patch, 4 partsrefactor the --color-words to make it more hackable

color.c, color.h

diff --git a/color.c b/color.c
index fc0b72a..bff24ac 100644
--- a/color.c
+++ b/color.c
@@ -191,3 +191,27 @@ int color_fprintf_ln(FILE *fp, const char *color, const char *fmt, ...)
 	va_end(args);
 	return r;
 }
+
+/*
+ * This function splits the buffer by newlines and colors the lines individually.
+ */
+void color_fwrite(FILE *f, const char *color, size_t count, const char *buf)
+{
+	if (!*color) {
+		fwrite(buf, count, 1, f);
+		return;
+	}
+	while (count) {
+		char *p = memchr(buf, '\n', count);
+		fputs(color, f);
+		fwrite(buf, p ? p - buf : count, 1, f);
+		fputs(COLOR_RESET, f);
+		if (!p)
+			return;
+		fputc('\n', f);
+		count -= p + 1 - buf;
+		buf = p + 1;
+	}
+}
+
+
diff --git a/color.h b/color.h
index 6cf5c88..9fb58f5 100644
--- a/color.h
+++ b/color.h
@@ -19,5 +19,6 @@ int git_config_colorbool(const char *var, const char *value, int stdout_is_tty);
 void color_parse(const char *var, const char *value, char *dst);
 int color_fprintf(FILE *fp, const char *color, const char *fmt, ...);
 int color_fprintf_ln(FILE *fp, const char *color, const char *fmt, ...);
+void color_fwrite(FILE *f, const char *color, size_t count, const char *buf);
 
 #endif /* COLOR_H */

color.c, color.h

diff --git a/color.c b/color.c
index fc0b72a..b028880 100644
--- a/color.c
+++ b/color.c
@@ -191,3 +191,31 @@ int color_fprintf_ln(FILE *fp, const char *color, const char *fmt, ...)
 	va_end(args);
 	return r;
 }
+
+/*
+ * This function splits the buffer by newlines and colors the lines individually.
+ *
+ * Returns 0 on success.
+ */
+int color_fwrite_lines(FILE *fp, const char *color,
+		size_t count, const char *buf)
+{
+	if (!*color)
+		return fwrite(buf, count, 1, fp) != 1;
+	while (count) {
+		char *p = memchr(buf, '\n', count);
+		if (fputs(color, fp) < 0 ||
+				fwrite(buf, p ? p - buf : count, 1, fp) != 1 ||
+				fputs(COLOR_RESET, fp) < 0)
+			return -1;
+		if (!p)
+			return 0;
+		if (fputc('\n', fp) < 0)
+			return -1;
+		count -= p + 1 - buf;
+		buf = p + 1;
+	}
+	return 0;
+}
+
+
diff --git a/color.h b/color.h
index 6cf5c88..cd5c985 100644
--- a/color.h
+++ b/color.h
@@ -19,5 +19,6 @@ int git_config_colorbool(const char *var, const char *value, int stdout_is_tty);
 void color_parse(const char *var, const char *value, char *dst);
 int color_fprintf(FILE *fp, const char *color, const char *fmt, ...);
 int color_fprintf_ln(FILE *fp, const char *color, const char *fmt, ...);
+int color_fwrite_lines(FILE *fp, const char *color, size_t count, const char *buf);
 
 #endif /* COLOR_H */

diff --git a/diff.c b/diff.c
index f67e0b2..6d87ea5 100644
--- a/diff.c
+++ b/diff.c
@@ -326,10 +326,7 @@ struct diff_words_buffer {
 static void diff_words_append(char *line, unsigned long len,
 		struct diff_words_buffer *buffer)
 {
-	if (buffer->text.size + len > buffer->alloc) {
-		buffer->alloc = (buffer->text.size + len) * 3 / 2;
-		buffer->text.ptr = xrealloc(buffer->text.ptr, buffer->alloc);
-	}
+	ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
 	line++;
 	len--;
 	memcpy(buffer->text.ptr + buffer->text.size, line, len);
@@ -398,6 +395,22 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 	}
 }
 
+/*
+ * This function splits the words in buffer->text, and stores the list with
+ * newline separator into out.
+ */
+static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
+{
+	int i;
+	out->size = buffer->text.size;
+	out->ptr = xmalloc(out->size);
+	memcpy(out->ptr, buffer->text.ptr, out->size);
+	for (i = 0; i < out->size; i++)
+		if (isspace(out->ptr[i]))
+			out->ptr[i] = '\n';
+	buffer->current = 0;
+}
+
 /* this executes the word diff on the accumulated buffers */
 static void diff_words_show(struct diff_words_data *diff_words)
 {
@@ -405,26 +418,11 @@ static void diff_words_show(struct diff_words_data *diff_words)
 	xdemitconf_t xecfg;
 	xdemitcb_t ecb;
 	mmfile_t minus, plus;
-	int i;
 
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
-	minus.size = diff_words->minus.text.size;
-	minus.ptr = xmalloc(minus.size);
-	memcpy(minus.ptr, diff_words->minus.text.ptr, minus.size);
-	for (i = 0; i < minus.size; i++)
-		if (isspace(minus.ptr[i]))
-			minus.ptr[i] = '\n';
-	diff_words->minus.current = 0;
-
-	plus.size = diff_words->plus.text.size;
-	plus.ptr = xmalloc(plus.size);
-	memcpy(plus.ptr, diff_words->plus.text.ptr, plus.size);
-	for (i = 0; i < plus.size; i++)
-		if (isspace(plus.ptr[i]))
-			plus.ptr[i] = '\n';
-	diff_words->plus.current = 0;
-
+	diff_words_fill(&diff_words->minus, &minus);
+	diff_words_fill(&diff_words->plus, &plus);
 	xpp.flags = XDF_NEED_MINIMAL;
 	xecfg.ctxlen = diff_words->minus.alloc + diff_words->plus.alloc;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,

diff --git a/diff.c b/diff.c
index 6d87ea5..2a3d301 100644
--- a/diff.c
+++ b/diff.c
@@ -319,8 +319,10 @@ static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
 struct diff_words_buffer {
 	mmfile_t text;
 	long alloc;
-	long current; /* output pointer */
-	int suppressed_newline;
+	struct diff_words_orig {
+		const char *begin, *end;
+	} *orig;
+	int orig_nr, orig_alloc;
 };
 
 static void diff_words_append(char *line, unsigned long len,
@@ -335,80 +337,79 @@ static void diff_words_append(char *line, unsigned long len,
 
 struct diff_words_data {
 	struct diff_words_buffer minus, plus;
+	const char *current_plus;
 	FILE *file;
 };
 
-static void print_word(FILE *file, struct diff_words_buffer *buffer, int len, int color,
-		int suppress_newline)
+static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 {
-	const char *ptr;
-	int eol = 0;
+	struct diff_words_data *diff_words = priv;
+	int minus_first, minus_len, plus_first, plus_len;
+	const char *minus_begin, *minus_end, *plus_begin, *plus_end;
 
-	if (len == 0)
+	if (line[0] != '@' || parse_hunk_header(line, len,
+			&minus_first, &minus_len, &plus_first, &plus_len))
 		return;
 
-	ptr  = buffer->text.ptr + buffer->current;
-	buffer->current += len;
-
-	if (ptr[len - 1] == '\n') {
-		eol = 1;
-		len--;
-	}
-
-	fputs(diff_get_color(1, color), file);
-	fwrite(ptr, len, 1, file);
-	fputs(diff_get_color(1, DIFF_RESET), file);
-
-	if (eol) {
-		if (suppress_newline)
-			buffer->suppressed_newline = 1;
-		else
-			putc('\n', file);
-	}
-}
-
-static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
-{
-	struct diff_words_data *diff_words = priv;
+	minus_begin = diff_words->minus.orig[minus_first].begin;
+	minus_end = minus_len == 0 ? minus_begin :
+		diff_words->minus.orig[minus_first + minus_len - 1].end;
+	plus_begin = diff_words->plus.orig[plus_first].begin;
+	plus_end = plus_len == 0 ? plus_begin :
+		diff_words->plus.orig[plus_first + plus_len - 1].end;
 
-	if (diff_words->minus.suppressed_newline) {
-		if (line[0] != '+')
-			putc('\n', diff_words->file);
-		diff_words->minus.suppressed_newline = 0;
-	}
+	if (diff_words->current_plus != plus_begin)
+		fwrite(diff_words->current_plus,
+				plus_begin - diff_words->current_plus, 1,
+				diff_words->file);
+	if (minus_begin != minus_end)
+		color_fwrite(diff_words->file, diff_get_color(1, DIFF_FILE_OLD),
+				minus_end - minus_begin, minus_begin);
+	if (plus_begin != plus_end)
+		color_fwrite(diff_words->file, diff_get_color(1, DIFF_FILE_NEW),
+				plus_end - plus_begin, plus_begin);
 
-	len--;
-	switch (line[0]) {
-		case '-':
-			print_word(diff_words->file,
-				   &diff_words->minus, len, DIFF_FILE_OLD, 1);
-			break;
-		case '+':
-			print_word(diff_words->file,
-				   &diff_words->plus, len, DIFF_FILE_NEW, 0);
-			break;
-		case ' ':
-			print_word(diff_words->file,
-				   &diff_words->plus, len, DIFF_PLAIN, 0);
-			diff_words->minus.current += len;
-			break;
-	}
+	diff_words->current_plus = plus_end;
 }
 
 /*
- * This function splits the words in buffer->text, and stores the list with
- * newline separator into out.
+ * This function splits the words in buffer->text, stores the list with
+ * newline separator into out, and saves the offsets of the original words
+ * in buffer->orig.
  */
 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 {
-	int i;
-	out->size = buffer->text.size;
-	out->ptr = xmalloc(out->size);
-	memcpy(out->ptr, buffer->text.ptr, out->size);
-	for (i = 0; i < out->size; i++)
-		if (isspace(out->ptr[i]))
-			out->ptr[i] = '\n';
-	buffer->current = 0;
+	int i, j;
+
+	out->size = 0;
+	out->ptr = xmalloc(buffer->text.size);
+
+	/* fake an empty "0th" word */
+	ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
+	buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
+	buffer->orig_nr = 1;
+
+	for (i = 0; i < buffer->text.size; i++) {
+		if (isspace(buffer->text.ptr[i]))
+			continue;
+		for (j = i + 1; j < buffer->text.size &&
+				!isspace(buffer->text.ptr[j]); j++)
+			; /* find the end of the word */
+
+		/* store original boundaries */
+		ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
+				buffer->orig_alloc);
+		buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
+		buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
+		buffer->orig_nr++;
+
+		/* store one word */
+		memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
+		out->ptr[out->size + j - i] = '\n';
+		out->size += j - i + 1;
+
+		i = j - 1;
+	}
 }
 
 /* this executes the word diff on the accumulated buffers */
@@ -419,22 +420,33 @@ static void diff_words_show(struct diff_words_data *diff_words)
 	xdemitcb_t ecb;
 	mmfile_t minus, plus;
 
+	/* special case: only removal */
+	if (!diff_words->plus.text.size) {
+		color_fwrite(diff_words->file, diff_get_color(1, DIFF_FILE_OLD),
+			diff_words->minus.text.size, diff_words->minus.text.ptr);
+		diff_words->minus.text.size = 0;
+		return;
+	}
+
+	diff_words->current_plus = diff_words->plus.text.ptr;
+
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
 	diff_words_fill(&diff_words->minus, &minus);
 	diff_words_fill(&diff_words->plus, &plus);
 	xpp.flags = XDF_NEED_MINIMAL;
-	xecfg.ctxlen = diff_words->minus.alloc + diff_words->plus.alloc;
+	xecfg.ctxlen = 0;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
 		      &xpp, &xecfg, &ecb);
 	free(minus.ptr);
 	free(plus.ptr);
+	if (diff_words->current_plus != diff_words->plus.text.ptr +
+			diff_words->plus.text.size)
+		fwrite(diff_words->current_plus,
+			diff_words->plus.text.ptr + diff_words->plus.text.size
+			- diff_words->current_plus, 1,
+			diff_words->file);
 	diff_words->minus.text.size = diff_words->plus.text.size = 0;
-
-	if (diff_words->minus.suppressed_newline) {
-		putc('\n', diff_words->file);
-		diff_words->minus.suppressed_newline = 0;
-	}
 }
 
 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
@@ -458,7 +470,9 @@ static void free_diff_words_data(struct emit_callback *ecbdata)
 			diff_words_show(ecbdata->diff_words);
 
 		free (ecbdata->diff_words->minus.text.ptr);
+		free (ecbdata->diff_words->minus.orig);
 		free (ecbdata->diff_words->plus.text.ptr);
+		free (ecbdata->diff_words->plus.orig);
 		free(ecbdata->diff_words);
 		ecbdata->diff_words = NULL;
 	}

color.c, color.h

diff --git a/color.c b/color.c
index fc0b72a..d4ae83f 100644
--- a/color.c
+++ b/color.c
@@ -191,3 +191,31 @@ int color_fprintf_ln(FILE *fp, const char *color, const char *fmt, ...)
 	va_end(args);
 	return r;
 }
+
+/*
+ * This function splits the buffer by newlines and colors the lines individually.
+ *
+ * Returns 0 on success.
+ */
+int color_fwrite_lines(FILE *fp, const char *color,
+		size_t count, const char *buf)
+{
+	if (!*color)
+		return fwrite(buf, count, 1, fp) != 1;
+	while (count) {
+		char *p = memchr(buf, '\n', count);
+		if (p != buf && (fputs(color, fp) < 0 ||
+				fwrite(buf, p ? p - buf : count, 1, fp) != 1 ||
+				fputs(COLOR_RESET, fp) < 0))
+			return -1;
+		if (!p)
+			return 0;
+		if (fputc('\n', fp) < 0)
+			return -1;
+		count -= p + 1 - buf;
+		buf = p + 1;
+	}
+	return 0;
+}
+
+
diff --git a/color.h b/color.h
index 6cf5c88..cd5c985 100644
--- a/color.h
+++ b/color.h
@@ -19,5 +19,6 @@ int git_config_colorbool(const char *var, const char *value, int stdout_is_tty);
 void color_parse(const char *var, const char *value, char *dst);
 int color_fprintf(FILE *fp, const char *color, const char *fmt, ...);
 int color_fprintf_ln(FILE *fp, const char *color, const char *fmt, ...);
+int color_fwrite_lines(FILE *fp, const char *color, size_t count, const char *buf);
 
 #endif /* COLOR_H */

diff --git a/diff.c b/diff.c
index f67e0b2..6d87ea5 100644
--- a/diff.c
+++ b/diff.c
@@ -326,10 +326,7 @@ struct diff_words_buffer {
 static void diff_words_append(char *line, unsigned long len,
 		struct diff_words_buffer *buffer)
 {
-	if (buffer->text.size + len > buffer->alloc) {
-		buffer->alloc = (buffer->text.size + len) * 3 / 2;
-		buffer->text.ptr = xrealloc(buffer->text.ptr, buffer->alloc);
-	}
+	ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
 	line++;
 	len--;
 	memcpy(buffer->text.ptr + buffer->text.size, line, len);
@@ -398,6 +395,22 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 	}
 }
 
+/*
+ * This function splits the words in buffer->text, and stores the list with
+ * newline separator into out.
+ */
+static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
+{
+	int i;
+	out->size = buffer->text.size;
+	out->ptr = xmalloc(out->size);
+	memcpy(out->ptr, buffer->text.ptr, out->size);
+	for (i = 0; i < out->size; i++)
+		if (isspace(out->ptr[i]))
+			out->ptr[i] = '\n';
+	buffer->current = 0;
+}
+
 /* this executes the word diff on the accumulated buffers */
 static void diff_words_show(struct diff_words_data *diff_words)
 {
@@ -405,26 +418,11 @@ static void diff_words_show(struct diff_words_data *diff_words)
 	xdemitconf_t xecfg;
 	xdemitcb_t ecb;
 	mmfile_t minus, plus;
-	int i;
 
