +
 +      if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
 +          !(reg_error = regexec(&regexp, line, 1, match, 0))) {
 +              const char *cp = line + match[0].rm_so;
 +              const char *nline;
 +
 +              while (begin++ < lno) {
 +                      nline = nth_line(sb, begin);
 +                      if (line <= cp && cp < nline)
 +                              break;
 +                      line = nline;
 +              }
 +              *ret = begin;
 +              regfree(&regexp);
 +              *term++ = '/';
 +              return term;
 +      }
 +      else {
 +              char errbuf[1024];
 +              regerror(reg_error, &regexp, errbuf, 1024);
 +              die("-L parameter '%s': %s", spec + 1, errbuf);
 +      }
 +}
 +
 +/*
 + * Parsing of -L option
 + */
 +static void prepare_blame_range(struct scoreboard *sb,
 +                              const char *bottomtop,
 +                              long lno,
 +                              long *bottom, long *top)
 +{
 +      const char *term;
 +
 +      term = parse_loc(bottomtop, sb, lno, 1, bottom);
 +      if (*term == ',') {
 +              term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
 +              if (*term)
 +                      usage(blame_usage);
 +      }
 +      if (*term)
 +              usage(blame_usage);
 +}
 +
 +static int git_blame_config(const char *var, const char *value, void *cb)
 +{
 +      if (!strcmp(var, "blame.showroot")) {
 +              show_root = git_config_bool(var, value);
 +              return 0;
 +      }
 +      if (!strcmp(var, "blame.blankboundary")) {
 +              blank_boundary = git_config_bool(var, value);
 +              return 0;
 +      }
 +      if (!strcmp(var, "blame.date")) {
 +              if (!value)
 +                      return config_error_nonbool(var);
 +              blame_date_mode = parse_date_format(value);
 +              return 0;
 +      }
 +      return git_default_config(var, value, cb);
 +}
 +
 +/*
 + * Prepare a dummy commit that represents the work tree (or staged) item.
 + * Note that annotating work tree item never works in the reverse.
 + */
 +static struct commit *fake_working_tree_commit(const char *path, const char *contents_from)
 +{
 +      struct commit *commit;
 +      struct origin *origin;
 +      unsigned char head_sha1[20];
 +      struct strbuf buf = STRBUF_INIT;
 +      const char *ident;
 +      time_t now;
 +      int size, len;
 +      struct cache_entry *ce;
 +      unsigned mode;
 +
 +      if (get_sha1("HEAD", head_sha1))
 +              die("No such ref: HEAD");
 +
 +      time(&now);
 +      commit = xcalloc(1, sizeof(*commit));
 +      commit->parents = xcalloc(1, sizeof(*commit->parents));
 +      commit->parents->item = lookup_commit_reference(head_sha1);
 +      commit->object.parsed = 1;
 +      commit->date = now;
 +      commit->object.type = OBJ_COMMIT;
 +
 +      origin = make_origin(commit, path);
 +
 +      if (!contents_from || strcmp("-", contents_from)) {
 +              struct stat st;
 +              const char *read_from;
 +
 +              if (contents_from) {
 +                      if (stat(contents_from, &st) < 0)
 +                              die_errno("Cannot stat '%s'", contents_from);
 +                      read_from = contents_from;
 +              }
 +              else {
 +                      if (lstat(path, &st) < 0)
 +                              die_errno("Cannot lstat '%s'", path);
 +                      read_from = path;
 +              }
 +              mode = canon_mode(st.st_mode);
 +              switch (st.st_mode & S_IFMT) {
 +              case S_IFREG:
 +                      if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
 +                              die_errno("cannot open or read '%s'", read_from);
 +                      break;
 +              case S_IFLNK:
 +                      if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
 +                              die_errno("cannot readlink '%s'", read_from);
 +                      break;
 +              default:
 +                      die("unsupported file type %s", read_from);
 +              }
 +      }
 +      else {
 +              /* Reading from stdin */
 +              contents_from = "standard input";
 +              mode = 0;
 +              if (strbuf_read(&buf, 0, 0) < 0)
 +                      die_errno("failed to read from stdin");
 +      }
 +      convert_to_git(path, buf.buf, buf.len, &buf, 0);
 +      origin->file.ptr = buf.buf;
 +      origin->file.size = buf.len;
 +      pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
 +      commit->util = origin;
 +
 +      /*
 +       * Read the current index, replace the path entry with
 +       * origin->blob_sha1 without mucking with its mode or type
 +       * bits; we are not going to write this index out -- we just
 +       * want to run "diff-index --cached".
 +       */
 +      discard_cache();
 +      read_cache();
 +
 +      len = strlen(path);
 +      if (!mode) {
 +              int pos = cache_name_pos(path, len);
 +              if (0 <= pos)
 +                      mode = active_cache[pos]->ce_mode;
 +              else
 +                      /* Let's not bother reading from HEAD tree */
 +                      mode = S_IFREG | 0644;
 +      }
 +      size = cache_entry_size(len);
 +      ce = xcalloc(1, size);
 +      hashcpy(ce->sha1, origin->blob_sha1);
 +      memcpy(ce->name, path, len);
 +      ce->ce_flags = create_ce_flags(len, 0);
 +      ce->ce_mode = create_ce_mode(mode);
 +      add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
 +
 +      /*
 +       * We are not going to write this out, so this does not matter
 +       * right now, but someday we might optimize diff-index --cached
 +       * with cache-tree information.
