mergetool: respect autocrlf by using checkout-index
[gitweb.git] / pretty.c
index a29c290009587a12cdc6aec335d508d29481e697..cc460b56970c3869e67feb816f76f4804e37e7ce 100644 (file)
--- a/pretty.c
+++ b/pretty.c
@@ -5,6 +5,8 @@
 #include "revision.h"
 #include "string-list.h"
 #include "mailmap.h"
+#include "log-tree.h"
+#include "color.h"
 
 static char *user_format;
 
@@ -180,6 +182,20 @@ static int is_empty_line(const char *line, int *len_p)
        return !len;
 }
 
+static const char *skip_empty_lines(const char *msg)
+{
+       for (;;) {
+               int linelen = get_one_line(msg);
+               int ll = linelen;
+               if (!linelen)
+                       break;
+               if (!is_empty_line(msg, &ll))
+                       break;
+               msg += linelen;
+       }
+       return msg;
+}
+
 static void add_merge_info(enum cmit_fmt fmt, struct strbuf *sb,
                        const struct commit *commit, int abbrev)
 {
@@ -233,7 +249,7 @@ static char *get_header(const struct commit *commit, const char *key)
 
 static char *replace_encoding_header(char *buf, const char *encoding)
 {
-       struct strbuf tmp;
+       struct strbuf tmp = STRBUF_INIT;
        size_t start, len;
        char *cp = buf;
 
@@ -249,7 +265,6 @@ static char *replace_encoding_header(char *buf, const char *encoding)
                return buf; /* should not happen but be defensive */
        len = cp + 1 - (buf + start);
 
-       strbuf_init(&tmp, 0);
        strbuf_attach(&tmp, buf, strlen(buf), strlen(buf) + 1);
        if (is_encoding_utf8(encoding)) {
                /* we have re-coded to UTF-8; drop the header */
@@ -410,13 +425,15 @@ struct chunk {
 struct format_commit_context {
        const struct commit *commit;
        enum date_mode dmode;
+       unsigned commit_header_parsed:1;
+       unsigned commit_message_parsed:1;
 
        /* These offsets are relative to the start of the commit message. */
-       int commit_header_parsed;
-       struct chunk subject;
        struct chunk author;
        struct chunk committer;
        struct chunk encoding;
+       size_t message_off;
+       size_t subject_off;
        size_t body_off;
 
        /* The following ones are relative to the result struct strbuf. */
@@ -446,23 +463,14 @@ static void parse_commit_header(struct format_commit_context *context)
 {
        const char *msg = context->commit->buffer;
        int i;
-       enum { HEADER, SUBJECT, BODY } state;
 
-       for (i = 0, state = HEADER; msg[i] && state < BODY; i++) {
+       for (i = 0; msg[i]; i++) {
                int eol;
                for (eol = i; msg[eol] && msg[eol] != '\n'; eol++)
                        ; /* do nothing */
 
-               if (state == SUBJECT) {
-                       context->subject.off = i;
-                       context->subject.len = eol - i;
-                       i = eol;
-               }
                if (i == eol) {
-                       state++;
-                       /* strip empty lines */
-                       while (msg[eol] == '\n' && msg[eol + 1] == '\n')
-                               eol++;
+                       break;
                } else if (!prefixcmp(msg + i, "author ")) {
                        context->author.off = i + 7;
                        context->author.len = eol - i - 7;
@@ -474,13 +482,67 @@ static void parse_commit_header(struct format_commit_context *context)
                        context->encoding.len = eol - i - 9;
                }
                i = eol;
-               if (!msg[i])
-                       break;
        }
-       context->body_off = i;
+       context->message_off = i;
        context->commit_header_parsed = 1;
 }
 
+const char *format_subject(struct strbuf *sb, const char *msg,
+                          const char *line_separator)
+{
+       int first = 1;
+
+       for (;;) {
+               const char *line = msg;
+               int linelen = get_one_line(line);
+
+               msg += linelen;
+               if (!linelen || is_empty_line(line, &linelen))
+                       break;
+
+               if (!sb)
+                       continue;
+               strbuf_grow(sb, linelen + 2);
+               if (!first)
+                       strbuf_addstr(sb, line_separator);
+               strbuf_add(sb, line, linelen);
+               first = 0;
+       }
+       return msg;
+}
+
+static void parse_commit_message(struct format_commit_context *c)
+{
+       const char *msg = c->commit->buffer + c->message_off;
+       const char *start = c->commit->buffer;
+
+       msg = skip_empty_lines(msg);
+       c->subject_off = msg - start;
+
+       msg = format_subject(NULL, msg, NULL);
+       msg = skip_empty_lines(msg);
+       c->body_off = msg - start;
+
+       c->commit_message_parsed = 1;
+}
+
+static void format_decoration(struct strbuf *sb, const struct commit *commit)
+{
+       struct name_decoration *d;
+       const char *prefix = " (";
+
+       load_ref_decorations();
+       d = lookup_decoration(&name_decoration, &commit->object);
+       while (d) {
+               strbuf_addstr(sb, prefix);
+               prefix = ", ";
+               strbuf_addstr(sb, d->name);
+               d = d->next;
+       }
+       if (prefix[0] == ',')
+               strbuf_addch(sb, ')');
+}
+
 static size_t format_commit_item(struct strbuf *sb, const char *placeholder,
                                void *context)
 {
@@ -493,6 +555,17 @@ static size_t format_commit_item(struct strbuf *sb, const char *placeholder,
        /* these are independent of the commit */
        switch (placeholder[0]) {
        case 'C':
+               if (placeholder[1] == '(') {
+                       const char *end = strchr(placeholder + 2, ')');
+                       char color[COLOR_MAXLEN];
+                       if (!end)
+                               return 0;
+                       color_parse_mem(placeholder + 2,
+                                       end - (placeholder + 2),
+                                       "--pretty format", color);
+                       strbuf_addstr(sb, color);
+                       return end - placeholder + 1;
+               }
                if (!prefixcmp(placeholder + 1, "red")) {
                        strbuf_addstr(sb, "\033[31m");
                        return 4;
@@ -573,6 +646,9 @@ static size_t format_commit_item(struct strbuf *sb, const char *placeholder,
                                 ? '<'
                                 : '>');
                return 1;
+       case 'd':
+               format_decoration(sb, commit);
+               return 1;
        }
 