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
-	minus.size = diff_words->minus.text.size;
-	minus.ptr = xmalloc(minus.size);
-	memcpy(minus.ptr, diff_words->minus.text.ptr, minus.size);
-	for (i = 0; i < minus.size; i++)
-		if (isspace(minus.ptr[i]))
-			minus.ptr[i] = '\n';
-	diff_words->minus.current = 0;
-
-	plus.size = diff_words->plus.text.size;
-	plus.ptr = xmalloc(plus.size);
-	memcpy(plus.ptr, diff_words->plus.text.ptr, plus.size);
-	for (i = 0; i < plus.size; i++)
-		if (isspace(plus.ptr[i]))
-			plus.ptr[i] = '\n';
-	diff_words->plus.current = 0;
-
+	diff_words_fill(&diff_words->minus, &minus);
+	diff_words_fill(&diff_words->plus, &plus);
 	xpp.flags = XDF_NEED_MINIMAL;
 	xecfg.ctxlen = diff_words->minus.alloc + diff_words->plus.alloc;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,

diff.c, t/t4034-diff-words.sh

diff --git a/diff.c b/diff.c
index 6d87ea5..fe8b1f0 100644
--- a/diff.c
+++ b/diff.c
@@ -319,8 +319,10 @@ static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
 struct diff_words_buffer {
 	mmfile_t text;
 	long alloc;
-	long current; /* output pointer */
-	int suppressed_newline;
+	struct diff_words_orig {
+		const char *begin, *end;
+	} *orig;
+	int orig_nr, orig_alloc;
 };
 
 static void diff_words_append(char *line, unsigned long len,
@@ -335,80 +337,81 @@ static void diff_words_append(char *line, unsigned long len,
 
 struct diff_words_data {
 	struct diff_words_buffer minus, plus;
+	const char *current_plus;
 	FILE *file;
 };
 
-static void print_word(FILE *file, struct diff_words_buffer *buffer, int len, int color,
-		int suppress_newline)
-{
-	const char *ptr;
-	int eol = 0;
-
-	if (len == 0)
-		return;
-
-	ptr  = buffer->text.ptr + buffer->current;
-	buffer->current += len;
-
-	if (ptr[len - 1] == '\n') {
-		eol = 1;
-		len--;
-	}
-
-	fputs(diff_get_color(1, color), file);
-	fwrite(ptr, len, 1, file);
-	fputs(diff_get_color(1, DIFF_RESET), file);
-
-	if (eol) {
-		if (suppress_newline)
-			buffer->suppressed_newline = 1;
-		else
-			putc('\n', file);
-	}
-}
-
 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 {
 	struct diff_words_data *diff_words = priv;
+	int minus_first, minus_len, plus_first, plus_len;
+	const char *minus_begin, *minus_end, *plus_begin, *plus_end;
 
-	if (diff_words->minus.suppressed_newline) {
-		if (line[0] != '+')
-			putc('\n', diff_words->file);
-		diff_words->minus.suppressed_newline = 0;
-	}
+	if (line[0] != '@' || parse_hunk_header(line, len,
+			&minus_first, &minus_len, &plus_first, &plus_len))
+		return;
 
-	len--;
-	switch (line[0]) {
-		case '-':
-			print_word(diff_words->file,
-				   &diff_words->minus, len, DIFF_FILE_OLD, 1);
-			break;
-		case '+':
-			print_word(diff_words->file,
-				   &diff_words->plus, len, DIFF_FILE_NEW, 0);
-			break;
-		case ' ':
-			print_word(diff_words->file,
-				   &diff_words->plus, len, DIFF_PLAIN, 0);
-			diff_words->minus.current += len;
-			break;
-	}
+	minus_begin = diff_words->minus.orig[minus_first].begin;
+	minus_end = minus_len == 0 ? minus_begin :
+		diff_words->minus.orig[minus_first + minus_len - 1].end;
+	plus_begin = diff_words->plus.orig[plus_first].begin;
+	plus_end = plus_len == 0 ? plus_begin :
+		diff_words->plus.orig[plus_first + plus_len - 1].end;
+
+	if (diff_words->current_plus != plus_begin)
+		fwrite(diff_words->current_plus,
+				plus_begin - diff_words->current_plus, 1,
+				diff_words->file);
+	if (minus_begin != minus_end)
+		color_fwrite_lines(diff_words->file,
+				diff_get_color(1, DIFF_FILE_OLD),
+				minus_end - minus_begin, minus_begin);
+	if (plus_begin != plus_end)
+		color_fwrite_lines(diff_words->file,
+				diff_get_color(1, DIFF_FILE_NEW),
+				plus_end - plus_begin, plus_begin);
+
+	diff_words->current_plus = plus_end;
 }
 
 /*
- * This function splits the words in buffer->text, and stores the list with
- * newline separator into out.
+ * This function splits the words in buffer->text, stores the list with
+ * newline separator into out, and saves the offsets of the original words
+ * in buffer->orig.
  */
 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 {
-	int i;
-	out->size = buffer->text.size;
-	out->ptr = xmalloc(out->size);
-	memcpy(out->ptr, buffer->text.ptr, out->size);
-	for (i = 0; i < out->size; i++)
-		if (isspace(out->ptr[i]))
-			out->ptr[i] = '\n';
-	buffer->current = 0;
+	int i, j;
+
+	out->size = 0;
+	out->ptr = xmalloc(buffer->text.size);
+
+	/* fake an empty "0th" word */
+	ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
+	buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
+	buffer->orig_nr = 1;
+
+	for (i = 0; i < buffer->text.size; i++) {
+		if (isspace(buffer->text.ptr[i]))
+			continue;
+		for (j = i + 1; j < buffer->text.size &&
+				!isspace(buffer->text.ptr[j]); j++)
+			; /* find the end of the word */
+
+		/* store original boundaries */
+		ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
+				buffer->orig_alloc);
+		buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
+		buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
+		buffer->orig_nr++;
+
+		/* store one word */
+		memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
+		out->ptr[out->size + j - i] = '\n';
+		out->size += j - i + 1;
+
+		i = j - 1;
+	}
 }
 
 /* this executes the word diff on the accumulated buffers */
@@ -419,22 +422,34 @@ static void diff_words_show(struct diff_words_data *diff_words)
 	xdemitcb_t ecb;
 	mmfile_t minus, plus;
 
+	/* special case: only removal */
+	if (!diff_words->plus.text.size) {
+		color_fwrite_lines(diff_words->file,
+			diff_get_color(1, DIFF_FILE_OLD),
+			diff_words->minus.text.size, diff_words->minus.text.ptr);
+		diff_words->minus.text.size = 0;
+		return;
+	}
+
+	diff_words->current_plus = diff_words->plus.text.ptr;
+
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
 	diff_words_fill(&diff_words->minus, &minus);
 	diff_words_fill(&diff_words->plus, &plus);
 	xpp.flags = XDF_NEED_MINIMAL;
-	xecfg.ctxlen = diff_words->minus.alloc + diff_words->plus.alloc;
+	xecfg.ctxlen = 0;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
 		      &xpp, &xecfg, &ecb);
 	free(minus.ptr);
 	free(plus.ptr);
+	if (diff_words->current_plus != diff_words->plus.text.ptr +
+			diff_words->plus.text.size)
+		fwrite(diff_words->current_plus,
+			diff_words->plus.text.ptr + diff_words->plus.text.size
+			- diff_words->current_plus, 1,
+			diff_words->file);
 	diff_words->minus.text.size = diff_words->plus.text.size = 0;
-
-	if (diff_words->minus.suppressed_newline) {
-		putc('\n', diff_words->file);
-		diff_words->minus.suppressed_newline = 0;
-	}
 }
 
 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
@@ -458,7 +473,9 @@ static void free_diff_words_data(struct emit_callback *ecbdata)
 			diff_words_show(ecbdata->diff_words);
 
 		free (ecbdata->diff_words->minus.text.ptr);
+		free (ecbdata->diff_words->minus.orig);
 		free (ecbdata->diff_words->plus.text.ptr);
+		free (ecbdata->diff_words->plus.orig);
 		free(ecbdata->diff_words);
 		ecbdata->diff_words = NULL;
 	}
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
new file mode 100755
index 0000000..b032bd3
--- /dev/null
+++ b/t/t4034-diff-words.sh
@@ -0,0 +1,62 @@
+#!/bin/sh
+
+test_description='word diff colors'
+
+. ./test-lib.sh
+
+test_expect_success setup '
+
+	git config diff.color.old red
+	git config diff.color.new green
+
+'
+
+decrypt_color () {
+	sed \
+		-e 's/.\[1m/<WHITE>/g' \
+		-e 's/.\[31m/<RED>/g' \
+		-e 's/.\[32m/<GREEN>/g' \
+		-e 's/.\[36m/<BROWN>/g' \
+		-e 's/.\[m/<RESET>/g'
+}
+
+cat > pre <<\EOF
+h(4)
+
+a = b + c
+EOF
+
+cat > post <<\EOF
+h(4),hh[44]
+
+a = b + c
+
+aa = a
+
+aeff = aeff * ( aaa )
+EOF
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+<RED>h(4)<RESET><GREEN>h(4),hh[44]<RESET>
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa )<RESET>
+EOF
+
+test_expect_success 'word diff with runs of whitespace' '
+
+	test_must_fail git diff --no-index --color-words pre post > output &&
+	decrypt_color < output > output.decrypted &&
+	test_cmp expect output.decrypted
+
+'
+
+test_done

Documentation/diff-options.txt, diff.c, diff.h, t/t4034-diff-words.sh

diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 1f8ce97..e546bfa 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -94,8 +94,12 @@ endif::git-format-patch[]
 	Turn off colored diff, even when the configuration file
 	gives the default to color output.
 
---color-words::
+--color-words[=regex]::
 	Show colored word diff, i.e. color words which have changed.
++
+Optionally, you can pass a regular expression that tells Git what the
+words are that you are looking for; The default is to interpret any
+stretch of non-whitespace as a word.
 
 --no-renames::
 	Turn off rename detection, even when the configuration
diff --git a/diff.c b/diff.c
index fe8b1f0..d5d7171 100644
--- a/diff.c
+++ b/diff.c
@@ -333,12 +333,14 @@ static void diff_words_append(char *line, unsigned long len,
 	len--;
 	memcpy(buffer->text.ptr + buffer->text.size, line, len);
 	buffer->text.size += len;
+	buffer->text.ptr[buffer->text.size] = '\0';
 }
 
 struct diff_words_data {
 	struct diff_words_buffer minus, plus;
 	const char *current_plus;
 	FILE *file;
+	regex_t *word_regex;
 };
 
 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
@@ -374,17 +376,49 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 	diff_words->current_plus = plus_end;
 }
 
+/* This function starts looking at *begin, and returns 0 iff a word was found. */
+static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
+		int *begin, int *end)
+{
+	if (word_regex && *begin < buffer->size) {
+		regmatch_t match[1];
+		if (!regexec(word_regex, buffer->ptr + *begin, 1, match, 0)) {
+			char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
+					'\n', match[0].rm_eo);
+			*end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
+			*begin += match[0].rm_so;
+			return *begin >= *end;
+		}
+		return -1;
+	}
+
+	/* find the next word */
+	while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
+		(*begin)++;
+	if (*begin >= buffer->size)
+		return -1;
+
+	/* find the end of the word */
+	*end = *begin + 1;
+	while (*end < buffer->size && !isspace(buffer->ptr[*end]))
+		(*end)++;
+
+	return 0;
+}
+
 /*
  * This function splits the words in buffer->text, stores the list with
  * newline separator into out, and saves the offsets of the original words
  * in buffer->orig.
  */
-static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
+static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
+		regex_t *word_regex)
 {
 	int i, j;
+	long alloc = 0;
 
 	out->size = 0;
-	out->ptr = xmalloc(buffer->text.size);
+	out->ptr = NULL;
 
 	/* fake an empty "0th" word */
 	ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
@@ -392,11 +426,8 @@ static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 	buffer->orig_nr = 1;
 
 	for (i = 0; i < buffer->text.size; i++) {
-		if (isspace(buffer->text.ptr[i]))
-			continue;
-		for (j = i + 1; j < buffer->text.size &&
-				!isspace(buffer->text.ptr[j]); j++)
-			; /* find the end of the word */
+		if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
+			return;
 
 		/* store original boundaries */
 		ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
@@ -406,6 +437,7 @@ static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 		buffer->orig_nr++;
 
 		/* store one word */
+		ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
 		memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
 		out->ptr[out->size + j - i] = '\n';
 		out->size += j - i + 1;
@@ -435,9 +467,10 @@ static void diff_words_show(struct diff_words_data *diff_words)
 
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
-	diff_words_fill(&diff_words->minus, &minus);
-	diff_words_fill(&diff_words->plus, &plus);
+	diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
+	diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
 	xpp.flags = XDF_NEED_MINIMAL;
+	/* as only the hunk header will be parsed, we need a 0-context */
 	xecfg.ctxlen = 0;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
 		      &xpp, &xecfg, &ecb);
@@ -476,6 +509,7 @@ static void free_diff_words_data(struct emit_callback *ecbdata)
 		free (ecbdata->diff_words->minus.orig);
 		free (ecbdata->diff_words->plus.text.ptr);
 		free (ecbdata->diff_words->plus.orig);
+		free(ecbdata->diff_words->word_regex);
 		free(ecbdata->diff_words);
 		ecbdata->diff_words = NULL;
 	}
@@ -1498,6 +1532,14 @@ static void builtin_diff(const char *name_a,
 			ecbdata.diff_words =
 				xcalloc(1, sizeof(struct diff_words_data));
 			ecbdata.diff_words->file = o->file;
+			if (o->word_regex) {
+				ecbdata.diff_words->word_regex = (regex_t *)
+					xmalloc(sizeof(regex_t));
+				if (regcomp(ecbdata.diff_words->word_regex,
+						o->word_regex, REG_EXTENDED))
+					die ("Invalid regular expression: %s",
+							o->word_regex);
+			}
 		}
 		xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
 			      &xpp, &xecfg, &ecb);
@@ -2513,6 +2555,10 @@ int diff_opt_parse(struct diff_options *options, const char **av, int ac)
 		DIFF_OPT_CLR(options, COLOR_DIFF);
 	else if (!strcmp(arg, "--color-words"))
 		options->flags |= DIFF_OPT_COLOR_DIFF | DIFF_OPT_COLOR_DIFF_WORDS;
+	else if (!prefixcmp(arg, "--color-words=")) {
+		options->flags |= DIFF_OPT_COLOR_DIFF | DIFF_OPT_COLOR_DIFF_WORDS;
+		options->word_regex = arg + 14;
+	}
 	else if (!strcmp(arg, "--exit-code"))
 		DIFF_OPT_SET(options, EXIT_WITH_STATUS);
 	else if (!strcmp(arg, "--quiet"))
diff --git a/diff.h b/diff.h
index 4d5a327..23cd90c 100644
--- a/diff.h
+++ b/diff.h
@@ -98,6 +98,7 @@ struct diff_options {
 
 	int stat_width;
 	int stat_name_width;
+	const char *word_regex;
 
 	/* this is set by diffcore for DIFF_FORMAT_PATCH */
 	int found_changes;
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index b032bd3..0ed7e53 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -59,4 +59,28 @@ test_expect_success 'word diff with runs of whitespace' '
 
 '
 
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+h(4),<GREEN>hh<RESET>[44]
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa )<RESET>
+EOF
+
+test_expect_success 'word diff with a regular expression' '
+
+	test_must_fail git diff --no-index --color-words='[a-z]+' \
+		pre post > output &&
+	decrypt_color < output > output.decrypted &&
+	test_cmp expect output.decrypted
+
+'
+
 test_done

diff.c, t/t4034-diff-words.sh

diff --git a/diff.c b/diff.c
index d5d7171..1408717 100644
--- a/diff.c
+++ b/diff.c
@@ -384,7 +384,7 @@ static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
 		regmatch_t match[1];
 		if (!regexec(word_regex, buffer->ptr + *begin, 1, match, 0)) {
 			char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
-					'\n', match[0].rm_eo);
+					'\n', match[0].rm_eo - match[0].rm_so);
 			*end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
 			*begin += match[0].rm_so;
 			return *begin >= *end;
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 0ed7e53..1137131 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -83,4 +83,24 @@ test_expect_success 'word diff with a regular expression' '
 
 '
 