 +       */
 +      cache_tree_invalidate_path(active_cache_tree, path);
 +
 +      commit->buffer = xmalloc(400);
 +      ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
 +      snprintf(commit->buffer, 400,
 +              "tree 0000000000000000000000000000000000000000\n"
 +              "parent %s\n"
 +              "author %s\n"
 +              "committer %s\n\n"
 +              "Version of %s from %s\n",
 +              sha1_to_hex(head_sha1),
 +              ident, ident, path, contents_from ? contents_from : path);
 +      return commit;
 +}
 +
 +static const char *prepare_final(struct scoreboard *sb)
 +{
 +      int i;
 +      const char *final_commit_name = NULL;
 +      struct rev_info *revs = sb->revs;
 +
 +      /*
 +       * There must be one and only one positive commit in the
 +       * revs->pending array.
 +       */
 +      for (i = 0; i < revs->pending.nr; i++) {
 +              struct object *obj = revs->pending.objects[i].item;
 +              if (obj->flags & UNINTERESTING)
 +                      continue;
 +              while (obj->type == OBJ_TAG)
 +                      obj = deref_tag(obj, NULL, 0);
 +              if (obj->type != OBJ_COMMIT)
 +                      die("Non commit %s?", revs->pending.objects[i].name);
 +              if (sb->final)
 +                      die("More than one commit to dig from %s and %s?",
 +                          revs->pending.objects[i].name,
 +                          final_commit_name);
 +              sb->final = (struct commit *) obj;
 +              final_commit_name = revs->pending.objects[i].name;
 +      }
 +      return final_commit_name;
 +}
 +
 +static const char *prepare_initial(struct scoreboard *sb)
 +{
 +      int i;
 +      const char *final_commit_name = NULL;
 +      struct rev_info *revs = sb->revs;
 +
 +      /*
 +       * There must be one and only one negative commit, and it must be
 +       * the boundary.
 +       */
 +      for (i = 0; i < revs->pending.nr; i++) {
 +              struct object *obj = revs->pending.objects[i].item;
 +              if (!(obj->flags & UNINTERESTING))
 +                      continue;
 +              while (obj->type == OBJ_TAG)
 +                      obj = deref_tag(obj, NULL, 0);
 +              if (obj->type != OBJ_COMMIT)
 +                      die("Non commit %s?", revs->pending.objects[i].name);
 +              if (sb->final)
 +                      die("More than one commit to dig down to %s and %s?",
 +                          revs->pending.objects[i].name,
 +                          final_commit_name);
 +              sb->final = (struct commit *) obj;
 +              final_commit_name = revs->pending.objects[i].name;
 +      }
 +      if (!final_commit_name)
 +              die("No commit to dig down to?");
 +      return final_commit_name;
 +}
 +
 +static int blame_copy_callback(const struct option *option, const char *arg, int unset)
 +{
 +      int *opt = option->value;
 +
 +      /*
 +       * -C enables copy from removed files;
 +       * -C -C enables copy from existing files, but only
 +       *       when blaming a new file;
 +       * -C -C -C enables copy from existing files for
 +       *          everybody
 +       */
 +      if (*opt & PICKAXE_BLAME_COPY_HARDER)
 +              *opt |= PICKAXE_BLAME_COPY_HARDEST;
 +      if (*opt & PICKAXE_BLAME_COPY)
 +              *opt |= PICKAXE_BLAME_COPY_HARDER;
 +      *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
 +
 +      if (arg)
 +              blame_copy_score = parse_score(arg);
 +      return 0;
 +}
 +
 +static int blame_move_callback(const struct option *option, const char *arg, int unset)
 +{
 +      int *opt = option->value;
 +
 +      *opt |= PICKAXE_BLAME_MOVE;
 +
 +      if (arg)
 +              blame_move_score = parse_score(arg);
 +      return 0;
 +}
 +
 +static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
 +{
 +      const char **bottomtop = option->value;
 +      if (!arg)
 +              return -1;
 +      if (*bottomtop)
 +              die("More than one '-L n,m' option given");
 +      *bottomtop = arg;
 +      return 0;
 +}
 +
 +int cmd_blame(int argc, const char **argv, const char *prefix)
 +{
 +      struct rev_info revs;
 +      const char *path;
 +      struct scoreboard sb;
 +      struct origin *o;
 +      struct blame_entry *ent;
 +      long dashdash_pos, bottom, top, lno;
 +      const char *final_commit_name = NULL;
 +      enum object_type type;
 +
 +      static const char *bottomtop = NULL;
 +      static int output_option = 0, opt = 0;
 +      static int show_stats = 0;
 +      static const char *revs_file = NULL;
 +      static const char *contents_from = NULL;
 +      static const struct option options[] = {
 +              OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
 +              OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
 +              OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
 +              OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
 +              OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
 +              OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
 +              OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
 +              OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
 +              OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
 +              OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
 +              OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
 +              OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
 +              OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
 +              OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
 +              OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
 +              { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
 +              { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
