Add small cache for author names to reduce memory foot-print

Numbers from an old linux kernel repository with 2740 unique authors
shows that it takes a little longer time to start up but the gain is
quite good. Furthermore, it lifts the name length restriction.

18191 fonseca   20   0 19712  17m  872 S    3  0.8   0:01.34 tig
18180 fonseca   20   0 23224  20m  868 S    0  1.0   0:01.26 tig-0.14.1

diff --git a/tig.c b/tig.c
index 180661a4fcfbc94067a795898df1a6fc6a6e0330..20f7d5498cf32ef16d0aba335985e6c8456a96f1 100644 (file)
--- a/tig.c
+++ b/tig.c
@@ -3328,6 +3328,46 @@ view_driver(struct view *view, enum request request)
  * View backend utilities
  */
 
+/* Small author cache to reduce memory consumption. It uses binary
+ * search to lookup or find place to position new entries. No entries
+ * are ever freed. */
+static const char *
+get_author(const char *name)
+{
+       static const char **authors;
+       static size_t authors_alloc;
+       static size_t authors_size;
+       const char **tmp;
+       int from = 0, to = authors_size - 1;
+
+       while (from <= to) {
+               size_t pos = (to + from) / 2;
+               int cmp = strcmp(name, authors[pos]);
+
+               if (!cmp)
+                       return authors[pos];
+
+               if (cmp < 0)
+                       to = pos - 1;
+               else
+                       from = pos + 1;
+       }
+
+       tmp = realloc_items(authors, &authors_alloc, authors_size + 1, sizeof(*authors));
+       if (!tmp)
+               return NULL;
+       name = strdup(name);
+       if (!name)
+               return NULL;
+
+       authors = tmp;
+       memmove(authors + from + 1, authors + from, (authors_size - from) * sizeof(*authors));
+       authors[from] = name;
+       authors_size++;
+
+       return name;
+}
+
 static void
 parse_timezone(time_t *time, const char *zone)
 {
@@ -3348,7 +3388,7 @@ parse_timezone(time_t *time, const char *zone)
  *     author  <email@address.tld> 1138474660 +0100
  */
 static void
-parse_author_line(char *ident, char *author, size_t authorsize, struct tm *tm)
+parse_author_line(char *ident, const char **author, struct tm *tm)
 {
        char *nameend = strchr(ident, '<');
        char *emailend = strchr(ident, '>');
@@ -3363,7 +3403,7 @@ parse_author_line(char *ident, char *author, size_t authorsize, struct tm *tm)
                        ident = "Unknown";
        }
 
-       string_ncopy_do(author, authorsize, ident, strlen(ident));
+       *author = get_author(ident);
 
        /* Parse epoch and timezone */
        if (emailend && emailend[1] == ' ') {
@@ -3796,7 +3836,7 @@ struct tree_entry {
        char id[SIZEOF_REV];
        mode_t mode;
        struct tm time;                 /* Date from the author ident. */
-       char author[75];                /* Author of the commit. */
+       const char *author;             /* Author of the commit. */
        char name[1];
 };
 
@@ -3839,7 +3879,7 @@ tree_entry(struct view *view, enum line_type type, const char *path,
 static bool
 tree_read_date(struct view *view, char *text, bool *read_date)
 {
-       static char author_name[SIZEOF_STR];
+       static const char *author_name;
        static struct tm author_time;
 
        if (!text && *read_date) {
@@ -3873,7 +3913,7 @@ tree_read_date(struct view *view, char *text, bool *read_date)
 
        } else if (*text == 'a' && get_line_type(text) == LINE_AUTHOR) {
                parse_author_line(text + STRING_SIZE("author "),
-                                 author_name, sizeof(author_name), &author_time);
+                                 &author_name, &author_time);
 
        } else if (*text == ':') {
                char *pos;
@@ -3896,11 +3936,11 @@ tree_read_date(struct view *view, char *text, bool *read_date)
                        struct line *line = &view->line[i];
                        struct tree_entry *entry = line->data;
 
-                       annotated += !!*entry->author;
-                       if (*entry->author || strcmp(entry->name, text))
+                       annotated += !!entry->author;
+                       if (entry->author || strcmp(entry->name, text))
                                continue;
 
-                       string_copy(entry->author, author_name);
+                       entry->author = author_name;
                        memcpy(&entry->time, &author_time, sizeof(entry->time));
                        line->dirty = 1;
                        break;
@@ -3989,7 +4029,7 @@ tree_draw(struct view *view, struct line *line, unsigned int lineno)
                if (opt_author && draw_author(view, entry->author))
                        return TRUE;
 
-               if (opt_date && draw_date(view, *entry->author ? &entry->time : NULL))
+               if (opt_date && draw_date(view, entry->author ? &entry->time : NULL))
                        return TRUE;
        }
        if (draw_text(view, line->type, entry->name, TRUE))
@@ -4185,7 +4225,7 @@ static const char *blame_cat_file_argv[] = {
 struct blame_commit {
        char id[SIZEOF_REV];            /* SHA1 ID. */
        char title[128];                /* First line of the commit message. */
-       char author[75];                /* Author of the commit. */
+       const char *author;             /* Author of the commit. */
        struct tm time;                 /* Date from the author ident. */
        char filename[128];             /* Name of file. */
        bool has_previous;              /* Was a "previous" line detected. */
@@ -4363,7 +4403,7 @@ blame_read(struct view *view, char *line)
                              view->lines ? blamed * 100 / view->lines : 0);
 
        } else if (match_blame_header("author ", &line)) {
-               string_ncopy(commit->author, line, strlen(line));
+               commit->author = get_author(line);
 
        } else if (match_blame_header("author-time ", &line)) {
                author_time = (time_t) atol(line);
@@ -5609,7 +5649,7 @@ static struct view_ops stage_ops = {
 struct commit {
        char id[SIZEOF_REV];            /* SHA1 ID. */
        char title[128];                /* First line of the commit message. */
-       char author[75];                /* Author of the commit. */
+       const char *author;             /* Author of the commit. */
        struct tm time;                 /* Date from the author ident. */
        struct ref **refs;              /* Repository references. */
        chtype graph[SIZEOF_REVGRAPH];  /* Ancestry chain graphics. */
@@ -5836,7 +5876,7 @@ main_draw(struct view *view, struct line *line, unsigned int lineno)
 {
        struct commit *commit = line->data;
 
-       if (!*commit->author)
+       if (!commit->author)
                return FALSE;
 
        if (opt_date && draw_date(view, &commit->time))
@@ -5898,7 +5938,7 @@ main_read(struct view *view, char *line)
                if (view->lines > 0) {
                        commit = view->line[view->lines - 1].data;
                        view->line[view->lines - 1].dirty = 1;
-                       if (!*commit->author) {
+                       if (!commit->author) {
                                view->lines--;
                                free(commit);
                                graph->commit = NULL;
@@ -5949,8 +5989,7 @@ main_read(struct view *view, char *line)
 
        case LINE_AUTHOR:
                parse_author_line(line + STRING_SIZE("author "),
-                                 commit->author, sizeof(commit->author),
-                                 &commit->time);
+                                 &commit->author, &commit->time);
                update_rev_graph(view, graph);
                graph = graph->next;
                break;
@@ -6026,7 +6065,7 @@ main_grep(struct view *view, struct line *line)
        regmatch_t pmatch;
 
        for (state = S_TITLE; state < S_END; state++) {
-               char *text;
+               const char *text;
 
                switch (state) {
                case S_TITLE:   text = commit->title;   break;