+echo 'aaa (aaa)' > pre
+echo 'aaa (aaa) aaa' > post
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index c29453b..be22f37 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1 +1 @@<RESET>
+aaa (aaa)<GREEN> aaa<RESET>
+EOF
+
+test_expect_success "Teemo's example" '
+
+	test_must_fail git diff --no-index --color-words='a+' pre post > output &&
+	decrypt_color < output > output.decrypted &&
+	test_cmp expect output.decrypted
+
+'
+
 test_done

color.c, color.h

diff --git a/color.c b/color.c
index fc0b72a..d4ae83f 100644
--- a/color.c
+++ b/color.c
@@ -191,3 +191,31 @@ int color_fprintf_ln(FILE *fp, const char *color, const char *fmt, ...)
 	va_end(args);
 	return r;
 }
+
+/*
+ * This function splits the buffer by newlines and colors the lines individually.
+ *
+ * Returns 0 on success.
+ */
+int color_fwrite_lines(FILE *fp, const char *color,
+		size_t count, const char *buf)
+{
+	if (!*color)
+		return fwrite(buf, count, 1, fp) != 1;
+	while (count) {
+		char *p = memchr(buf, '\n', count);
+		if (p != buf && (fputs(color, fp) < 0 ||
+				fwrite(buf, p ? p - buf : count, 1, fp) != 1 ||
+				fputs(COLOR_RESET, fp) < 0))
+			return -1;
+		if (!p)
+			return 0;
+		if (fputc('\n', fp) < 0)
+			return -1;
+		count -= p + 1 - buf;
+		buf = p + 1;
+	}
+	return 0;
+}
+
+
diff --git a/color.h b/color.h
index 6cf5c88..cd5c985 100644
--- a/color.h
+++ b/color.h
@@ -19,5 +19,6 @@
 void color_parse(const char *var, const char *value, char *dst);
 int color_fprintf(FILE *fp, const char *color, const char *fmt, ...);
 int color_fprintf_ln(FILE *fp, const char *color, const char *fmt, ...);
+int color_fwrite_lines(FILE *fp, const char *color, size_t count, const char *buf);
 
 #endif /* COLOR_H */

diff --git a/diff.c b/diff.c
index d235482..c111eef 100644
--- a/diff.c
+++ b/diff.c
@@ -326,10 +326,7 @@ struct diff_words_buffer {
 static void diff_words_append(char *line, unsigned long len,
 		struct diff_words_buffer *buffer)
 {
-	if (buffer->text.size + len > buffer->alloc) {
-		buffer->alloc = (buffer->text.size + len) * 3 / 2;
-		buffer->text.ptr = xrealloc(buffer->text.ptr, buffer->alloc);
-	}
+	ALLOC_GROW(buffer->text.ptr, buffer->text.size + len, buffer->alloc);
 	line++;
 	len--;
 	memcpy(buffer->text.ptr + buffer->text.size, line, len);
@@ -398,6 +395,22 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 	}
 }
 
+/*
+ * This function splits the words in buffer->text, and stores the list with
+ * newline separator into out.
+ */
+static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
+{
+	int i;
+	out->size = buffer->text.size;
+	out->ptr = xmalloc(out->size);
+	memcpy(out->ptr, buffer->text.ptr, out->size);
+	for (i = 0; i < out->size; i++)
+		if (isspace(out->ptr[i]))
+			out->ptr[i] = '\n';
+	buffer->current = 0;
+}
+
 /* this executes the word diff on the accumulated buffers */
 static void diff_words_show(struct diff_words_data *diff_words)
 {
@@ -405,26 +418,11 @@ static void diff_words_show(struct diff_words_data *diff_words)
 	xdemitconf_t xecfg;
 	xdemitcb_t ecb;
 	mmfile_t minus, plus;
-	int i;
 
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
-	minus.size = diff_words->minus.text.size;
-	minus.ptr = xmalloc(minus.size);
-	memcpy(minus.ptr, diff_words->minus.text.ptr, minus.size);
-	for (i = 0; i < minus.size; i++)
-		if (isspace(minus.ptr[i]))
-			minus.ptr[i] = '\n';
-	diff_words->minus.current = 0;
-
-	plus.size = diff_words->plus.text.size;
-	plus.ptr = xmalloc(plus.size);
-	memcpy(plus.ptr, diff_words->plus.text.ptr, plus.size);
-	for (i = 0; i < plus.size; i++)
-		if (isspace(plus.ptr[i]))
-			plus.ptr[i] = '\n';
-	diff_words->plus.current = 0;
-
+	diff_words_fill(&diff_words->minus, &minus);
+	diff_words_fill(&diff_words->plus, &plus);
 	xpp.flags = XDF_NEED_MINIMAL;
 	xecfg.ctxlen = diff_words->minus.alloc + diff_words->plus.alloc;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,

diff.c, t/t4034-diff-words.sh

diff --git a/diff.c b/diff.c
index c111eef..37c886a 100644
--- a/diff.c
+++ b/diff.c
@@ -319,8 +319,10 @@ static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
 struct diff_words_buffer {
 	mmfile_t text;
 	long alloc;
-	long current; /* output pointer */
-	int suppressed_newline;
+	struct diff_words_orig {
+		const char *begin, *end;
+	} *orig;
+	int orig_nr, orig_alloc;
 };
 
 static void diff_words_append(char *line, unsigned long len,
@@ -335,80 +337,89 @@ static void diff_words_append(char *line, unsigned long len,
 
 struct diff_words_data {
 	struct diff_words_buffer minus, plus;
+	const char *current_plus;
 	FILE *file;
 };
 
-static void print_word(FILE *file, struct diff_words_buffer *buffer, int len, int color,
-		int suppress_newline)
+static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 {
-	const char *ptr;
-	int eol = 0;
+	struct diff_words_data *diff_words = priv;
+	int minus_first, minus_len, plus_first, plus_len;
+	const char *minus_begin, *minus_end, *plus_begin, *plus_end;
 
-	if (len == 0)
+	if (line[0] != '@' || parse_hunk_header(line, len,
+			&minus_first, &minus_len, &plus_first, &plus_len))
 		return;
 
-	ptr  = buffer->text.ptr + buffer->current;
-	buffer->current += len;
+	/* POSIX requires that first be decremented by one if len == 0... */
+	if (minus_len) {
+		minus_begin = diff_words->minus.orig[minus_first].begin;
+		minus_end =
+			diff_words->minus.orig[minus_first + minus_len - 1].end;
+	} else
+		minus_begin = minus_end =
+			diff_words->minus.orig[minus_first].end;
 
-	if (ptr[len - 1] == '\n') {
-		eol = 1;
-		len--;
-	}
+	if (plus_len) {
+		plus_begin = diff_words->plus.orig[plus_first].begin;
+		plus_end = diff_words->plus.orig[plus_first + plus_len - 1].end;
+	} else
+		plus_begin = plus_end = diff_words->plus.orig[plus_first].end;
 
-	fputs(diff_get_color(1, color), file);
-	fwrite(ptr, len, 1, file);
-	fputs(diff_get_color(1, DIFF_RESET), file);
+	if (diff_words->current_plus != plus_begin)
+		fwrite(diff_words->current_plus,
+				plus_begin - diff_words->current_plus, 1,
+				diff_words->file);
+	if (minus_begin != minus_end)
+		color_fwrite_lines(diff_words->file,
+				diff_get_color(1, DIFF_FILE_OLD),
+				minus_end - minus_begin, minus_begin);
+	if (plus_begin != plus_end)
+		color_fwrite_lines(diff_words->file,
+				diff_get_color(1, DIFF_FILE_NEW),
+				plus_end - plus_begin, plus_begin);
 
-	if (eol) {
-		if (suppress_newline)
-			buffer->suppressed_newline = 1;
-		else
-			putc('\n', file);
-	}
-}
-
-static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
-{
-	struct diff_words_data *diff_words = priv;
-
-	if (diff_words->minus.suppressed_newline) {
-		if (line[0] != '+')
-			putc('\n', diff_words->file);
-		diff_words->minus.suppressed_newline = 0;
-	}
-
-	len--;
-	switch (line[0]) {
-		case '-':
-			print_word(diff_words->file,
-				   &diff_words->minus, len, DIFF_FILE_OLD, 1);
-			break;
-		case '+':
-			print_word(diff_words->file,
-				   &diff_words->plus, len, DIFF_FILE_NEW, 0);
-			break;
-		case ' ':
-			print_word(diff_words->file,
-				   &diff_words->plus, len, DIFF_PLAIN, 0);
-			diff_words->minus.current += len;
-			break;
-	}
+	diff_words->current_plus = plus_end;
 }
 
 /*
- * This function splits the words in buffer->text, and stores the list with
- * newline separator into out.
+ * This function splits the words in buffer->text, stores the list with
+ * newline separator into out, and saves the offsets of the original words
+ * in buffer->orig.
  */
 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 {
-	int i;
-	out->size = buffer->text.size;
-	out->ptr = xmalloc(out->size);
-	memcpy(out->ptr, buffer->text.ptr, out->size);
-	for (i = 0; i < out->size; i++)
-		if (isspace(out->ptr[i]))
-			out->ptr[i] = '\n';
-	buffer->current = 0;
+	int i, j;
+
+	out->size = 0;
+	out->ptr = xmalloc(buffer->text.size);
+
+	/* fake an empty "0th" word */
+	ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
+	buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
+	buffer->orig_nr = 1;
+
+	for (i = 0; i < buffer->text.size; i++) {
+		if (isspace(buffer->text.ptr[i]))
+			continue;
+		for (j = i + 1; j < buffer->text.size &&
+				!isspace(buffer->text.ptr[j]); j++)
+			; /* find the end of the word */
+
+		/* store original boundaries */
+		ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
+				buffer->orig_alloc);
+		buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
+		buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
+		buffer->orig_nr++;
+
+		/* store one word */
+		memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
+		out->ptr[out->size + j - i] = '\n';
+		out->size += j - i + 1;
+
+		i = j - 1;
+	}
 }
 
 /* this executes the word diff on the accumulated buffers */
@@ -419,22 +430,34 @@ static void diff_words_show(struct diff_words_data *diff_words)
 	xdemitcb_t ecb;
 	mmfile_t minus, plus;
 
+	/* special case: only removal */
+	if (!diff_words->plus.text.size) {
+		color_fwrite_lines(diff_words->file,
+			diff_get_color(1, DIFF_FILE_OLD),
+			diff_words->minus.text.size, diff_words->minus.text.ptr);
+		diff_words->minus.text.size = 0;
+		return;
+	}
+
+	diff_words->current_plus = diff_words->plus.text.ptr;
+
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
 	diff_words_fill(&diff_words->minus, &minus);
 	diff_words_fill(&diff_words->plus, &plus);
 	xpp.flags = XDF_NEED_MINIMAL;
-	xecfg.ctxlen = diff_words->minus.alloc + diff_words->plus.alloc;
+	xecfg.ctxlen = 0;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
 		      &xpp, &xecfg, &ecb);
 	free(minus.ptr);
 	free(plus.ptr);
+	if (diff_words->current_plus != diff_words->plus.text.ptr +
+			diff_words->plus.text.size)
+		fwrite(diff_words->current_plus,
+			diff_words->plus.text.ptr + diff_words->plus.text.size
+			- diff_words->current_plus, 1,
+			diff_words->file);
 	diff_words->minus.text.size = diff_words->plus.text.size = 0;
-
-	if (diff_words->minus.suppressed_newline) {
-		putc('\n', diff_words->file);
-		diff_words->minus.suppressed_newline = 0;
-	}
 }
 
 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
@@ -458,7 +481,9 @@ static void free_diff_words_data(struct emit_callback *ecbdata)
 			diff_words_show(ecbdata->diff_words);
 
 		free (ecbdata->diff_words->minus.text.ptr);
+		free (ecbdata->diff_words->minus.orig);
 		free (ecbdata->diff_words->plus.text.ptr);
+		free (ecbdata->diff_words->plus.orig);
 		free(ecbdata->diff_words);
 		ecbdata->diff_words = NULL;
 	}
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
new file mode 100755
index 0000000..b22195f
--- /dev/null
+++ b/t/t4034-diff-words.sh
@@ -0,0 +1,66 @@
+#!/bin/sh
+
+test_description='word diff colors'
+
+. ./test-lib.sh
+
+test_expect_success setup '
+
+	git config diff.color.old red
+	git config diff.color.new green
+
+'
+
+decrypt_color () {
+	sed \
+		-e 's/.\[1m/<WHITE>/g' \
+		-e 's/.\[31m/<RED>/g' \
+		-e 's/.\[32m/<GREEN>/g' \
+		-e 's/.\[36m/<BROWN>/g' \
+		-e 's/.\[m/<RESET>/g'
+}
+
+word_diff () {
+	test_must_fail git diff --no-index "$@" pre post > output &&
+	decrypt_color < output > output.decrypted &&
+	test_cmp expect output.decrypted
+}
+
+cat > pre <<\EOF
+h(4)
+
+a = b + c
+EOF
+
+cat > post <<\EOF
+h(4),hh[44]
+
+a = b + c
+
+aa = a
+
+aeff = aeff * ( aaa )
+EOF
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+<RED>h(4)<RESET><GREEN>h(4),hh[44]<RESET>
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa )<RESET>
+EOF
+
+test_expect_success 'word diff with runs of whitespace' '
+
+	word_diff --color-words
+
+'
+
+test_done

Documentation/diff-options.txt, diff.c, diff.h, t/t4034-diff-words.sh

diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 43793d7..2c1fa4b 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -91,8 +91,12 @@ endif::git-format-patch[]
 	Turn off colored diff, even when the configuration file
 	gives the default to color output.
 
---color-words::
+--color-words[=regex]::
 	Show colored word diff, i.e. color words which have changed.
++
+Optionally, you can pass a regular expression that tells Git what the
+words are that you are looking for; The default is to interpret any
+stretch of non-whitespace as a word.
 