 +              OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
 +              OPT_END()
 +      };
 +
 +      struct parse_opt_ctx_t ctx;
 +      int cmd_is_annotate = !strcmp(argv[0], "annotate");
 +
 +      git_config(git_blame_config, NULL);
 +      init_revisions(&revs, NULL);
 +      revs.date_mode = blame_date_mode;
 +
 +      save_commit_buffer = 0;
 +      dashdash_pos = 0;
 +
 +      parse_options_start(&ctx, argc, argv, prefix, PARSE_OPT_KEEP_DASHDASH |
 +                          PARSE_OPT_KEEP_ARGV0);
 +      for (;;) {
 +              switch (parse_options_step(&ctx, options, blame_opt_usage)) {
 +              case PARSE_OPT_HELP:
 +                      exit(129);
 +              case PARSE_OPT_DONE:
 +                      if (ctx.argv[0])
 +                              dashdash_pos = ctx.cpidx;
 +                      goto parse_done;
 +              }
 +
 +              if (!strcmp(ctx.argv[0], "--reverse")) {
 +                      ctx.argv[0] = "--children";
 +                      reverse = 1;
 +              }
 +              parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
 +      }
 +parse_done:
 +      argc = parse_options_end(&ctx);
 +
 +      if (revs_file && read_ancestry(revs_file))
 +              die_errno("reading graft file '%s' failed", revs_file);
 +
 +      if (cmd_is_annotate) {
 +              output_option |= OUTPUT_ANNOTATE_COMPAT;
 +              blame_date_mode = DATE_ISO8601;
 +      } else {
 +              blame_date_mode = revs.date_mode;
 +      }
 +
 +      /* The maximum width used to show the dates */
 +      switch (blame_date_mode) {
 +      case DATE_RFC2822:
 +              blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
 +              break;
 +      case DATE_ISO8601:
 +              blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
 +              break;
 +      case DATE_RAW:
 +              blame_date_width = sizeof("1161298804 -0700");
 +              break;
 +      case DATE_SHORT:
 +              blame_date_width = sizeof("2006-10-19");
 +              break;
 +      case DATE_RELATIVE:
 +              /* "normal" is used as the fallback for "relative" */
 +      case DATE_LOCAL:
 +      case DATE_NORMAL:
 +              blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
 +              break;
 +      }
 +      blame_date_width -= 1; /* strip the null */
 +
 +      if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
 +              opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
 +                      PICKAXE_BLAME_COPY_HARDER);
 +
 +      if (!blame_move_score)
 +              blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
 +      if (!blame_copy_score)
 +              blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
 +
 +      /*
 +       * We have collected options unknown to us in argv[1..unk]
 +       * which are to be passed to revision machinery if we are
 +       * going to do the "bottom" processing.
 +       *
 +       * The remaining are:
 +       *
 +       * (1) if dashdash_pos != 0, its either
 +       *     "blame [revisions] -- <path>" or
 +       *     "blame -- <path> <rev>"
 +       *
 +       * (2) otherwise, its one of the two:
 +       *     "blame [revisions] <path>"
 +       *     "blame <path> <rev>"
 +       *
 +       * Note that we must strip out <path> from the arguments: we do not
 +       * want the path pruning but we may want "bottom" processing.
 +       */
 +      if (dashdash_pos) {
 +              switch (argc - dashdash_pos - 1) {
 +              case 2: /* (1b) */
 +                      if (argc != 4)
 +                              usage_with_options(blame_opt_usage, options);
 +                      /* reorder for the new way: <rev> -- <path> */
 +                      argv[1] = argv[3];
 +                      argv[3] = argv[2];
 +                      argv[2] = "--";
 +                      /* FALLTHROUGH */
 +              case 1: /* (1a) */
 +                      path = add_prefix(prefix, argv[--argc]);
 +                      argv[argc] = NULL;
 +                      break;
 +              default:
 +                      usage_with_options(blame_opt_usage, options);
 +              }
 +      } else {
 +              if (argc < 2)
 +                      usage_with_options(blame_opt_usage, options);
 +              path = add_prefix(prefix, argv[argc - 1]);
 +              if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
 +                      path = add_prefix(prefix, argv[1]);
 +                      argv[1] = argv[2];
 +              }
 +              argv[argc - 1] = "--";
 +
 +              setup_work_tree();
 +              if (!has_string_in_work_tree(path))
 +                      die_errno("cannot stat path '%s'", path);
 +      }
 +
 +      revs.disable_stdin = 1;
 +      setup_revisions(argc, argv, &revs, NULL);
 +      memset(&sb, 0, sizeof(sb));
 +
 +      sb.revs = &revs;
 +      if (!reverse)
 +              final_commit_name = prepare_final(&sb);
 +      else if (contents_from)
 +              die("--contents and --children do not blend well.");
 +      else
 +              final_commit_name = prepare_initial(&sb);
 +
 +      if (!sb.final) {
 +              /*
 +               * "--not A B -- path" without anything positive;
 +               * do not default to HEAD, but use the working tree
 +               * or "--contents".
 +               */
 +              setup_work_tree();
 +              sb.final = fake_working_tree_commit(path, contents_from);
 +              add_pending_object(&revs, &(sb.final->object), ":");
 +      }
 +      else if (contents_from)
 +              die("Cannot use --contents with final commit object name");
 +
 +      /*
 +       * If we have bottom, this will mark the ancestors of the
 +       * bottom commits we would reach while traversing as
 +       * uninteresting.