        /* For the rest we have to parse the commit header. */
@@ -580,9 +656,6 @@ static size_t format_commit_item(struct strbuf *sb, const char *placeholder,
                parse_commit_header(c);
 
        switch (placeholder[0]) {
-       case 's':       /* subject */
-               strbuf_add(sb, msg + c->subject.off, c->subject.len);
-               return 1;
        case 'a':       /* author ... */
                return format_person_part(sb, placeholder[1],
                                   msg + c->author.off, c->author.len,
@@ -594,6 +667,16 @@ static size_t format_commit_item(struct strbuf *sb, const char *placeholder,
        case 'e':       /* encoding */
                strbuf_add(sb, msg + c->encoding.off, c->encoding.len);
                return 1;
+       }
+
+       /* Now we need to parse the commit message. */
+       if (!c->commit_message_parsed)
+               parse_commit_message(c);
+
+       switch (placeholder[0]) {
+       case 's':       /* subject */
+               format_subject(sb, msg + c->subject_off, " ");
+               return 1;
        case 'b':       /* body */
                strbuf_addstr(sb, msg + c->body_off);
                return 1;
@@ -684,27 +767,11 @@ void pp_title_line(enum cmit_fmt fmt,
                   const char *encoding,
                   int need_8bit_cte)
 {
+       const char *line_separator = (fmt == CMIT_FMT_EMAIL) ? "\n " : " ";
        struct strbuf title;
 
        strbuf_init(&title, 80);
-
-       for (;;) {
-               const char *line = *msg_p;
-               int linelen = get_one_line(line);
-
-               *msg_p += linelen;
-               if (!linelen || is_empty_line(line, &linelen))
-                       break;
-
-               strbuf_grow(&title, linelen + 2);
-               if (title.len) {
-                       if (fmt == CMIT_FMT_EMAIL) {
-                               strbuf_addch(&title, '\n');
-                       }
-                       strbuf_addch(&title, ' ');
-               }
-               strbuf_add(&title, line, linelen);
-       }
+       *msg_p = format_subject(&title, *msg_p, line_separator);
 
        strbuf_grow(sb, title.len + 1024);
        if (subject) {
@@ -763,6 +830,20 @@ void pp_remainder(enum cmit_fmt fmt,
        }
 }
 
+char *reencode_commit_message(const struct commit *commit, const char **encoding_p)
+{
+       const char *encoding;
+
+       encoding = (git_log_output_encoding
+                   ? git_log_output_encoding
+                   : git_commit_encoding);
+       if (!encoding)
+               encoding = "utf-8";
+       if (encoding_p)
+               *encoding_p = encoding;
+       return logmsg_reencode(commit, encoding);
+}
+
 void pretty_print_commit(enum cmit_fmt fmt, const struct commit *commit,
                         struct strbuf *sb, int abbrev,
                         const char *subject, const char *after_subject,
@@ -779,12 +860,7 @@ void pretty_print_commit(enum cmit_fmt fmt, const struct commit *commit,
                return;
        }
 
-       encoding = (git_log_output_encoding
-                   ? git_log_output_encoding
-                   : git_commit_encoding);
-       if (!encoding)
-               encoding = "utf-8";
-       reencoded = logmsg_reencode(commit, encoding);
+       reencoded = reencode_commit_message(commit, &encoding);
        if (reencoded) {
                msg = reencoded;
        }
@@ -821,15 +897,7 @@ void pretty_print_commit(enum cmit_fmt fmt, const struct commit *commit,
        }
 
        /* Skip excess blank lines at the beginning of body, if any... */
-       for (;;) {
-               int linelen = get_one_line(msg);
-               int ll = linelen;
-               if (!linelen)
-                       break;
-               if (!is_empty_line(msg, &ll))
-                       break;
-               msg += linelen;
-       }
+       msg = skip_empty_lines(msg);
 
        /* These formats treat the title line specially. */
        if (fmt == CMIT_FMT_ONELINE || fmt == CMIT_FMT_EMAIL)