 --no-renames::
 	Turn off rename detection, even when the configuration
diff --git a/diff.c b/diff.c
index 37c886a..9fb3d0d 100644
--- a/diff.c
+++ b/diff.c
@@ -333,12 +333,14 @@ static void diff_words_append(char *line, unsigned long len,
 	len--;
 	memcpy(buffer->text.ptr + buffer->text.size, line, len);
 	buffer->text.size += len;
+	buffer->text.ptr[buffer->text.size] = '\0';
 }
 
 struct diff_words_data {
 	struct diff_words_buffer minus, plus;
 	const char *current_plus;
 	FILE *file;
+	regex_t *word_regex;
 };
 
 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
@@ -382,17 +384,49 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 	diff_words->current_plus = plus_end;
 }
 
+/* This function starts looking at *begin, and returns 0 iff a word was found. */
+static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
+		int *begin, int *end)
+{
+	if (word_regex && *begin < buffer->size) {
+		regmatch_t match[1];
+		if (!regexec(word_regex, buffer->ptr + *begin, 1, match, 0)) {
+			char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
+					'\n', match[0].rm_eo - match[0].rm_so);
+			*end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
+			*begin += match[0].rm_so;
+			return *begin >= *end;
+		}
+		return -1;
+	}
+
+	/* find the next word */
+	while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
+		(*begin)++;
+	if (*begin >= buffer->size)
+		return -1;
+
+	/* find the end of the word */
+	*end = *begin + 1;
+	while (*end < buffer->size && !isspace(buffer->ptr[*end]))
+		(*end)++;
+
+	return 0;
+}
+
 /*
  * This function splits the words in buffer->text, stores the list with
  * newline separator into out, and saves the offsets of the original words
  * in buffer->orig.
  */
-static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
+static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
+		regex_t *word_regex)
 {
 	int i, j;
+	long alloc = 0;
 
 	out->size = 0;
-	out->ptr = xmalloc(buffer->text.size);
+	out->ptr = NULL;
 
 	/* fake an empty "0th" word */
 	ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
@@ -400,11 +434,8 @@ static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 	buffer->orig_nr = 1;
 
 	for (i = 0; i < buffer->text.size; i++) {
-		if (isspace(buffer->text.ptr[i]))
-			continue;
-		for (j = i + 1; j < buffer->text.size &&
-				!isspace(buffer->text.ptr[j]); j++)
-			; /* find the end of the word */
+		if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
+			return;
 
 		/* store original boundaries */
 		ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
@@ -414,6 +445,7 @@ static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 		buffer->orig_nr++;
 
 		/* store one word */
+		ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
 		memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
 		out->ptr[out->size + j - i] = '\n';
 		out->size += j - i + 1;
@@ -443,9 +475,10 @@ static void diff_words_show(struct diff_words_data *diff_words)
 
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
-	diff_words_fill(&diff_words->minus, &minus);
-	diff_words_fill(&diff_words->plus, &plus);
+	diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
+	diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
 	xpp.flags = XDF_NEED_MINIMAL;
+	/* as only the hunk header will be parsed, we need a 0-context */
 	xecfg.ctxlen = 0;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
 		      &xpp, &xecfg, &ecb);
@@ -484,6 +517,7 @@ static void free_diff_words_data(struct emit_callback *ecbdata)
 		free (ecbdata->diff_words->minus.orig);
 		free (ecbdata->diff_words->plus.text.ptr);
 		free (ecbdata->diff_words->plus.orig);
+		free(ecbdata->diff_words->word_regex);
 		free(ecbdata->diff_words);
 		ecbdata->diff_words = NULL;
 	}
@@ -1506,6 +1540,14 @@ static void builtin_diff(const char *name_a,
 			ecbdata.diff_words =
 				xcalloc(1, sizeof(struct diff_words_data));
 			ecbdata.diff_words->file = o->file;
+			if (o->word_regex) {
+				ecbdata.diff_words->word_regex = (regex_t *)
+					xmalloc(sizeof(regex_t));
+				if (regcomp(ecbdata.diff_words->word_regex,
+						o->word_regex, REG_EXTENDED))
+					die ("Invalid regular expression: %s",
+							o->word_regex);
+			}
 		}
 		xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
 			      &xpp, &xecfg, &ecb);
@@ -2517,6 +2559,10 @@ int diff_opt_parse(struct diff_options *options, const char **av, int ac)
 		DIFF_OPT_CLR(options, COLOR_DIFF);
 	else if (!strcmp(arg, "--color-words"))
 		options->flags |= DIFF_OPT_COLOR_DIFF | DIFF_OPT_COLOR_DIFF_WORDS;
+	else if (!prefixcmp(arg, "--color-words=")) {
+		options->flags |= DIFF_OPT_COLOR_DIFF | DIFF_OPT_COLOR_DIFF_WORDS;
+		options->word_regex = arg + 14;
+	}
 	else if (!strcmp(arg, "--exit-code"))
 		DIFF_OPT_SET(options, EXIT_WITH_STATUS);
 	else if (!strcmp(arg, "--quiet"))
diff --git a/diff.h b/diff.h
index 4d5a327..23cd90c 100644
--- a/diff.h
+++ b/diff.h
@@ -98,6 +98,7 @@ struct diff_options {
 
 	int stat_width;
 	int stat_name_width;
+	const char *word_regex;
 
 	/* this is set by diffcore for DIFF_FORMAT_PATCH */
 	int found_changes;
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index b22195f..4873486 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -63,4 +63,61 @@ test_expect_success 'word diff with runs of whitespace' '
 
 '
 
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+h(4),<GREEN>hh<RESET>[44]
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa<RESET> )
+EOF
+
+test_expect_success 'word diff with a regular expression' '
+
+	word_diff --color-words="[a-z]+"
+
+'
+
+echo 'aaa (aaa)' > pre
+echo 'aaa (aaa) aaa' > post
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index c29453b..be22f37 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1 +1 @@<RESET>
+aaa (aaa) <GREEN>aaa<RESET>
+EOF
+
+test_expect_success 'test parsing words for newline' '
+
+	word_diff --color-words="a+"
+
+'
+
+echo '(:' > pre
+echo '(' > post
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 289cb9d..2d06f37 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1 +1 @@<RESET>
+(<RED>:<RESET>
+EOF
+
+test_expect_success 'test when words are only removed at the end' '
+
+	word_diff --color-words=.
+
+'
+
 test_done

diff --git a/diff.c b/diff.c
index 9fb3d0d..00c661f 100644
--- a/diff.c
+++ b/diff.c
@@ -1544,7 +1544,8 @@ static void builtin_diff(const char *name_a,
 				ecbdata.diff_words->word_regex = (regex_t *)
 					xmalloc(sizeof(regex_t));
 				if (regcomp(ecbdata.diff_words->word_regex,
-						o->word_regex, REG_EXTENDED))
+						o->word_regex,
+						REG_EXTENDED | REG_NEWLINE))
 					die ("Invalid regular expression: %s",
 							o->word_regex);
 			}

diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 2c1fa4b..8689a92 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -91,12 +91,17 @@ endif::git-format-patch[]
 	Turn off colored diff, even when the configuration file
 	gives the default to color output.
 
---color-words[=regex]::
-	Show colored word diff, i.e. color words which have changed.
+--color-words[=<regex>]::
+	Show colored word diff, i.e., color words which have changed.
+	By default, words are separated by whitespace.
 +
-Optionally, you can pass a regular expression that tells Git what the
-words are that you are looking for; The default is to interpret any
-stretch of non-whitespace as a word.
+When a <regex> is specified, every non-overlapping match of the
+<regex> is considered a word.  Anything between these matches is
+considered whitespace and ignored(!) for the purposes of finding
+differences.  You may want to append `|[^[:space:]]` to your regular
+expression to make sure that it matches all non-whitespace characters.
+A match that contains a newline is silently truncated(!) at the
+newline.
 
 --no-renames::
 	Turn off rename detection, even when the configuration

Documentation/diff-options.txt, Documentation/gitattributes.txt, diff.c, t/t4034-diff-words.sh, userdiff.c, userdiff.h

diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 8689a92..1edb82e 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -102,6 +102,10 @@ differences.  You may want to append `|[^[:space:]]` to your regular
 expression to make sure that it matches all non-whitespace characters.
 A match that contains a newline is silently truncated(!) at the
 newline.
++
+The regex can also be set via a diff driver, see
+linkgit:gitattributes[1]; giving it explicitly overrides any diff
+driver setting.
 
 --no-renames::
 	Turn off rename detection, even when the configuration
diff --git a/Documentation/gitattributes.txt b/Documentation/gitattributes.txt
index 8af22ec..ba3ba12 100644
--- a/Documentation/gitattributes.txt
+++ b/Documentation/gitattributes.txt
@@ -317,6 +317,8 @@ patterns are available:
 
 - `bibtex` suitable for files with BibTeX coded references.
 
+- `cpp` suitable for source code in the C and C++ languages.
+
 - `html` suitable for HTML/XHTML documents.
 
 - `java` suitable for source code in the Java language.
@@ -334,6 +336,25 @@ patterns are available:
 - `tex` suitable for source code for LaTeX documents.
 
 
+Customizing word diff
+^^^^^^^^^^^^^^^^^^^^^
+
+You can customize the rules that `git diff --color-words` uses to
+split words in a line, by specifying an appropriate regular expression
+in the "diff.*.wordregex" configuration variable.  For example, in TeX
+a backslash followed by a sequence of letters forms a command, but
+several such commands can be run together without intervening
+whitespace.  To separate them, use a regular expression such as
+
+------------------------
+[diff "tex"]
+	wordregex = "\\\\[a-zA-Z]+|[{}]|\\\\.|[^\\{}[:space:]]+"
+------------------------
+
+A built-in pattern is provided for all languages listed in the
+previous section.
+
+
 Performing text diffs of binary files
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
diff --git a/diff.c b/diff.c
index 00c661f..9fcde96 100644
--- a/diff.c
+++ b/diff.c
@@ -1380,6 +1380,12 @@ int diff_filespec_is_binary(struct diff_filespec *one)
 	return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
 }
 
+static const char *userdiff_word_regex(struct diff_filespec *one)
+{
+	diff_filespec_load_driver(one);
+	return one->driver->word_regex;
+}
+
 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
 {
 	if (!options->a_prefix)
@@ -1540,6 +1546,10 @@ static void builtin_diff(const char *name_a,
 			ecbdata.diff_words =
 				xcalloc(1, sizeof(struct diff_words_data));
 			ecbdata.diff_words->file = o->file;
+			if (!o->word_regex)
+				o->word_regex = userdiff_word_regex(one);
+			if (!o->word_regex)
+				o->word_regex = userdiff_word_regex(two);
 			if (o->word_regex) {
 				ecbdata.diff_words->word_regex = (regex_t *)
 					xmalloc(sizeof(regex_t));
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 4873486..744221b 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -84,6 +84,41 @@ test_expect_success 'word diff with a regular expression' '
 
 '
 
+test_expect_success 'set a diff driver' '
+	git config diff.testdriver.wordregex "[^[:space:]]" &&
+	cat <<EOF > .gitattributes
+pre diff=testdriver
+post diff=testdriver
+EOF
+'
+
+test_expect_success 'option overrides default' '
+
+	word_diff --color-words="[a-z]+"
+
+'
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+h(4)<GREEN>,hh[44]<RESET>
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa )<RESET>
+EOF
+
+test_expect_success 'use default supplied by driver' '
+
+	word_diff --color-words
+
+'
+
 echo 'aaa (aaa)' > pre
 echo 'aaa (aaa) aaa' > post
 
@@ -100,6 +135,7 @@ test_expect_success 'test parsing words for newline' '
 
 	word_diff --color-words="a+"
 
+
 '
 
 echo '(:' > pre
diff --git a/userdiff.c b/userdiff.c
index 3681062..2b55509 100644
--- a/userdiff.c
+++ b/userdiff.c
@@ -6,14 +6,20 @@
 static int ndrivers;
 static int drivers_alloc;
 
-#define FUNCNAME(name, pattern) \
-	{ name, NULL, -1, { pattern, REG_EXTENDED } }
+#define PATTERNS(name, pattern, wordregex)			\
+	{ name, NULL, -1, { pattern, REG_EXTENDED }, wordregex }
 static struct userdiff_driver builtin_drivers[] = {
-FUNCNAME("html", "^[ \t]*(<[Hh][1-6][ \t].*>.*)$"),
-FUNCNAME("java",
+PATTERNS("html", "^[ \t]*(<[Hh][1-6][ \t].*>.*)$",
+	 "[^<>= \t]+|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("java",
 	 "!^[ \t]*(catch|do|for|if|instanceof|new|return|switch|throw|while)\n"
-	 "^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\\([^;]*)$"),
-FUNCNAME("objc",
+	 "^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\\([^;]*)$",
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[fFlL]?|0[xXbB]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]="
+	 "|--|\\+\\+|<<=?|>>>?=?|&&|\\|\\|"
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("objc",
 	 /* Negate C statements that can look like functions */
 	 "!^[ \t]*(do|for|if|else|return|switch|while)\n"
 	 /* Objective-C methods */
@@ -21,20 +27,60 @@
 	 /* C functions */
 	 "^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\\([^;]*)$\n"
 	 /* Objective-C class/protocol definitions */
-	 "^(@(implementation|interface|protocol)[ \t].*)$"),
-FUNCNAME("pascal",
+	 "^(@(implementation|interface|protocol)[ \t].*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[fFlL]?|0[xXbB]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]=|--|\\+\\+|<<=?|>>=?|&&|\\|\\||::|->"
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("pascal",
 	 "^((procedure|function|constructor|destructor|interface|"
 		"implementation|initialization|finalization)[ \t]*.*)$"
 	 "\n"
-	 "^(.*=[ \t]*(class|record).*)$"),
-FUNCNAME("php", "^[\t ]*((function|class).*)"),
-FUNCNAME("python", "^[ \t]*((class|def)[ \t].*)$"),
-FUNCNAME("ruby", "^[ \t]*((class|module|def)[ \t].*)$"),
-FUNCNAME("bibtex", "(@[a-zA-Z]{1,}[ \t]*\\{{0,1}[ \t]*[^ \t\"@',\\#}{~%]*).*$"),
-FUNCNAME("tex", "^(\\\\((sub)*section|chapter|part)\\*{0,1}\\{.*)$"),
+	 "^(.*=[ \t]*(class|record).*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+|0[xXbB]?[0-9a-fA-F]+"
+	 "|<>|<=|>=|:=|\\.\\."
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("php", "^[\t ]*((function|class).*)",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+|0[xXbB]?[0-9a-fA-F]+"
+	 "|[-+*/<>%&^|=!.]=|--|\\+\\+|<<=?|>>=?|===|&&|\\|\\||::|->"
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("python", "^[ \t]*((class|def)[ \t].*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[jJlL]?|0[xX]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]=|//=?|<<=?|>>=?|\\*\\*=?"
+	 "|[^[:space:]|[\x80-\xff]+"),
+	 /* -- */
+PATTERNS("ruby", "^[ \t]*((class|module|def)[ \t].*)$",
+	 /* -- */
+	 "(@|@@|\\$)?[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+|0[xXbB]?[0-9a-fA-F]+|\\?(\\\\C-)?(\\\\M-)?."
+	 "|//=?|[-+*/<>%&^|=!]=|<<=?|>>=?|===|\\.{1,3}|::|[!=]~"
+	 "|[^[:space:]|[\x80-\xff]+"),
+PATTERNS("bibtex", "(@[a-zA-Z]{1,}[ \t]*\\{{0,1}[ \t]*[^ \t\"@',\\#}{~%]*).*$",
+	 "[={}\"]|[^={}\" \t]+"),
+PATTERNS("tex", "^(\\\\((sub)*section|chapter|part)\\*{0,1}\\{.*)$",
+	 "\\\\[a-zA-Z@]+|\\\\.|[a-zA-Z0-9\x80-\xff]+|[^[:space:]]"),
+PATTERNS("cpp",
+	 /* Jump targets or access declarations */
+	 "!^[ \t]*[A-Za-z_][A-Za-z_0-9]*:.*$\n"
+	 /* C/++ functions/methods at top level */
+	 "^([A-Za-z_][A-Za-z_0-9]*([ \t]+[A-Za-z_][A-Za-z_0-9]*([ \t]*::[ \t]*[^[:space:]]+)?){1,}[ \t]*\\([^;]*)$\n"
+	 /* compound type at top level */
+	 "^((struct|class|enum)[^;]*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[fFlL]?|0[xXbB]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]=|--|\\+\\+|<<=?|>>=?|&&|\\|\\||::|->"
+	 "|[^[:space:]]|[\x80-\xff]+"),
 { "default", NULL, -1, { NULL, 0 } },
 };
-#undef FUNCNAME
+#undef PATTERNS
 
 static struct userdiff_driver driver_true = {
 	"diff=true",
@@ -134,6 +180,8 @@ int userdiff_config(const char *k, const char *v)
 		return parse_string(&drv->external, k, v);
 	if ((drv = parse_driver(k, v, "textconv")))
 		return parse_string(&drv->textconv, k, v);
+	if ((drv = parse_driver(k, v, "wordregex")))
+		return parse_string(&drv->word_regex, k, v);
 
 	return 0;
 }
diff --git a/userdiff.h b/userdiff.h
index ba29457..c315159 100644
--- a/userdiff.h
+++ b/userdiff.h
@@ -11,6 +11,7 @@ struct userdiff_driver {
 	const char *external;
 	int binary;
 	struct userdiff_funcname funcname;
+	const char *word_regex;
 	const char *textconv;
 };
 

diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 744221b..6ebce9d 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -63,7 +63,7 @@ test_expect_success 'word diff with runs of whitespace' '
 
 '
 
-cat > expect <<\EOF
+cat > expect.letter-runs-are-words <<\EOF
 <WHITE>diff --git a/pre b/post<RESET>
 <WHITE>index 330b04f..5ed8eff 100644<RESET>
 <WHITE>--- a/pre<RESET>
@@ -77,6 +77,7 @@ a = b + c<RESET>
 
 <GREEN>aeff = aeff * ( aaa<RESET> )
 EOF
+cp expect.letter-runs-are-words expect
 
 test_expect_success 'word diff with a regular expression' '
 
@@ -84,21 +85,11 @@ test_expect_success 'word diff with a regular expression' '
 
 '
 