 +       */
 +      if (prepare_revision_walk(&revs))
 +              die("revision walk setup failed");
 +
 +      if (is_null_sha1(sb.final->object.sha1)) {
 +              char *buf;
 +              o = sb.final->util;
 +              buf = xmalloc(o->file.size + 1);
 +              memcpy(buf, o->file.ptr, o->file.size + 1);
 +              sb.final_buf = buf;
 +              sb.final_buf_size = o->file.size;
 +      }
 +      else {
 +              o = get_origin(&sb, sb.final, path);
 +              if (fill_blob_sha1(o))
 +                      die("no such path %s in %s", path, final_commit_name);
 +
 +              sb.final_buf = read_sha1_file(o->blob_sha1, &type,
 +                                            &sb.final_buf_size);
 +              if (!sb.final_buf)
 +                      die("Cannot read blob %s for path %s",
 +                          sha1_to_hex(o->blob_sha1),
 +                          path);
 +      }
 +      num_read_blob++;
 +      lno = prepare_lines(&sb);
 +
 +      bottom = top = 0;
 +      if (bottomtop)
 +              prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
 +      if (bottom && top && top < bottom) {
 +              long tmp;
 +              tmp = top; top = bottom; bottom = tmp;
 +      }
 +      if (bottom < 1)
 +              bottom = 1;
 +      if (top < 1)
 +              top = lno;
 +      bottom--;
 +      if (lno < top || lno < bottom)
 +              die("file %s has only %lu lines", path, lno);
 +
 +      ent = xcalloc(1, sizeof(*ent));
 +      ent->lno = bottom;
 +      ent->num_lines = top - bottom;
 +      ent->suspect = o;
 +      ent->s_lno = bottom;
 +
 +      sb.ent = ent;
 +      sb.path = path;
 +
 +      read_mailmap(&mailmap, NULL);
 +
 +      if (!incremental)
 +              setup_pager();
 +
 +      assign_blame(&sb, opt);
 +
 +      if (incremental)
 +              return 0;
 +
 +      coalesce(&sb);
 +
 +      if (!(output_option & OUTPUT_PORCELAIN))
 +              find_alignment(&sb, &output_option);
 +
 +      output(&sb, output_option);
 +      free((void *)sb.final_buf);
 +      for (ent = sb.ent; ent; ) {
 +              struct blame_entry *e = ent->next;
 +              free(ent);
 +              ent = e;
 +      }
 +
 +      if (show_stats) {
 +              printf("num read blob: %d\n", num_read_blob);
 +              printf("num get patch: %d\n", num_get_patch);
 +              printf("num commits: %d\n", num_commits);
 +      }
 +      return 0;
 +}
index 610849a6533c6fd2d3d2e8d3f8d233757174aabd,0000000000000000000000000000000000000000..b8e9e5ba01658bbdd6e9d712b00da902b6f33fbe
mode 100644,000000..100644
--- /dev/null
@@@ -1,103 -1,0 +1,103 @@@
-       xmparam_t xmp = {{XDF_NEED_MINIMAL}};
 +#include "builtin.h"
 +#include "cache.h"
 +#include "xdiff/xdiff.h"
 +#include "xdiff-interface.h"
 +#include "parse-options.h"
 +
 +static const char *const merge_file_usage[] = {
 +      "git merge-file [options] [-L name1 [-L orig [-L name2]]] file1 orig_file file2",
 +      NULL
 +};
 +
 +static int label_cb(const struct option *opt, const char *arg, int unset)
 +{
 +      static int label_count = 0;
 +      const char **names = (const char **)opt->value;
 +
 +      if (label_count >= 3)
 +              return error("too many labels on the command line");
 +      names[label_count++] = arg;
 +      return 0;
 +}
 +
 +int cmd_merge_file(int argc, const char **argv, const char *prefix)
 +{
 +      const char *names[3] = { NULL, NULL, NULL };
 +      mmfile_t mmfs[3];
 +      mmbuffer_t result = {NULL, 0};
++      xmparam_t xmp = {{0}};
 +      int ret = 0, i = 0, to_stdout = 0;
 +      int quiet = 0;
 +      int nongit;
 +      struct option options[] = {
 +              OPT_BOOLEAN('p', "stdout", &to_stdout, "send results to standard output"),
 +              OPT_SET_INT(0, "diff3", &xmp.style, "use a diff3 based merge", XDL_MERGE_DIFF3),
 +              OPT_SET_INT(0, "ours", &xmp.favor, "for conflicts, use our version",
 +                          XDL_MERGE_FAVOR_OURS),
 +              OPT_SET_INT(0, "theirs", &xmp.favor, "for conflicts, use their version",
 +                          XDL_MERGE_FAVOR_THEIRS),
 +              OPT_SET_INT(0, "union", &xmp.favor, "for conflicts, use a union version",
 +                          XDL_MERGE_FAVOR_UNION),
 +              OPT_INTEGER(0, "marker-size", &xmp.marker_size,
 +                          "for conflicts, use this marker size"),
 +              OPT__QUIET(&quiet),
 +              OPT_CALLBACK('L', NULL, names, "name",
 +                           "set labels for file1/orig_file/file2", &label_cb),
 +              OPT_END(),
 +      };
 +
 +      xmp.level = XDL_MERGE_ZEALOUS_ALNUM;
 +      xmp.style = 0;
 +      xmp.favor = 0;
 +
 +      prefix = setup_git_directory_gently(&nongit);
 +      if (!nongit) {
 +              /* Read the configuration file */
 +              git_config(git_xmerge_config, NULL);
 +              if (0 <= git_xmerge_style)
 +                      xmp.style = git_xmerge_style;
 +      }
 +
 +      argc = parse_options(argc, argv, prefix, options, merge_file_usage, 0);
 +      if (argc != 3)
 +              usage_with_options(merge_file_usage, options);
 +      if (quiet) {
 +              if (!freopen("/dev/null", "w", stderr))
 +                      return error("failed to redirect stderr to /dev/null: "
 +                                   "%s\n", strerror(errno));
 +      }
 +
 +      for (i = 0; i < 3; i++) {
 +              if (!names[i])
 +                      names[i] = argv[i];
 +              if (read_mmfile(mmfs + i, argv[i]))
 +                      return -1;
 +              if (buffer_is_binary(mmfs[i].ptr, mmfs[i].size))
 +                      return error("Cannot merge binary files: %s\n",
 +                                      argv[i]);
 +      }
 +
 +      xmp.ancestor = names[1];
 +      xmp.file1 = names[0];
 +      xmp.file2 = names[2];
 +      ret = xdl_merge(mmfs + 1, mmfs + 0, mmfs + 2, &xmp, &result);
 +
 +      for (i = 0; i < 3; i++)
 +              free(mmfs[i].ptr);
 +
 +      if (ret >= 0) {
 +              const char *filename = argv[0];
 +              FILE *f = to_stdout ? stdout : fopen(filename, "wb");
 +
 +              if (!f)
 +                      ret = error("Could not open %s for writing", filename);
 +              else if (result.size &&
 +                       fwrite(result.ptr, result.size, 1, f) != 1)
 +                      ret = error("Could not write to %s", filename);
 +              else if (fclose(f))
 +                      ret = error("Could not close %s", filename);
 +              free(result.ptr);
 +      }
 +
 +      return ret;
 +}
index a4a4f2ce4c3f147062070c2acc08eaf9f4d40be8,0000000000000000000000000000000000000000..fc00d794d641c146023a78d7d8e4143f0fdac302
mode 100644,000000..100644
--- /dev/null
@@@ -1,358 -1,0 +1,358 @@@
-       xpp.flags = XDF_NEED_MINIMAL;
 +#include "cache.h"
 +#include "tree-walk.h"
 +#include "xdiff-interface.h"
 +#include "blob.h"
 +#include "exec_cmd.h"
 +
 +static const char merge_tree_usage[] = "git merge-tree <base-tree> <branch1> <branch2>";
 +static int resolve_directories = 1;
 +
 +struct merge_list {
 +      struct merge_list *next;
 +      struct merge_list *link;        /* other stages for this object */
 +
 +      unsigned int stage : 2,
 +                   flags : 30;
 +      unsigned int mode;
 +      const char *path;
 +      struct blob *blob;
 +};
 +
 +static struct merge_list *merge_result, **merge_result_end = &merge_result;
 +
 +static void add_merge_entry(struct merge_list *entry)
 +{
 +      *merge_result_end = entry;
 +      merge_result_end = &entry->next;
 +}
 +
 +static void merge_trees(struct tree_desc t[3], const char *base);
 +
 +static const char *explanation(struct merge_list *entry)
 +{
 +      switch (entry->stage) {
 +      case 0:
 +              return "merged";
 +      case 3:
 +              return "added in remote";
 +      case 2:
 +              if (entry->link)
 +                      return "added in both";
 +              return "added in local";
 +      }
 +
 +      /* Existed in base */
 +      entry = entry->link;
 +      if (!entry)
 +              return "removed in both";
 +
 +      if (entry->link)
 +              return "changed in both";
 +
 +      if (entry->stage == 3)
 +              return "removed in local";
 +      return "removed in remote";
 +}
 +
 +extern void *merge_file(const char *, struct blob *, struct blob *, struct blob *, unsigned long *);
 +
 +static void *result(struct merge_list *entry, unsigned long *size)
 +{
 +      enum object_type type;
 +      struct blob *base, *our, *their;
 +
 +      if (!entry->stage)
 +              return read_sha1_file(entry->blob->object.sha1, &type, size);
 +      base = NULL;
 +      if (entry->stage == 1) {
 +              base = entry->blob;
 +              entry = entry->link;
 +      }
 +      our = NULL;
 +      if (entry && entry->stage == 2) {
 +              our = entry->blob;
 +              entry = entry->link;
 +      }
 +      their = NULL;
 +      if (entry)
 +              their = entry->blob;
 +      return merge_file(entry->path, base, our, their, size);
 +}
 +
 +static void *origin(struct merge_list *entry, unsigned long *size)
 +{
 +      enum object_type type;
 +      while (entry) {
 +              if (entry->stage == 2)
 +                      return read_sha1_file(entry->blob->object.sha1, &type, size);
 +              entry = entry->link;
 +      }
 +      return NULL;
 +}
 +
 +static int show_outf(void *priv_, mmbuffer_t *mb, int nbuf)
 +{
 +      int i;
 +      for (i = 0; i < nbuf; i++)
 +              printf("%.*s", (int) mb[i].size, mb[i].ptr);
 +      return 0;
 +}
 +
 +static void show_diff(struct merge_list *entry)
 +{
 +      unsigned long size;
 +      mmfile_t src, dst;
 +      xpparam_t xpp;
 +      xdemitconf_t xecfg;
 +      xdemitcb_t ecb;
 +
++      xpp.flags = 0;