-test_expect_success 'set a diff driver' '
-	git config diff.testdriver.wordregex "[^[:space:]]" &&
-	cat <<EOF > .gitattributes
-pre diff=testdriver
-post diff=testdriver
-EOF
-'
-
-test_expect_success 'option overrides default' '
-
-	word_diff --color-words="[a-z]+"
-
+test_expect_success 'add configuration for default regex' '
+	git config diff.color-words "[^[:space:]]"
 '
 
-cat > expect <<\EOF
+cat > expect.non-whitespace-is-word <<\EOF
 <WHITE>diff --git a/pre b/post<RESET>
 <WHITE>index 330b04f..5ed8eff 100644<RESET>
 <WHITE>--- a/pre<RESET>
@@ -112,6 +103,37 @@ a = b + c<RESET>
 
 <GREEN>aeff = aeff * ( aaa )<RESET>
 EOF
+cp expect.non-whitespace-is-word expect
+
+test_expect_failure 'use default supplied by config' '
+
+	word_diff --color-words
+
+'
+
+cp expect.letter-runs-are-words expect
+
+test_expect_success 'option overrides config-default' '
+
+	word_diff --color-words="[a-z]+"
+
+'
+
+test_expect_success 'set a diff driver' '
+	git config diff.testdriver.wordregex "[^[:space:]]" &&
+	cat <<EOF > .gitattributes
+pre diff=testdriver
+post diff=testdriver
+EOF
+'
+
+test_expect_success 'option overrides default' '
+
+	word_diff --color-words="[a-z]+"
+
+'
+
+cp expect.non-whitespace-is-word expect
 
 test_expect_success 'use default supplied by driver' '
 

diff.c, t/t4034-diff-words.sh

diff --git a/diff.c b/diff.c
index 9fcde96..c53e1d1 100644
--- a/diff.c
+++ b/diff.c
@@ -23,6 +23,7 @@ static int diff_detect_rename_default;
 static int diff_rename_limit_default = 200;
 static int diff_suppress_blank_empty;
 int diff_use_color_default = -1;
+static const char *diff_color_words_cfg = NULL;
 static const char *external_diff_cmd_cfg;
 int diff_auto_refresh_index = 1;
 static int diff_mnemonic_prefix;
@@ -92,6 +93,8 @@ int git_diff_ui_config(const char *var, const char *value, void *cb)
 	}
 	if (!strcmp(var, "diff.external"))
 		return git_config_string(&external_diff_cmd_cfg, var, value);
+	if (!strcmp(var, "diff.color-words"))
+		return git_config_string(&diff_color_words_cfg, var, value);
 
 	return git_diff_basic_config(var, value, cb);
 }
@@ -1550,6 +1553,8 @@ static void builtin_diff(const char *name_a,
 				o->word_regex = userdiff_word_regex(one);
 			if (!o->word_regex)
 				o->word_regex = userdiff_word_regex(two);
+			if (!o->word_regex)
+				o->word_regex = diff_color_words_cfg;
 			if (o->word_regex) {
 				ecbdata.diff_words->word_regex = (regex_t *)
 					xmalloc(sizeof(regex_t));
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 6ebce9d..a207d9e 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -105,7 +105,7 @@ a = b + c<RESET>
 EOF
 cp expect.non-whitespace-is-word expect
 
-test_expect_failure 'use default supplied by config' '
+test_expect_success 'use default supplied by config' '
 
 	word_diff --color-words
 

Documentation/config.txt, diff.c, t/t4029-diff-trailing-space.sh

diff --git a/Documentation/config.txt b/Documentation/config.txt
index c92e7e6..4f0a0b1 100644
--- a/Documentation/config.txt
+++ b/Documentation/config.txt
@@ -652,7 +652,7 @@ diff.renames::
 	will enable basic rename detection.  If set to "copies" or
 	"copy", it will detect copies, as well.
 
-diff.suppress-blank-empty::
+diff.suppressBlankEmpty::
 	A boolean to inhibit the standard behavior of printing a space
 	before each empty output line. Defaults to false.
 
diff --git a/diff.c b/diff.c
index c6a992d..0100b59 100644
--- a/diff.c
+++ b/diff.c
@@ -118,7 +118,9 @@ int git_diff_basic_config(const char *var, const char *value, void *cb)
 	}
 
 	/* like GNU diff's --suppress-blank-empty option  */
-	if (!strcmp(var, "diff.suppress-blank-empty")) {
+	if (!strcmp(var, "diff.suppressblankempty") ||
+			/* for backwards compatibility */
+			!strcmp(var, "diff.suppress-blank-empty")) {
 		diff_suppress_blank_empty = git_config_bool(var, value);
 		return 0;
 	}
diff --git a/t/t4029-diff-trailing-space.sh b/t/t4029-diff-trailing-space.sh
index 4ca65e0..9ddbbcd 100755
--- a/t/t4029-diff-trailing-space.sh
+++ b/t/t4029-diff-trailing-space.sh
@@ -2,7 +2,7 @@
 #
 # Copyright (c) Jim Meyering
 #
-test_description='diff honors config option, diff.suppress-blank-empty'
+test_description='diff honors config option, diff.suppressBlankEmpty'
 
 . ./test-lib.sh
 
@@ -24,14 +24,14 @@ test_expect_success \
      git add f &&
      git commit -q -m. f &&
      printf "\ny\n" > f &&
-     git config --bool diff.suppress-blank-empty true &&
+     git config --bool diff.suppressBlankEmpty true &&
      git diff f > actual &&
      test_cmp exp actual &&
      perl -i.bak -p -e "s/^\$/ /" exp &&
-     git config --bool diff.suppress-blank-empty false &&
+     git config --bool diff.suppressBlankEmpty false &&
      git diff f > actual &&
      test_cmp exp actual &&
-     git config --bool --unset diff.suppress-blank-empty &&
+     git config --bool --unset diff.suppressBlankEmpty &&
      git diff f > actual &&
      test_cmp exp actual
      '

Documentation/config.txt, Documentation/diff-options.txt, diff.c, t/t4034-diff-words.sh

diff --git a/Documentation/config.txt b/Documentation/config.txt
index 7408bb2..0ca983a 100644
--- a/Documentation/config.txt
+++ b/Documentation/config.txt
@@ -639,6 +639,12 @@ diff.suppress-blank-empty::
 	A boolean to inhibit the standard behavior of printing a space
 	before each empty output line. Defaults to false.
 
+diff.wordregex::
+	A POSIX Extended Regular Expression used to determine what is a "word"
+	when performing word-by-word difference calculations.  Character
+	sequences that match the regular expression are "words", all other
+	characters are *ignorable* whitespace.
+
 fetch.unpackLimit::
 	If the number of objects fetched over the git native
 	transfer is below this
diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 1edb82e..164e2c5 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -103,9 +103,10 @@ expression to make sure that it matches all non-whitespace characters.
 A match that contains a newline is silently truncated(!) at the
 newline.
 +
-The regex can also be set via a diff driver, see
-linkgit:gitattributes[1]; giving it explicitly overrides any diff
-driver setting.
+The regex can also be set via a diff driver or configuration option, see
+linkgit:gitattributes[1] or linkgit:git-config[1].  Giving it explicitly
+overrides any diff driver or configuration setting.  Diff drivers
+override configuration settings.
 
 --no-renames::
 	Turn off rename detection, even when the configuration
diff --git a/diff.c b/diff.c
index 9fcde96..ed8b83c 100644
--- a/diff.c
+++ b/diff.c
@@ -23,6 +23,7 @@ static int diff_detect_rename_default;
 static int diff_rename_limit_default = 200;
 static int diff_suppress_blank_empty;
 int diff_use_color_default = -1;
+static const char *diff_word_regex_cfg;
 static const char *external_diff_cmd_cfg;
 int diff_auto_refresh_index = 1;
 static int diff_mnemonic_prefix;
@@ -92,6 +93,8 @@ int git_diff_ui_config(const char *var, const char *value, void *cb)
 	}
 	if (!strcmp(var, "diff.external"))
 		return git_config_string(&external_diff_cmd_cfg, var, value);
+	if (!strcmp(var, "diff.wordregex"))
+		return git_config_string(&diff_word_regex_cfg, var, value);
 
 	return git_diff_basic_config(var, value, cb);
 }
@@ -1550,6 +1553,8 @@ static void builtin_diff(const char *name_a,
 				o->word_regex = userdiff_word_regex(one);
 			if (!o->word_regex)
 				o->word_regex = userdiff_word_regex(two);
+			if (!o->word_regex)
+				o->word_regex = diff_word_regex_cfg;
 			if (o->word_regex) {
 				ecbdata.diff_words->word_regex = (regex_t *)
 					xmalloc(sizeof(regex_t));
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 744221b..6bcc153 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -77,6 +77,7 @@ a = b + c<RESET>
 
 <GREEN>aeff = aeff * ( aaa<RESET> )
 EOF
+cp expect expect.letter-runs-are-words
 
 test_expect_success 'word diff with a regular expression' '
 
@@ -92,7 +93,7 @@ post diff=testdriver
 EOF
 '
 
-test_expect_success 'option overrides default' '
+test_expect_success 'option overrides .gitattributes' '
 
 	word_diff --color-words="[a-z]+"
 
@@ -112,13 +113,53 @@ a = b + c<RESET>
 
 <GREEN>aeff = aeff * ( aaa )<RESET>
 EOF
+cp expect expect.non-whitespace-is-word
 
-test_expect_success 'use default supplied by driver' '
+test_expect_success 'use regex supplied by driver' '
 
 	word_diff --color-words
 
 '
 
+test_expect_success 'set diff.wordregex option' '
+	git config diff.wordregex "[[:alnum:]]+"
+'
+
+cp expect.letter-runs-are-words expect
+
+test_expect_success 'command-line overrides config' '
+	word_diff --color-words="[a-z]+"
+'
+
+cp expect.non-whitespace-is-word expect
+
+test_expect_success '.gitattributes override config' '
+	word_diff --color-words
+'
+
+test_expect_success 'remove diff driver regex' '
+	git config --unset diff.testdriver.wordregex
+'
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+h(4),<GREEN>hh[44<RESET>]
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa<RESET> )
+EOF
+
+test_expect_success 'use configured regex' '
+	word_diff --color-words
+'
+
 echo 'aaa (aaa)' > pre
 echo 'aaa (aaa) aaa' > post
 

Documentation/config.txt, Documentation/gitattributes.txt, t/t4034-diff-words.sh, userdiff.c

diff --git a/Documentation/config.txt b/Documentation/config.txt
index 0ca983a..332213e 100644
--- a/Documentation/config.txt
+++ b/Documentation/config.txt
@@ -639,7 +639,7 @@ diff.suppress-blank-empty::
 	A boolean to inhibit the standard behavior of printing a space
 	before each empty output line. Defaults to false.
 
-diff.wordregex::
+diff.wordRegex::
 	A POSIX Extended Regular Expression used to determine what is a "word"
 	when performing word-by-word difference calculations.  Character
 	sequences that match the regular expression are "words", all other
diff --git a/Documentation/gitattributes.txt b/Documentation/gitattributes.txt
index ba3ba12..227934f 100644
--- a/Documentation/gitattributes.txt
+++ b/Documentation/gitattributes.txt
@@ -341,14 +341,14 @@ Customizing word diff
 
 You can customize the rules that `git diff --color-words` uses to
 split words in a line, by specifying an appropriate regular expression
-in the "diff.*.wordregex" configuration variable.  For example, in TeX
+in the "diff.*.wordRegex" configuration variable.  For example, in TeX
 a backslash followed by a sequence of letters forms a command, but
 several such commands can be run together without intervening
 whitespace.  To separate them, use a regular expression such as
 
 ------------------------
 [diff "tex"]
-	wordregex = "\\\\[a-zA-Z]+|[{}]|\\\\.|[^\\{}[:space:]]+"
+	wordRegex = "\\\\[a-zA-Z]+|[{}]|\\\\.|[^\\{}[:space:]]+"
 ------------------------
 
 A built-in pattern is provided for all languages listed in the
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 6bcc153..4508eff 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -86,7 +86,7 @@ test_expect_success 'word diff with a regular expression' '
 '
 
 test_expect_success 'set a diff driver' '
-	git config diff.testdriver.wordregex "[^[:space:]]" &&
+	git config diff.testdriver.wordRegex "[^[:space:]]" &&
 	cat <<EOF > .gitattributes
 pre diff=testdriver
 post diff=testdriver
@@ -121,8 +121,8 @@ test_expect_success 'use regex supplied by driver' '
 
 '
 
-test_expect_success 'set diff.wordregex option' '
-	git config diff.wordregex "[[:alnum:]]+"
+test_expect_success 'set diff.wordRegex option' '
+	git config diff.wordRegex "[[:alnum:]]+"
 '
 
 cp expect.letter-runs-are-words expect
@@ -138,7 +138,7 @@ test_expect_success '.gitattributes override config' '
 '
 
 test_expect_success 'remove diff driver regex' '
-	git config --unset diff.testdriver.wordregex
+	git config --unset diff.testdriver.wordRegex
 '
 
 cat > expect <<\EOF
diff --git a/userdiff.c b/userdiff.c
index 2b55509..d556da9 100644
--- a/userdiff.c
+++ b/userdiff.c
@@ -6,8 +6,8 @@ static struct userdiff_driver *drivers;
 static int ndrivers;
 static int drivers_alloc;
 
-#define PATTERNS(name, pattern, wordregex)			\
-	{ name, NULL, -1, { pattern, REG_EXTENDED }, wordregex }
+#define PATTERNS(name, pattern, word_regex)			\
+	{ name, NULL, -1, { pattern, REG_EXTENDED }, word_regex }
 static struct userdiff_driver builtin_drivers[] = {
 PATTERNS("html", "^[ \t]*(<[Hh][1-6][ \t].*>.*)$",
 	 "[^<>= \t]+|[^[:space:]]|[\x80-\xff]+"),

Documentation/diff-options.txt, Documentation/gitattributes.txt, diff.c, t/t4034-diff-words.sh, userdiff.c, userdiff.h

diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 2c1fa4b..ef0e2f5 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -97,6 +97,9 @@ endif::git-format-patch[]
 Optionally, you can pass a regular expression that tells Git what the
 words are that you are looking for; The default is to interpret any
 stretch of non-whitespace as a word.
+The regex can also be set via a diff driver, see
+linkgit:gitattributes[1]; giving it explicitly overrides any diff
+driver setting.
 
 --no-renames::
 	Turn off rename detection, even when the configuration
diff --git a/Documentation/gitattributes.txt b/Documentation/gitattributes.txt
index 8af22ec..17707ba 100644
--- a/Documentation/gitattributes.txt
+++ b/Documentation/gitattributes.txt
@@ -317,6 +317,8 @@ patterns are available:
 
 - `bibtex` suitable for files with BibTeX coded references.
 
+- `cpp` suitable for source code in the C and C++ languages.
+
 - `html` suitable for HTML/XHTML documents.
 
 - `java` suitable for source code in the Java language.
@@ -334,6 +336,25 @@ patterns are available:
 - `tex` suitable for source code for LaTeX documents.
 
 
+Customizing word diff
+^^^^^^^^^^^^^^^^^^^^^
+
+You can customize the rules that `git diff --color-words` uses to
+split words in a line, by specifying an appropriate regular expression
+in the "diff.*.wordregex" configuration variable.  For example, in TeX
+a backslash followed by a sequence of letters forms a command, but
+several such commands can be run together without intervening
+whitespace.  To separate them, use a regular expression such as
+
+------------------------
+[diff "tex"]
+	wordregex = "\\\\[a-zA-Z]+|[{}]|\\\\.|[^\\{}[:space:]]+"
+------------------------
+
+A built-in pattern is provided for all languages listed in the last
+section.
+
+
 Performing text diffs of binary files
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
diff --git a/diff.c b/diff.c
index eb67431..08bdc86 100644
--- a/diff.c
+++ b/diff.c
@@ -1372,6 +1372,12 @@ int diff_filespec_is_binary(struct diff_filespec *one)
 	return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
 }
 
+static const char *userdiff_word_regex(struct diff_filespec *one)
+{
+	diff_filespec_load_driver(one);
+	return one->driver->word_regex;
+}
+
 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
 {
 	if (!options->a_prefix)
@@ -1532,6 +1538,10 @@ static void builtin_diff(const char *name_a,
 			ecbdata.diff_words =
 				xcalloc(1, sizeof(struct diff_words_data));
 			ecbdata.diff_words->file = o->file;
+			if (!o->word_regex)
+				o->word_regex = userdiff_word_regex(one);
+			if (!o->word_regex)
+				o->word_regex = userdiff_word_regex(two);
 			if (o->word_regex) {
 				ecbdata.diff_words->word_regex = (regex_t *)
 					xmalloc(sizeof(regex_t));
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 0ed7e53..d6731d1 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -83,4 +83,44 @@ test_expect_success 'word diff with a regular expression' '
 
 '
 
+cat > expect-by-chars <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+<RED>h(4)<RESET><GREEN>h(4),hh[44]<RESET>
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa )<RESET>
+EOF
+
+test_expect_success 'set a diff driver' '
+	git config diff.testdriver.wordregex "[^[:space:]]" &&
+	cat <<EOF > .gitattributes
+test_* diff=testdriver
+EOF
+'
+
+test_expect_success 'use default supplied by driver' '
+
+	test_must_fail git diff --no-index --color-words \
+		pre post > output &&
+	decrypt_color < output > output.decrypted &&
+	test_cmp expect-by-chars output.decrypted
+
+'
+
+test_expect_success 'option overrides default' '
+
+	test_must_fail git diff --no-index --color-words="[a-z]+" \
+		pre post > output &&
+	decrypt_color < output > output.decrypted &&
+	test_cmp expect output.decrypted
+
+'
+
 test_done
diff --git a/userdiff.c b/userdiff.c
index 3681062..79f9cb9 100644
--- a/userdiff.c
+++ b/userdiff.c
@@ -6,14 +6,20 @@
 static int ndrivers;
 static int drivers_alloc;
 
-#define FUNCNAME(name, pattern) \
-	{ name, NULL, -1, { pattern, REG_EXTENDED } }
+#define PATTERNS(name, pattern, wordregex)			\
+	{ name, NULL, -1, { pattern, REG_EXTENDED }, NULL, wordregex }
 static struct userdiff_driver builtin_drivers[] = {
-FUNCNAME("html", "^[ \t]*(<[Hh][1-6][ \t].*>.*)$"),
-FUNCNAME("java",
+PATTERNS("html", "^[ \t]*(<[Hh][1-6][ \t].*>.*)$",
+	 "[^<>= \t]+|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("java",
 	 "!^[ \t]*(catch|do|for|if|instanceof|new|return|switch|throw|while)\n"
-	 "^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\\([^;]*)$"),
-FUNCNAME("objc",
+	 "^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\\([^;]*)$",
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[fFlL]?|0[xXbB]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]="
+	 "|--|\\+\\+|<<=?|>>>?=?|&&|\\|\\|"
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("objc",
 	 /* Negate C statements that can look like functions */
 	 "!^[ \t]*(do|for|if|else|return|switch|while)\n"
 	 /* Objective-C methods */
@@ -21,20 +27,60 @@
 	 /* C functions */
 	 "^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\\([^;]*)$\n"
 	 /* Objective-C class/protocol definitions */
-	 "^(@(implementation|interface|protocol)[ \t].*)$"),
-FUNCNAME("pascal",
+	 "^(@(implementation|interface|protocol)[ \t].*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[fFlL]?|0[xXbB]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]=|--|\\+\\+|<<=?|>>=?|&&|\\|\\||::|->"
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("pascal",
 	 "^((procedure|function|constructor|destructor|interface|"
 		"implementation|initialization|finalization)[ \t]*.*)$"
 	 "\n"
-	 "^(.*=[ \t]*(class|record).*)$"),
-FUNCNAME("php", "^[\t ]*((function|class).*)"),
-FUNCNAME("python", "^[ \t]*((class|def)[ \t].*)$"),
-FUNCNAME("ruby", "^[ \t]*((class|module|def)[ \t].*)$"),
-FUNCNAME("bibtex", "(@[a-zA-Z]{1,}[ \t]*\\{{0,1}[ \t]*[^ \t\"@',\\#}{~%]*).*$"),
-FUNCNAME("tex", "^(\\\\((sub)*section|chapter|part)\\*{0,1}\\{.*)$"),
+	 "^(.*=[ \t]*(class|record).*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+|0[xXbB]?[0-9a-fA-F]+"
+	 "|<>|<=|>=|:=|\\.\\."
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("php", "^[\t ]*((function|class).*)",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+|0[xXbB]?[0-9a-fA-F]+"
+	 "|[-+*/<>%&^|=!.]=|--|\\+\\+|<<=?|>>=?|===|&&|\\|\\||::|->"
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("python", "^[ \t]*((class|def)[ \t].*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[jJlL]?|0[xX]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]=|//=?|<<=?|>>=?|\\*\\*=?"
+	 "|[^[:space:]|[\x80-\xff]+"),
+	 /* -- */
+PATTERNS("ruby", "^[ \t]*((class|module|def)[ \t].*)$",
+	 /* -- */
+	 "(@|@@|\\$)?[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+|0[xXbB]?[0-9a-fA-F]+|\\?(\\\\C-)?(\\\\M-)?."
+	 "|//=?|[-+*/<>%&^|=!]=|<<=?|>>=?|===|\\.{1,3}|::|[!=]~"
+	 "|[^[:space:]|[\x80-\xff]+"),
+PATTERNS("bibtex", "(@[a-zA-Z]{1,}[ \t]*\\{{0,1}[ \t]*[^ \t\"@',\\#}{~%]*).*$",
+	 "[={}\"]|[^={}\" \t]+"),
+PATTERNS("tex", "^(\\\\((sub)*section|chapter|part)\\*{0,1}\\{.*)$",
+	 "\\\\[a-zA-Z@]+|[{}]|\\\\.|[^\\{} \t]+"),
+PATTERNS("cpp",
+	 /* Jump targets or access declarations */
+	 "!^[ \t]*[A-Za-z_][A-Za-z_0-9]*:.*$\n"
+	 /* C functions at top level */
+	 "^([A-Za-z_][A-Za-z_0-9]*([ \t]+[A-Za-z_][A-Za-z_0-9]*){1,}[ \t]*\\([^;]*)$\n"
+	 /* compound type at top level */
+	 "^((struct|class|enum)[^;]*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[fFlL]?|0[xXbB]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]=|--|\\+\\+|<<=?|>>=?|&&|\\|\\||::|->"
+	 "|[^[:space:]]|[\x80-\xff]+"),
 { "default", NULL, -1, { NULL, 0 } },
 };
-#undef FUNCNAME
+#undef PATTERNS
 
 static struct userdiff_driver driver_true = {
 	"diff=true",
@@ -134,6 +180,8 @@ int userdiff_config(const char *k, const char *v)
 		return parse_string(&drv->external, k, v);
 	if ((drv = parse_driver(k, v, "textconv")))
 		return parse_string(&drv->textconv, k, v);
+	if ((drv = parse_driver(k, v, "wordregex")))
+		return parse_string(&drv->word_regex, k, v);
 
 	return 0;
 }
diff --git a/userdiff.h b/userdiff.h
index ba29457..2aab13e 100644
--- a/userdiff.h
+++ b/userdiff.h
@@ -12,6 +12,7 @@ struct userdiff_driver {
 	int binary;
 	struct userdiff_funcname funcname;
 	const char *textconv;
+	const char *word_regex;
 };
 
 int userdiff_config(const char *k, const char *v);

diff.c, t/t4034-diff-words.sh

diff --git a/diff.c b/diff.c
index 6d87ea5..fe8b1f0 100644
--- a/diff.c
+++ b/diff.c
@@ -319,8 +319,10 @@ static int fill_mmfile(mmfile_t *mf, struct diff_filespec *one)
 struct diff_words_buffer {
 	mmfile_t text;
 	long alloc;
-	long current; /* output pointer */
-	int suppressed_newline;
+	struct diff_words_orig {
+		const char *begin, *end;
+	} *orig;
+	int orig_nr, orig_alloc;
 };
 
 static void diff_words_append(char *line, unsigned long len,
@@ -335,80 +337,81 @@ static void diff_words_append(char *line, unsigned long len,
 
 struct diff_words_data {
 	struct diff_words_buffer minus, plus;
+	const char *current_plus;
 	FILE *file;
 };
 
-static void print_word(FILE *file, struct diff_words_buffer *buffer, int len, int color,
-		int suppress_newline)
-{
-	const char *ptr;
-	int eol = 0;
-
-	if (len == 0)
-		return;
-
-	ptr  = buffer->text.ptr + buffer->current;
-	buffer->current += len;
-
-	if (ptr[len - 1] == '\n') {
-		eol = 1;
-		len--;
-	}
-
-	fputs(diff_get_color(1, color), file);
-	fwrite(ptr, len, 1, file);
-	fputs(diff_get_color(1, DIFF_RESET), file);
-
-	if (eol) {
-		if (suppress_newline)
-			buffer->suppressed_newline = 1;
-		else
-			putc('\n', file);
-	}
-}
-
 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 {
 	struct diff_words_data *diff_words = priv;
+	int minus_first, minus_len, plus_first, plus_len;
+	const char *minus_begin, *minus_end, *plus_begin, *plus_end;
 
-	if (diff_words->minus.suppressed_newline) {
-		if (line[0] != '+')
-			putc('\n', diff_words->file);
-		diff_words->minus.suppressed_newline = 0;
-	}
+	if (line[0] != '@' || parse_hunk_header(line, len,
+			&minus_first, &minus_len, &plus_first, &plus_len))
+		return;
 
-	len--;
-	switch (line[0]) {
-		case '-':
-			print_word(diff_words->file,
-				   &diff_words->minus, len, DIFF_FILE_OLD, 1);
-			break;
-		case '+':
-			print_word(diff_words->file,
-				   &diff_words->plus, len, DIFF_FILE_NEW, 0);
-			break;
-		case ' ':
-			print_word(diff_words->file,
-				   &diff_words->plus, len, DIFF_PLAIN, 0);
-			diff_words->minus.current += len;
-			break;
-	}
+	minus_begin = diff_words->minus.orig[minus_first].begin;
+	minus_end = minus_len == 0 ? minus_begin :
+		diff_words->minus.orig[minus_first + minus_len - 1].end;
+	plus_begin = diff_words->plus.orig[plus_first].begin;
+	plus_end = plus_len == 0 ? plus_begin :
+		diff_words->plus.orig[plus_first + plus_len - 1].end;
+
+	if (diff_words->current_plus != plus_begin)
+		fwrite(diff_words->current_plus,
+				plus_begin - diff_words->current_plus, 1,
+				diff_words->file);
+	if (minus_begin != minus_end)
+		color_fwrite_lines(diff_words->file,
+				diff_get_color(1, DIFF_FILE_OLD),
+				minus_end - minus_begin, minus_begin);
+	if (plus_begin != plus_end)
+		color_fwrite_lines(diff_words->file,
+				diff_get_color(1, DIFF_FILE_NEW),
+				plus_end - plus_begin, plus_begin);
+
+	diff_words->current_plus = plus_end;
 }
 
 /*
- * This function splits the words in buffer->text, and stores the list with
- * newline separator into out.
+ * This function splits the words in buffer->text, stores the list with
+ * newline separator into out, and saves the offsets of the original words
+ * in buffer->orig.
  */
 static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 {
-	int i;
-	out->size = buffer->text.size;
-	out->ptr = xmalloc(out->size);
-	memcpy(out->ptr, buffer->text.ptr, out->size);
-	for (i = 0; i < out->size; i++)
-		if (isspace(out->ptr[i]))
-			out->ptr[i] = '\n';
-	buffer->current = 0;
+	int i, j;
+
+	out->size = 0;
+	out->ptr = xmalloc(buffer->text.size);
+
+	/* fake an empty "0th" word */
+	ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
+	buffer->orig[0].begin = buffer->orig[0].end = buffer->text.ptr;
+	buffer->orig_nr = 1;
+
+	for (i = 0; i < buffer->text.size; i++) {
+		if (isspace(buffer->text.ptr[i]))
+			continue;
+		for (j = i + 1; j < buffer->text.size &&
+				!isspace(buffer->text.ptr[j]); j++)
+			; /* find the end of the word */
+
+		/* store original boundaries */
+		ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
+				buffer->orig_alloc);
+		buffer->orig[buffer->orig_nr].begin = buffer->text.ptr + i;
+		buffer->orig[buffer->orig_nr].end = buffer->text.ptr + j;
+		buffer->orig_nr++;
+
+		/* store one word */
+		memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
+		out->ptr[out->size + j - i] = '\n';
+		out->size += j - i + 1;
+
+		i = j - 1;
+	}
 }
 
 /* this executes the word diff on the accumulated buffers */
@@ -419,22 +422,34 @@ static void diff_words_show(struct diff_words_data *diff_words)
 	xdemitcb_t ecb;
 	mmfile_t minus, plus;
 
+	/* special case: only removal */
+	if (!diff_words->plus.text.size) {
+		color_fwrite_lines(diff_words->file,
+			diff_get_color(1, DIFF_FILE_OLD),
+			diff_words->minus.text.size, diff_words->minus.text.ptr);
+		diff_words->minus.text.size = 0;
+		return;
+	}
+
+	diff_words->current_plus = diff_words->plus.text.ptr;
+
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
 	diff_words_fill(&diff_words->minus, &minus);
 	diff_words_fill(&diff_words->plus, &plus);
 	xpp.flags = XDF_NEED_MINIMAL;
-	xecfg.ctxlen = diff_words->minus.alloc + diff_words->plus.alloc;
+	xecfg.ctxlen = 0;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
 		      &xpp, &xecfg, &ecb);
 	free(minus.ptr);
 	free(plus.ptr);
+	if (diff_words->current_plus != diff_words->plus.text.ptr +
+			diff_words->plus.text.size)
+		fwrite(diff_words->current_plus,
+			diff_words->plus.text.ptr + diff_words->plus.text.size
+			- diff_words->current_plus, 1,
+			diff_words->file);
 	diff_words->minus.text.size = diff_words->plus.text.size = 0;
-
-	if (diff_words->minus.suppressed_newline) {
-		putc('\n', diff_words->file);
-		diff_words->minus.suppressed_newline = 0;
-	}
 }
 
 typedef unsigned long (*sane_truncate_fn)(char *line, unsigned long len);
@@ -458,7 +473,9 @@ static void free_diff_words_data(struct emit_callback *ecbdata)
 			diff_words_show(ecbdata->diff_words);
 
 		free (ecbdata->diff_words->minus.text.ptr);
+		free (ecbdata->diff_words->minus.orig);
 		free (ecbdata->diff_words->plus.text.ptr);
+		free (ecbdata->diff_words->plus.orig);
 		free(ecbdata->diff_words);
 		ecbdata->diff_words = NULL;
 	}
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
new file mode 100755
index 0000000..b22195f
--- /dev/null
+++ b/t/t4034-diff-words.sh
@@ -0,0 +1,66 @@
+#!/bin/sh
+
+test_description='word diff colors'
+
+. ./test-lib.sh
+
+test_expect_success setup '
+
+	git config diff.color.old red
+	git config diff.color.new green
+
+'
+
+decrypt_color () {
+	sed \
+		-e 's/.\[1m/<WHITE>/g' \
+		-e 's/.\[31m/<RED>/g' \
+		-e 's/.\[32m/<GREEN>/g' \
+		-e 's/.\[36m/<BROWN>/g' \
+		-e 's/.\[m/<RESET>/g'
+}
+
+word_diff () {
+	test_must_fail git diff --no-index "$@" pre post > output &&
+	decrypt_color < output > output.decrypted &&
+	test_cmp expect output.decrypted
+}
+
+cat > pre <<\EOF
+h(4)
+
+a = b + c
+EOF
+
+cat > post <<\EOF
+h(4),hh[44]
+
+a = b + c
+
+aa = a
+
+aeff = aeff * ( aaa )
+EOF
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+<RED>h(4)<RESET><GREEN>h(4),hh[44]<RESET>
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa )<RESET>
+EOF
+
+test_expect_success 'word diff with runs of whitespace' '
+
+	word_diff --color-words
+
+'
+
+test_done

Documentation/diff-options.txt, diff.c, diff.h, t/t4034-diff-words.sh

diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 1f8ce97..e546bfa 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -94,8 +94,12 @@ endif::git-format-patch[]
 	Turn off colored diff, even when the configuration file
 	gives the default to color output.
 