 +      memset(&xecfg, 0, sizeof(xecfg));
 +      xecfg.ctxlen = 3;
 +      ecb.outf = show_outf;
 +      ecb.priv = NULL;
 +
 +      src.ptr = origin(entry, &size);
 +      if (!src.ptr)
 +              size = 0;
 +      src.size = size;
 +      dst.ptr = result(entry, &size);
 +      if (!dst.ptr)
 +              size = 0;
 +      dst.size = size;
 +      xdi_diff(&src, &dst, &xpp, &xecfg, &ecb);
 +      free(src.ptr);
 +      free(dst.ptr);
 +}
 +
 +static void show_result_list(struct merge_list *entry)
 +{
 +      printf("%s\n", explanation(entry));
 +      do {
 +              struct merge_list *link = entry->link;
 +              static const char *desc[4] = { "result", "base", "our", "their" };
 +              printf("  %-6s %o %s %s\n", desc[entry->stage], entry->mode, sha1_to_hex(entry->blob->object.sha1), entry->path);
 +              entry = link;
 +      } while (entry);
 +}
 +
 +static void show_result(void)
 +{
 +      struct merge_list *walk;
 +
 +      walk = merge_result;
 +      while (walk) {
 +              show_result_list(walk);
 +              show_diff(walk);
 +              walk = walk->next;
 +      }
 +}
 +
 +/* An empty entry never compares same, not even to another empty entry */
 +static int same_entry(struct name_entry *a, struct name_entry *b)
 +{
 +      return  a->sha1 &&
 +              b->sha1 &&
 +              !hashcmp(a->sha1, b->sha1) &&
 +              a->mode == b->mode;
 +}
 +
 +static struct merge_list *create_entry(unsigned stage, unsigned mode, const unsigned char *sha1, const char *path)
 +{
 +      struct merge_list *res = xcalloc(1, sizeof(*res));
 +
 +      res->stage = stage;
 +      res->path = path;
 +      res->mode = mode;
 +      res->blob = lookup_blob(sha1);
 +      return res;
 +}
 +
 +static char *traverse_path(const struct traverse_info *info, const struct name_entry *n)
 +{
 +      char *path = xmalloc(traverse_path_len(info, n) + 1);
 +      return make_traverse_path(path, info, n);
 +}
 +
 +static void resolve(const struct traverse_info *info, struct name_entry *branch1, struct name_entry *result)
 +{
 +      struct merge_list *orig, *final;
 +      const char *path;
 +
 +      /* If it's already branch1, don't bother showing it */
 +      if (!branch1)
 +              return;
 +
 +      path = traverse_path(info, result);
 +      orig = create_entry(2, branch1->mode, branch1->sha1, path);
 +      final = create_entry(0, result->mode, result->sha1, path);
 +
 +      final->link = orig;
 +
 +      add_merge_entry(final);
 +}
 +
 +static int unresolved_directory(const struct traverse_info *info, struct name_entry n[3])
 +{
 +      char *newbase;
 +      struct name_entry *p;
 +      struct tree_desc t[3];
 +      void *buf0, *buf1, *buf2;
 +
 +      if (!resolve_directories)
 +              return 0;
 +      p = n;
 +      if (!p->mode) {
 +              p++;
 +              if (!p->mode)
 +                      p++;
 +      }
 +      if (!S_ISDIR(p->mode))
 +              return 0;
 +      newbase = traverse_path(info, p);
 +      buf0 = fill_tree_descriptor(t+0, n[0].sha1);
 +      buf1 = fill_tree_descriptor(t+1, n[1].sha1);
 +      buf2 = fill_tree_descriptor(t+2, n[2].sha1);
 +      merge_trees(t, newbase);
 +
 +      free(buf0);
 +      free(buf1);
 +      free(buf2);
 +      free(newbase);
 +      return 1;
 +}
 +
 +
 +static struct merge_list *link_entry(unsigned stage, const struct traverse_info *info, struct name_entry *n, struct merge_list *entry)
 +{
 +      const char *path;
 +      struct merge_list *link;
 +
 +      if (!n->mode)
 +              return entry;
 +      if (entry)
 +              path = entry->path;
 +      else
 +              path = traverse_path(info, n);
 +      link = create_entry(stage, n->mode, n->sha1, path);
 +      link->link = entry;
 +      return link;
 +}
 +
 +static void unresolved(const struct traverse_info *info, struct name_entry n[3])
 +{
 +      struct merge_list *entry = NULL;
 +
 +      if (unresolved_directory(info, n))
 +              return;
 +
 +      /*
 +       * Do them in reverse order so that the resulting link
 +       * list has the stages in order - link_entry adds new
 +       * links at the front.
 +       */
 +      entry = link_entry(3, info, n + 2, entry);
 +      entry = link_entry(2, info, n + 1, entry);
 +      entry = link_entry(1, info, n + 0, entry);
 +
 +      add_merge_entry(entry);
 +}
 +
 +/*
 + * Merge two trees together (t[1] and t[2]), using a common base (t[0])
 + * as the origin.
 + *
 + * This walks the (sorted) trees in lock-step, checking every possible
 + * name. Note that directories automatically sort differently from other
 + * files (see "base_name_compare"), so you'll never see file/directory
 + * conflicts, because they won't ever compare the same.
 + *
 + * IOW, if a directory changes to a filename, it will automatically be
 + * seen as the directory going away, and the filename being created.
 + *
 + * Think of this as a three-way diff.