---color-words::
+--color-words[=regex]::
 	Show colored word diff, i.e. color words which have changed.
++
+Optionally, you can pass a regular expression that tells Git what the
+words are that you are looking for; The default is to interpret any
+stretch of non-whitespace as a word.
 
 --no-renames::
 	Turn off rename detection, even when the configuration
diff --git a/diff.c b/diff.c
index fe8b1f0..1408717 100644
--- a/diff.c
+++ b/diff.c
@@ -333,12 +333,14 @@ static void diff_words_append(char *line, unsigned long len,
 	len--;
 	memcpy(buffer->text.ptr + buffer->text.size, line, len);
 	buffer->text.size += len;
+	buffer->text.ptr[buffer->text.size] = '\0';
 }
 
 struct diff_words_data {
 	struct diff_words_buffer minus, plus;
 	const char *current_plus;
 	FILE *file;
+	regex_t *word_regex;
 };
 
 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
@@ -374,17 +376,49 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 	diff_words->current_plus = plus_end;
 }
 
+/* This function starts looking at *begin, and returns 0 iff a word was found. */
+static int find_word_boundaries(mmfile_t *buffer, regex_t *word_regex,
+		int *begin, int *end)
+{
+	if (word_regex && *begin < buffer->size) {
+		regmatch_t match[1];
+		if (!regexec(word_regex, buffer->ptr + *begin, 1, match, 0)) {
+			char *p = memchr(buffer->ptr + *begin + match[0].rm_so,
+					'\n', match[0].rm_eo - match[0].rm_so);
+			*end = p ? p - buffer->ptr : match[0].rm_eo + *begin;
+			*begin += match[0].rm_so;
+			return *begin >= *end;
+		}
+		return -1;
+	}
+
+	/* find the next word */
+	while (*begin < buffer->size && isspace(buffer->ptr[*begin]))
+		(*begin)++;
+	if (*begin >= buffer->size)
+		return -1;
+
+	/* find the end of the word */
+	*end = *begin + 1;
+	while (*end < buffer->size && !isspace(buffer->ptr[*end]))
+		(*end)++;
+
+	return 0;
+}
+
 /*
  * This function splits the words in buffer->text, stores the list with
  * newline separator into out, and saves the offsets of the original words
  * in buffer->orig.
  */
-static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
+static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
+		regex_t *word_regex)
 {
 	int i, j;
+	long alloc = 0;
 
 	out->size = 0;
-	out->ptr = xmalloc(buffer->text.size);
+	out->ptr = NULL;
 
 	/* fake an empty "0th" word */
 	ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
@@ -392,11 +426,8 @@ static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 	buffer->orig_nr = 1;
 
 	for (i = 0; i < buffer->text.size; i++) {
-		if (isspace(buffer->text.ptr[i]))
-			continue;
-		for (j = i + 1; j < buffer->text.size &&
-				!isspace(buffer->text.ptr[j]); j++)
-			; /* find the end of the word */
+		if (find_word_boundaries(&buffer->text, word_regex, &i, &j))
+			return;
 
 		/* store original boundaries */
 		ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
@@ -406,6 +437,7 @@ static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 		buffer->orig_nr++;
 
 		/* store one word */
+		ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
 		memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
 		out->ptr[out->size + j - i] = '\n';
 		out->size += j - i + 1;
@@ -435,9 +467,10 @@ static void diff_words_show(struct diff_words_data *diff_words)
 
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
-	diff_words_fill(&diff_words->minus, &minus);
-	diff_words_fill(&diff_words->plus, &plus);
+	diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
+	diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
 	xpp.flags = XDF_NEED_MINIMAL;
+	/* as only the hunk header will be parsed, we need a 0-context */
 	xecfg.ctxlen = 0;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
 		      &xpp, &xecfg, &ecb);
@@ -476,6 +509,7 @@ static void free_diff_words_data(struct emit_callback *ecbdata)
 		free (ecbdata->diff_words->minus.orig);
 		free (ecbdata->diff_words->plus.text.ptr);
 		free (ecbdata->diff_words->plus.orig);
+		free(ecbdata->diff_words->word_regex);
 		free(ecbdata->diff_words);
 		ecbdata->diff_words = NULL;
 	}
@@ -1498,6 +1532,14 @@ static void builtin_diff(const char *name_a,
 			ecbdata.diff_words =
 				xcalloc(1, sizeof(struct diff_words_data));
 			ecbdata.diff_words->file = o->file;
+			if (o->word_regex) {
+				ecbdata.diff_words->word_regex = (regex_t *)
+					xmalloc(sizeof(regex_t));
+				if (regcomp(ecbdata.diff_words->word_regex,
+						o->word_regex, REG_EXTENDED))
+					die ("Invalid regular expression: %s",
+							o->word_regex);
+			}
 		}
 		xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
 			      &xpp, &xecfg, &ecb);
@@ -2513,6 +2555,10 @@ int diff_opt_parse(struct diff_options *options, const char **av, int ac)
 		DIFF_OPT_CLR(options, COLOR_DIFF);
 	else if (!strcmp(arg, "--color-words"))
 		options->flags |= DIFF_OPT_COLOR_DIFF | DIFF_OPT_COLOR_DIFF_WORDS;
+	else if (!prefixcmp(arg, "--color-words=")) {
+		options->flags |= DIFF_OPT_COLOR_DIFF | DIFF_OPT_COLOR_DIFF_WORDS;
+		options->word_regex = arg + 14;
+	}
 	else if (!strcmp(arg, "--exit-code"))
 		DIFF_OPT_SET(options, EXIT_WITH_STATUS);
 	else if (!strcmp(arg, "--quiet"))
diff --git a/diff.h b/diff.h
index 4d5a327..23cd90c 100644
--- a/diff.h
+++ b/diff.h
@@ -98,6 +98,7 @@ struct diff_options {
 
 	int stat_width;
 	int stat_name_width;
+	const char *word_regex;
 
 	/* this is set by diffcore for DIFF_FORMAT_PATCH */
 	int found_changes;
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index b22195f..f4810e9 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -63,4 +63,43 @@ test_expect_success 'word diff with runs of whitespace' '
 
 '
 
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+h(4),<GREEN>hh<RESET>[44]
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa<RESET> )
+EOF
+
+test_expect_success 'word diff with a regular expression' '
+
+	word_diff --color-words='[a-z]+'
+
+'
+
+echo 'aaa (aaa)' > pre
+echo 'aaa (aaa) aaa' > post
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index c29453b..be22f37 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1 +1 @@<RESET>
+aaa (aaa) <GREEN>aaa<RESET>
+EOF
+
+test_expect_success "test parsing words for newline" '
+
+	word_diff --color-words='a+'
+
+'
+
 test_done

a1, a2

diff --git a/a1 b/a2
index 8309acb..2d06f37 100644
--- a/a1
+++ b/a2
@@ -2 +1,0 @@
-:


Notice the "+1,0"?  I fully expected this to be "+2,0", but apparently I 
was mistaken...

Can anybody explain to me why this is so?

Ciao,
Dscho

diff.c, t/t4034-diff-words.sh

diff --git a/diff.c b/diff.c
index c5f7c57..3709651 100644
--- a/diff.c
+++ b/diff.c
@@ -360,6 +360,24 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 	plus_end = plus_len == 0 ? plus_begin :
 		diff_words->plus.orig[plus_first + plus_len - 1].end;
 
+	/*
+	 * since this is a --unified=0 diff, it can result in a single hunk
+	 * with a header like this: @@ -2 +1,0 @@
+	 *
+	 * This breaks the assumption that minus_first == plus_first.
+	 *
+	 * So we have to fix it: whenever we reach the end of pre and post
+	 * texts, but nothing was added, we need to shift the plus part
+	 * to the end of the buffer.
+	 *
+	 * It is only necessary for the plus part, as we show the common
+	 * words from that buffer.
+	 */
+	if (plus_len == 0 && minus_first + minus_len
+			== diff_words->minus.orig_nr)
+		plus_begin = plus_end =
+			diff_words->plus.orig[diff_words->plus.orig_nr - 1].end;
+
 	if (diff_words->current_plus != plus_begin)
 		fwrite(diff_words->current_plus,
 				plus_begin - diff_words->current_plus, 1,
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 07e48d1..817fba6 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -135,6 +135,22 @@ test_expect_success 'test parsing words for newline' '
 
 	word_diff --color-words="a+"
 
+'
+
+echo '(:' > pre
+echo '(' > post
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 289cb9d..2d06f37 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1 +1 @@<RESET>
+(<RED>:<RESET>
+EOF
+
+test_expect_success 'test when words are only removed at the end' '
+
 	word_diff --color-words=.
 
 '

xdiff/xutils.c, diff.c, t/t4034-diff-words.sh

diff --git a/xdiff/xutils.c b/xdiff/xutils.c
[...]
  @@ -244,7 +257,7 @@ int xdl_emit_hunk_hdr(long s1, long c1, long s2, long c2,
          memcpy(buf, "@@ -", 4);
          nb += 4;

  -       nb += xdl_num_out(buf + nb, c1 ? s1: 0);
  +       nb += xdl_num_out(buf + nb, c1 ? s1: s1 - 1);

          if (c1 != 1) {
                  memcpy(buf + nb, ",", 1);
  @@ -256,7 +269,7 @@ int xdl_emit_hunk_hdr(long s1, long c1, long s2, long c2,
          memcpy(buf + nb, " +", 2);
          nb += 2;

  -       nb += xdl_num_out(buf + nb, c2 ? s2: 0);
  +       nb += xdl_num_out(buf + nb, c2 ? s2: s2 - 1);

          if (c2 != 1) {
                  memcpy(buf + nb, ",", 1);


Note how (for some reason I don't quite understand yet) "correcting"
the offsets involves subtracting 1 if there were no changes on that
side.

But skipping ahead to the end doesn't work if there are several such
instances where nothing was added.  So I think it must be fixed as
follows.

---- 8< ----
diff --git a/diff.c b/diff.c
index 4174d88..d7bbf74 100644
--- a/diff.c
+++ b/diff.c
@@ -361,8 +361,9 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 		diff_words->plus.orig[plus_first + plus_len - 1].end;
 
 	/*
-	 * since this is a --unified=0 diff, it can result in a single hunk
-	 * with a header like this: @@ -2 +1,0 @@
+	 * libxdiff subtracts one from the offset if the corresponding
+	 * length is 0.	 (This can only happen because we use
+	 * --unified=0.)
 	 *
 	 * This breaks the assumption that minus_first == plus_first.
 	 *
@@ -373,10 +374,9 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 	 * It is only necessary for the plus part, as we show the common
 	 * words from that buffer.
 	 */
-	if (plus_len == 0 && minus_first + minus_len
-			== diff_words->minus.orig_nr)
+	if (plus_len == 0)
 		plus_begin = plus_end =
-			diff_words->plus.orig[diff_words->plus.orig_nr - 1].end;
+			diff_words->plus.orig[plus_first + plus_len].end;
 
 	if (diff_words->current_plus != plus_begin)
 		fwrite(diff_words->current_plus,
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 744221b..875b464 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -156,4 +156,40 @@ test_expect_success 'test when words are only removed at the end' '
 
 '
 
+echo 'abcd(Xefghijklmn(YZopqrst' > pre
+echo 'abcd(efghijklmn(opqrst' > post
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 434ff54..c4bb9f1 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1 +1 @@<RESET>
+abcd(<RED>X<RESET>efghijklmn(<RED>YZ<RESET>opqrst
+EOF
+
+test_expect_success 'no added words' '
+
+	word_diff --color-words=.
+
+'
+
+echo 'abcd(efghijklmn(opqrst' > pre
+echo 'abcd(Xefghijklmn(YZopqrst' > post
+
+cat > expect <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index c4bb9f1..434ff54 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1 +1 @@<RESET>
+abcd(<GREEN>X<RESET>efghijklmn(<GREEN>YZ<RESET>opqrst
+EOF
+
+test_expect_success 'no removed words' '
+
+	word_diff --color-words=.
+
+'
+
 test_done

diff --git a/diff.c b/diff.c
index 3709651..219a242 100644
--- a/diff.c
+++ b/diff.c
@@ -353,30 +353,20 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 			&minus_first, &minus_len, &plus_first, &plus_len))
 		return;
 
-	minus_begin = diff_words->minus.orig[minus_first].begin;
-	minus_end = minus_len == 0 ? minus_begin :
-		diff_words->minus.orig[minus_first + minus_len - 1].end;
-	plus_begin = diff_words->plus.orig[plus_first].begin;
-	plus_end = plus_len == 0 ? plus_begin :
-		diff_words->plus.orig[plus_first + plus_len - 1].end;
+	/* POSIX requires that first be decremented by one if len == 0... */
+	if (minus_len) {
+		minus_begin = diff_words->minus.orig[minus_first].begin;
+		minus_end =
+			diff_words->minus.orig[minus_first + minus_len - 1].end;
+	} else
+		minus_begin = minus_end =
+			diff_words->minus.orig[minus_first].end;
 
-	/*
-	 * since this is a --unified=0 diff, it can result in a single hunk
-	 * with a header like this: @@ -2 +1,0 @@
-	 *
-	 * This breaks the assumption that minus_first == plus_first.
-	 *
-	 * So we have to fix it: whenever we reach the end of pre and post
-	 * texts, but nothing was added, we need to shift the plus part
-	 * to the end of the buffer.
-	 *
-	 * It is only necessary for the plus part, as we show the common
-	 * words from that buffer.
-	 */
-	if (plus_len == 0 && minus_first + minus_len
-			== diff_words->minus.orig_nr)
-		plus_begin = plus_end =
-			diff_words->plus.orig[diff_words->plus.orig_nr - 1].end;
+	if (plus_len) {
+		plus_begin = diff_words->plus.orig[plus_first].begin;
+		plus_end = diff_words->plus.orig[plus_first + plus_len - 1].end;
+	} else
+		plus_begin = plus_end = diff_words->plus.orig[plus_first].end;
 
 	if (diff_words->current_plus != plus_begin)
 		fwrite(diff_words->current_plus,

diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index f4810e9..6ad1c1f 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -80,7 +80,7 @@ EOF
 
 test_expect_success 'word diff with a regular expression' '
 
-	word_diff --color-words='[a-z]+'
+	word_diff --color-words="[a-z]+"
 
 '
 
@@ -98,7 +98,7 @@ EOF
 
 test_expect_success "test parsing words for newline" '
 
-	word_diff --color-words='a+'
+	word_diff --color-words="a+"
 
 '
 

diff --git a/diff.c b/diff.c
index cc42adf..3f07ac1 100644
--- a/diff.c
+++ b/diff.c
@@ -1536,7 +1536,8 @@ static void builtin_diff(const char *name_a,
 				ecbdata.diff_words->word_regex = (regex_t *)
 					xmalloc(sizeof(regex_t));
 				if (regcomp(ecbdata.diff_words->word_regex,
-						o->word_regex, REG_EXTENDED))
+						o->word_regex,
+						REG_EXTENDED | REG_NEWLINE))
 					die ("Invalid regular expression: %s",
 							o->word_regex);
 			}

diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 2c1fa4b..8689a92 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -91,12 +91,17 @@ endif::git-format-patch[]
 	Turn off colored diff, even when the configuration file
 	gives the default to color output.
 
---color-words[=regex]::
-	Show colored word diff, i.e. color words which have changed.
+--color-words[=<regex>]::
+	Show colored word diff, i.e., color words which have changed.
+	By default, words are separated by whitespace.
 +
-Optionally, you can pass a regular expression that tells Git what the
-words are that you are looking for; The default is to interpret any
-stretch of non-whitespace as a word.
+When a <regex> is specified, every non-overlapping match of the
+<regex> is considered a word.  Anything between these matches is
+considered whitespace and ignored(!) for the purposes of finding
+differences.  You may want to append `|[^[:space:]]` to your regular
+expression to make sure that it matches all non-whitespace characters.
+A match that contains a newline is silently truncated(!) at the
+newline.
 