 + *
 + * The output will be either:
 + *  - successful merge
 + *     "0 mode sha1 filename"
 + *    NOTE NOTE NOTE! FIXME! We really really need to walk the index
 + *    in parallel with this too!
 + *
 + *  - conflict:
 + *    "1 mode sha1 filename"
 + *    "2 mode sha1 filename"
 + *    "3 mode sha1 filename"
 + *    where not all of the 1/2/3 lines may exist, of course.
 + *
 + * The successful merge rules are the same as for the three-way merge
 + * in git-read-tree.
 + */
 +static int threeway_callback(int n, unsigned long mask, unsigned long dirmask, struct name_entry *entry, struct traverse_info *info)
 +{
 +      /* Same in both? */
 +      if (same_entry(entry+1, entry+2)) {
 +              if (entry[0].sha1) {
 +                      resolve(info, NULL, entry+1);
 +                      return mask;
 +              }
 +      }
 +
 +      if (same_entry(entry+0, entry+1)) {
 +              if (entry[2].sha1 && !S_ISDIR(entry[2].mode)) {
 +                      resolve(info, entry+1, entry+2);
 +                      return mask;
 +              }
 +      }
 +
 +      if (same_entry(entry+0, entry+2)) {
 +              if (entry[1].sha1 && !S_ISDIR(entry[1].mode)) {
 +                      resolve(info, NULL, entry+1);
 +                      return mask;
 +              }
 +      }
 +
 +      unresolved(info, entry);
 +      return mask;
 +}
 +
 +static void merge_trees(struct tree_desc t[3], const char *base)
 +{
 +      struct traverse_info info;
 +
 +      setup_traverse_info(&info, base);
 +      info.fn = threeway_callback;
 +      traverse_trees(3, t, &info);
 +}
 +
 +static void *get_tree_descriptor(struct tree_desc *desc, const char *rev)
 +{
 +      unsigned char sha1[20];
 +      void *buf;
 +
 +      if (get_sha1(rev, sha1))
 +              die("unknown rev %s", rev);
 +      buf = fill_tree_descriptor(desc, sha1);
 +      if (!buf)
 +              die("%s is not a tree", rev);
 +      return buf;
 +}
 +
 +int cmd_merge_tree(int argc, const char **argv, const char *prefix)
 +{
 +      struct tree_desc t[3];
 +      void *buf1, *buf2, *buf3;
 +
 +      if (argc != 4)
 +              usage(merge_tree_usage);
 +
 +      buf1 = get_tree_descriptor(t+0, argv[1]);
 +      buf2 = get_tree_descriptor(t+1, argv[2]);
 +      buf3 = get_tree_descriptor(t+2, argv[3]);
 +      merge_trees(t, "");
 +      free(buf1);
 +      free(buf2);
 +      free(buf3);
 +
 +      show_result();
 +      return 0;
 +}
index 34f9acee910406c7ba0bf9eed76267a0cf8f46c2,0000000000000000000000000000000000000000..0048f9ef7fee24e5e058ef226f3b0fc93703fcf1
mode 100644,000000..100644
--- /dev/null
@@@ -1,155 -1,0 +1,155 @@@
-       xpp.flags = XDF_NEED_MINIMAL;
 +#include "builtin.h"
 +#include "cache.h"
 +#include "dir.h"
 +#include "string-list.h"
 +#include "rerere.h"
 +#include "xdiff/xdiff.h"
 +#include "xdiff-interface.h"
 +
 +static const char git_rerere_usage[] =
 +"git rerere [clear | status | diff | gc]";
 +
 +/* these values are days */
 +static int cutoff_noresolve = 15;
 +static int cutoff_resolve = 60;
 +
 +static time_t rerere_created_at(const char *name)
 +{
 +      struct stat st;
 +      return stat(rerere_path(name, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
 +}
 +
 +static void unlink_rr_item(const char *name)
 +{
 +      unlink(rerere_path(name, "thisimage"));
 +      unlink(rerere_path(name, "preimage"));
 +      unlink(rerere_path(name, "postimage"));
 +      rmdir(git_path("rr-cache/%s", name));
 +}
 +
 +static int git_rerere_gc_config(const char *var, const char *value, void *cb)
 +{
 +      if (!strcmp(var, "gc.rerereresolved"))
 +              cutoff_resolve = git_config_int(var, value);
 +      else if (!strcmp(var, "gc.rerereunresolved"))
 +              cutoff_noresolve = git_config_int(var, value);
 +      else
 +              return git_default_config(var, value, cb);
 +      return 0;
 +}
 +
 +static void garbage_collect(struct string_list *rr)
 +{
 +      struct string_list to_remove = { NULL, 0, 0, 1 };
 +      DIR *dir;
 +      struct dirent *e;
 +      int i, cutoff;
 +      time_t now = time(NULL), then;
 +
 +      git_config(git_rerere_gc_config, NULL);
 +      dir = opendir(git_path("rr-cache"));
 +      if (!dir)
 +              die_errno("unable to open rr-cache directory");
 +      while ((e = readdir(dir))) {
 +              if (is_dot_or_dotdot(e->d_name))
 +                      continue;
 +              then = rerere_created_at(e->d_name);
 +              if (!then)
 +                      continue;
 +              cutoff = (has_rerere_resolution(e->d_name)
 +                        ? cutoff_resolve : cutoff_noresolve);
 +              if (then < now - cutoff * 86400)
 +                      string_list_append(e->d_name, &to_remove);
 +      }
 +      for (i = 0; i < to_remove.nr; i++)
 +              unlink_rr_item(to_remove.items[i].string);
 +      string_list_clear(&to_remove, 0);
 +}
 +
 +static int outf(void *dummy, mmbuffer_t *ptr, int nbuf)
 +{
 +      int i;
 +      for (i = 0; i < nbuf; i++)
 +              if (write_in_full(1, ptr[i].ptr, ptr[i].size) != ptr[i].size)
 +                      return -1;
 +      return 0;
 +}
 +
 +static int diff_two(const char *file1, const char *label1,
 +              const char *file2, const char *label2)
 +{
 +      xpparam_t xpp;
 +      xdemitconf_t xecfg;
 +      xdemitcb_t ecb;
 +      mmfile_t minus, plus;
 +
 +      if (read_mmfile(&minus, file1) || read_mmfile(&plus, file2))
 +              return 1;
 +
 +      printf("--- a/%s\n+++ b/%s\n", label1, label2);
 +      fflush(stdout);
 +      memset(&xpp, 0, sizeof(xpp));
++      xpp.flags = 0;
 +      memset(&xecfg, 0, sizeof(xecfg));
 +      xecfg.ctxlen = 3;
 +      ecb.outf = outf;
 +      xdi_diff(&minus, &plus, &xpp, &xecfg, &ecb);
 +
 +      free(minus.ptr);
 +      free(plus.ptr);
 +      return 0;
 +}
 +
 +int cmd_rerere(int argc, const char **argv, const char *prefix)
 +{
 +      struct string_list merge_rr = { NULL, 0, 0, 1 };
 +      int i, fd, flags = 0;
 +
 +      if (2 < argc) {
 +              if (!strcmp(argv[1], "-h"))
 +                      usage(git_rerere_usage);
 +              if (!strcmp(argv[1], "--rerere-autoupdate"))
 +                      flags = RERERE_AUTOUPDATE;
 +              else if (!strcmp(argv[1], "--no-rerere-autoupdate"))
 +                      flags = RERERE_NOAUTOUPDATE;
 +              if (flags) {
 +                      argc--;
 +                      argv++;
 +              }
 +      }
 +      if (argc < 2)
 +              return rerere(flags);
 +
 +      if (!strcmp(argv[1], "forget")) {
 +              const char **pathspec = get_pathspec(prefix, argv + 2);
 +              return rerere_forget(pathspec);
 +      }
 +
 +      fd = setup_rerere(&merge_rr, flags);
 +      if (fd < 0)
 +              return 0;
 +
 +      if (!strcmp(argv[1], "clear")) {
 +              for (i = 0; i < merge_rr.nr; i++) {
 +                      const char *name = (const char *)merge_rr.items[i].util;
 +                      if (!has_rerere_resolution(name))
 +                              unlink_rr_item(name);
 +              }
 +              unlink_or_warn(git_path("rr-cache/MERGE_RR"));
 +      } else if (!strcmp(argv[1], "gc"))
 +              garbage_collect(&merge_rr);
 +      else if (!strcmp(argv[1], "status"))
 +              for (i = 0; i < merge_rr.nr; i++)
 +                      printf("%s\n", merge_rr.items[i].string);
 +      else if (!strcmp(argv[1], "diff"))
 +              for (i = 0; i < merge_rr.nr; i++) {
 +                      const char *path = merge_rr.items[i].string;
 +                      const char *name = (const char *)merge_rr.items[i].util;
 +                      diff_two(rerere_path(name, "preimage"), path, path, path);
 +              }
 +      else
 +              usage(git_rerere_usage);
 +
 +      string_list_clear(&merge_rr, 1);
 +      return 0;
 +}
diff --cc combine-diff.c
Simple merge
diff --cc diff.c
index e40c1271da3f2ea433b55f770408e8d3a0f8a521,0924274dd3c21e6cf247a974cf0e492d907d8736..50cf8fb556c743390ac02d57d1bcbf7b4edd485d
--- 1/diff.c
--- 2/diff.c
+++ b/diff.c
@@@ -1834,9 -1832,10 +1834,9 @@@ static void builtin_diffstat(const cha
  
                memset(&xpp, 0, sizeof(xpp));
                memset(&xecfg, 0, sizeof(xecfg));
-               xpp.flags = XDF_NEED_MINIMAL | o->xdl_opts;
+               xpp.flags = o->xdl_opts;
                xdi_diff_outf(&mf1, &mf2, diffstat_consume, diffstat,
 -                            &xpp, &xecfg, &ecb);
 +                            &xpp, &xecfg);
        }
  
   free_and_return:
@@@ -1882,9 -1882,9 +1882,9 @@@ static void builtin_checkdiff(const cha
                memset(&xpp, 0, sizeof(xpp));
                memset(&xecfg, 0, sizeof(xecfg));
                xecfg.ctxlen = 1; /* at least one context line */
-               xpp.flags = XDF_NEED_MINIMAL;
+               xpp.flags = 0;
                xdi_diff_outf(&mf1, &mf2, checkdiff_consume, &data,
 -                            &xpp, &xecfg, &ecb);
 +                            &xpp, &xecfg);
  
                if (data.ws_rule & WS_BLANK_AT_EOF) {
                        struct emit_callback ecbdata;
diff --cc merge-file.c
Simple merge