 --no-renames::
 	Turn off rename detection, even when the configuration

Documentation/diff-options.txt, Documentation/gitattributes.txt, diff.c, t/t4034-diff-words.sh, userdiff.c, userdiff.h

diff --git a/Documentation/diff-options.txt b/Documentation/diff-options.txt
index 8689a92..1edb82e 100644
--- a/Documentation/diff-options.txt
+++ b/Documentation/diff-options.txt
@@ -102,6 +102,10 @@ differences.  You may want to append `|[^[:space:]]` to your regular
 expression to make sure that it matches all non-whitespace characters.
 A match that contains a newline is silently truncated(!) at the
 newline.
++
+The regex can also be set via a diff driver, see
+linkgit:gitattributes[1]; giving it explicitly overrides any diff
+driver setting.
 
 --no-renames::
 	Turn off rename detection, even when the configuration
diff --git a/Documentation/gitattributes.txt b/Documentation/gitattributes.txt
index 8af22ec..17707ba 100644
--- a/Documentation/gitattributes.txt
+++ b/Documentation/gitattributes.txt
@@ -317,6 +317,8 @@ patterns are available:
 
 - `bibtex` suitable for files with BibTeX coded references.
 
+- `cpp` suitable for source code in the C and C++ languages.
+
 - `html` suitable for HTML/XHTML documents.
 
 - `java` suitable for source code in the Java language.
@@ -334,6 +336,25 @@ patterns are available:
 - `tex` suitable for source code for LaTeX documents.
 
 
+Customizing word diff
+^^^^^^^^^^^^^^^^^^^^^
+
+You can customize the rules that `git diff --color-words` uses to
+split words in a line, by specifying an appropriate regular expression
+in the "diff.*.wordregex" configuration variable.  For example, in TeX
+a backslash followed by a sequence of letters forms a command, but
+several such commands can be run together without intervening
+whitespace.  To separate them, use a regular expression such as
+
+------------------------
+[diff "tex"]
+	wordregex = "\\\\[a-zA-Z]+|[{}]|\\\\.|[^\\{}[:space:]]+"
+------------------------
+
+A built-in pattern is provided for all languages listed in the last
+section.
+
+
 Performing text diffs of binary files
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
diff --git a/diff.c b/diff.c
index 3f07ac1..0e82e18 100644
--- a/diff.c
+++ b/diff.c
@@ -1372,6 +1372,12 @@ int diff_filespec_is_binary(struct diff_filespec *one)
 	return one->driver->funcname.pattern ? &one->driver->funcname : NULL;
 }
 
+static const char *userdiff_word_regex(struct diff_filespec *one)
+{
+	diff_filespec_load_driver(one);
+	return one->driver->word_regex;
+}
+
 void diff_set_mnemonic_prefix(struct diff_options *options, const char *a, const char *b)
 {
 	if (!options->a_prefix)
@@ -1532,6 +1538,10 @@ static void builtin_diff(const char *name_a,
 			ecbdata.diff_words =
 				xcalloc(1, sizeof(struct diff_words_data));
 			ecbdata.diff_words->file = o->file;
+			if (!o->word_regex)
+				o->word_regex = userdiff_word_regex(one);
+			if (!o->word_regex)
+				o->word_regex = userdiff_word_regex(two);
 			if (o->word_regex) {
 				ecbdata.diff_words->word_regex = (regex_t *)
 					xmalloc(sizeof(regex_t));
diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 6ad1c1f..631ca44 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -22,8 +22,10 @@ decrypt_color () {
 
 word_diff () {
 	test_must_fail git diff --no-index "$@" pre post > output &&
-	decrypt_color < output > output.decrypted &&
-	test_cmp expect output.decrypted
+	decrypt_color < output > output.decrypted
+}
+word_diff_check () {
+	test_cmp "$1" output.decrypted
 }
 
 cat > pre <<\EOF
@@ -80,7 +82,45 @@ EOF
 
 test_expect_success 'word diff with a regular expression' '
 
-	word_diff --color-words="[a-z]+"
+	word_diff --color-words="[a-z]+" &&
+	word_diff_check expect
+
+'
+
+cat > expect-by-chars <<\EOF
+<WHITE>diff --git a/pre b/post<RESET>
+<WHITE>index 330b04f..5ed8eff 100644<RESET>
+<WHITE>--- a/pre<RESET>
+<WHITE>+++ b/post<RESET>
+<BROWN>@@ -1,3 +1,7 @@<RESET>
+h(4)<GREEN>,hh[44]<RESET>
+<RESET>
+a = b + c<RESET>
+
+<GREEN>aa = a<RESET>
+
+<GREEN>aeff = aeff * ( aaa )<RESET>
+EOF
+
+test_expect_success 'set a diff driver' '
+	git config diff.testdriver.wordregex "[^[:space:]]" &&
+	cat <<EOF > .gitattributes
+pre diff=testdriver
+post diff=testdriver
+EOF
+'
+
+test_expect_success 'use default supplied by driver' '
+
+	word_diff --color-words &&
+	word_diff_check expect-by-chars
+
+'
+
+test_expect_success 'option overrides default' '
+
+	word_diff --color-words="[a-z]+" &&
+	word_diff_check expect
 
 '
 
@@ -98,7 +138,8 @@ EOF
 
 test_expect_success "test parsing words for newline" '
 
-	word_diff --color-words="a+"
+	word_diff --color-words="a+" &&
+	word_diff_check expect
 
 '
 
diff --git a/userdiff.c b/userdiff.c
index 3681062..dbfda6d 100644
--- a/userdiff.c
+++ b/userdiff.c
@@ -6,14 +6,20 @@
 static int ndrivers;
 static int drivers_alloc;
 
-#define FUNCNAME(name, pattern) \
-	{ name, NULL, -1, { pattern, REG_EXTENDED } }
+#define PATTERNS(name, pattern, wordregex)			\
+	{ name, NULL, -1, { pattern, REG_EXTENDED }, wordregex }
 static struct userdiff_driver builtin_drivers[] = {
-FUNCNAME("html", "^[ \t]*(<[Hh][1-6][ \t].*>.*)$"),
-FUNCNAME("java",
+PATTERNS("html", "^[ \t]*(<[Hh][1-6][ \t].*>.*)$",
+	 "[^<>= \t]+|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("java",
 	 "!^[ \t]*(catch|do|for|if|instanceof|new|return|switch|throw|while)\n"
-	 "^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\\([^;]*)$"),
-FUNCNAME("objc",
+	 "^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\\([^;]*)$",
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[fFlL]?|0[xXbB]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]="
+	 "|--|\\+\\+|<<=?|>>>?=?|&&|\\|\\|"
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("objc",
 	 /* Negate C statements that can look like functions */
 	 "!^[ \t]*(do|for|if|else|return|switch|while)\n"
 	 /* Objective-C methods */
@@ -21,20 +27,60 @@
 	 /* C functions */
 	 "^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\\([^;]*)$\n"
 	 /* Objective-C class/protocol definitions */
-	 "^(@(implementation|interface|protocol)[ \t].*)$"),
-FUNCNAME("pascal",
+	 "^(@(implementation|interface|protocol)[ \t].*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[fFlL]?|0[xXbB]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]=|--|\\+\\+|<<=?|>>=?|&&|\\|\\||::|->"
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("pascal",
 	 "^((procedure|function|constructor|destructor|interface|"
 		"implementation|initialization|finalization)[ \t]*.*)$"
 	 "\n"
-	 "^(.*=[ \t]*(class|record).*)$"),
-FUNCNAME("php", "^[\t ]*((function|class).*)"),
-FUNCNAME("python", "^[ \t]*((class|def)[ \t].*)$"),
-FUNCNAME("ruby", "^[ \t]*((class|module|def)[ \t].*)$"),
-FUNCNAME("bibtex", "(@[a-zA-Z]{1,}[ \t]*\\{{0,1}[ \t]*[^ \t\"@',\\#}{~%]*).*$"),
-FUNCNAME("tex", "^(\\\\((sub)*section|chapter|part)\\*{0,1}\\{.*)$"),
+	 "^(.*=[ \t]*(class|record).*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+|0[xXbB]?[0-9a-fA-F]+"
+	 "|<>|<=|>=|:=|\\.\\."
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("php", "^[\t ]*((function|class).*)",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+|0[xXbB]?[0-9a-fA-F]+"
+	 "|[-+*/<>%&^|=!.]=|--|\\+\\+|<<=?|>>=?|===|&&|\\|\\||::|->"
+	 "|[^[:space:]]|[\x80-\xff]+"),
+PATTERNS("python", "^[ \t]*((class|def)[ \t].*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[jJlL]?|0[xX]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]=|//=?|<<=?|>>=?|\\*\\*=?"
+	 "|[^[:space:]|[\x80-\xff]+"),
+	 /* -- */
+PATTERNS("ruby", "^[ \t]*((class|module|def)[ \t].*)$",
+	 /* -- */
+	 "(@|@@|\\$)?[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+|0[xXbB]?[0-9a-fA-F]+|\\?(\\\\C-)?(\\\\M-)?."
+	 "|//=?|[-+*/<>%&^|=!]=|<<=?|>>=?|===|\\.{1,3}|::|[!=]~"
+	 "|[^[:space:]|[\x80-\xff]+"),
+PATTERNS("bibtex", "(@[a-zA-Z]{1,}[ \t]*\\{{0,1}[ \t]*[^ \t\"@',\\#}{~%]*).*$",
+	 "[={}\"]|[^={}\" \t]+"),
+PATTERNS("tex", "^(\\\\((sub)*section|chapter|part)\\*{0,1}\\{.*)$",
+	 "\\\\[a-zA-Z@]+|[{}]|\\\\.|[^\\{} \t]+"),
+PATTERNS("cpp",
+	 /* Jump targets or access declarations */
+	 "!^[ \t]*[A-Za-z_][A-Za-z_0-9]*:.*$\n"
+	 /* C functions at top level */
+	 "^([A-Za-z_][A-Za-z_0-9]*([ \t]+[A-Za-z_][A-Za-z_0-9]*){1,}[ \t]*\\([^;]*)$\n"
+	 /* compound type at top level */
+	 "^((struct|class|enum)[^;]*)$",
+	 /* -- */
+	 "[a-zA-Z_][a-zA-Z0-9_]*"
+	 "|[-+0-9.e]+[fFlL]?|0[xXbB]?[0-9a-fA-F]+[lL]?"
+	 "|[-+*/<>%&^|=!]=|--|\\+\\+|<<=?|>>=?|&&|\\|\\||::|->"
+	 "|[^[:space:]]|[\x80-\xff]+"),
 { "default", NULL, -1, { NULL, 0 } },
 };
-#undef FUNCNAME
+#undef PATTERNS
 
 static struct userdiff_driver driver_true = {
 	"diff=true",
@@ -134,6 +180,8 @@ int userdiff_config(const char *k, const char *v)
 		return parse_string(&drv->external, k, v);
 	if ((drv = parse_driver(k, v, "textconv")))
 		return parse_string(&drv->textconv, k, v);
+	if ((drv = parse_driver(k, v, "wordregex")))
+		return parse_string(&drv->word_regex, k, v);
 
 	return 0;
 }
diff --git a/userdiff.h b/userdiff.h
index ba29457..c315159 100644
--- a/userdiff.h
+++ b/userdiff.h
@@ -11,6 +11,7 @@ struct userdiff_driver {
 	const char *external;
 	int binary;
 	struct userdiff_funcname funcname;
+	const char *word_regex;
 	const char *textconv;
 };
 

diff --git a/t/t4034-diff-words.sh b/t/t4034-diff-words.sh
index 631ca44..07e48d1 100755
--- a/t/t4034-diff-words.sh
+++ b/t/t4034-diff-words.sh
@@ -22,10 +22,8 @@ decrypt_color () {
 
 word_diff () {
 	test_must_fail git diff --no-index "$@" pre post > output &&
-	decrypt_color < output > output.decrypted
-}
-word_diff_check () {
-	test_cmp "$1" output.decrypted
+	decrypt_color < output > output.decrypted &&
+	test_cmp expect output.decrypted
 }
 
 cat > pre <<\EOF
@@ -82,12 +80,25 @@ EOF
 
 test_expect_success 'word diff with a regular expression' '
 
-	word_diff --color-words="[a-z]+" &&
-	word_diff_check expect
+	word_diff --color-words="[a-z]+"
 
 '
 
-cat > expect-by-chars <<\EOF
+test_expect_success 'set a diff driver' '
+	git config diff.testdriver.wordregex "[^[:space:]]" &&
+	cat <<EOF > .gitattributes
+pre diff=testdriver
+post diff=testdriver
+EOF
+'
+
+test_expect_success 'option overrides default' '
+
+	word_diff --color-words="[a-z]+"
+
+'
+
+cat > expect <<\EOF
 <WHITE>diff --git a/pre b/post<RESET>
 <WHITE>index 330b04f..5ed8eff 100644<RESET>
 <WHITE>--- a/pre<RESET>
@@ -102,25 +113,9 @@ a = b + c<RESET>
 <GREEN>aeff = aeff * ( aaa )<RESET>
 EOF
 
-test_expect_success 'set a diff driver' '
-	git config diff.testdriver.wordregex "[^[:space:]]" &&
-	cat <<EOF > .gitattributes
-pre diff=testdriver
-post diff=testdriver
-EOF
-'
-
 test_expect_success 'use default supplied by driver' '
 
-	word_diff --color-words &&
-	word_diff_check expect-by-chars
-
-'
-
-test_expect_success 'option overrides default' '
-
-	word_diff --color-words="[a-z]+" &&
-	word_diff_check expect
+	word_diff --color-words
 
 '
 
@@ -136,10 +131,11 @@ cat > expect <<\EOF
 aaa (aaa) <GREEN>aaa<RESET>
 EOF
 
-test_expect_success "test parsing words for newline" '
+test_expect_success 'test parsing words for newline' '
+
+	word_diff --color-words="a+"
 
-	word_diff --color-words="a+" &&
-	word_diff_check expect
+	word_diff --color-words=.
 
 '
 

diff.c, diff.h

diff --git a/diff.c b/diff.c
index 2a3d301..d6bba72 100644
--- a/diff.c
+++ b/diff.c
@@ -333,12 +333,14 @@ static void diff_words_append(char *line, unsigned long len,
 	len--;
 	memcpy(buffer->text.ptr + buffer->text.size, line, len);
 	buffer->text.size += len;
+	buffer->text.ptr[buffer->text.size] = '\0';
 }
 
 struct diff_words_data {
 	struct diff_words_buffer minus, plus;
 	const char *current_plus;
 	FILE *file;
+	regex_t *word_regex;
 };
 
 static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
@@ -372,17 +374,36 @@ static void fn_out_diff_words_aux(void *priv, char *line, unsigned long len)
 	diff_words->current_plus = plus_end;
 }
 
+static int find_word_boundary(mmfile_t *buffer, int i, regex_t *word_regex)
+{
+	if (i >= buffer->size)
+		return i;
+
+	if (word_regex) {
+		regmatch_t match[1];
+		if (!regexec(word_regex, buffer->ptr + i, 1, match, 0))
+			i += match[0].rm_eo;
+	}
+	else
+		while (i < buffer->size && !isspace(buffer->ptr[i]))
+			i++;
+
+	return i;
+}
+
 /*
  * This function splits the words in buffer->text, stores the list with
  * newline separator into out, and saves the offsets of the original words
  * in buffer->orig.
  */
-static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
+static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out,
+		regex_t *word_regex)
 {
 	int i, j;
+	long alloc = 0;
 
 	out->size = 0;
-	out->ptr = xmalloc(buffer->text.size);
+	out->ptr = NULL;
 
 	/* fake an empty "0th" word */
 	ALLOC_GROW(buffer->orig, 1, buffer->orig_alloc);
@@ -390,11 +411,9 @@ static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 	buffer->orig_nr = 1;
 
 	for (i = 0; i < buffer->text.size; i++) {
-		if (isspace(buffer->text.ptr[i]))
+		j = find_word_boundary(&buffer->text, i, word_regex);
+		if (i == j)
 			continue;
-		for (j = i + 1; j < buffer->text.size &&
-				!isspace(buffer->text.ptr[j]); j++)
-			; /* find the end of the word */
 
 		/* store original boundaries */
 		ALLOC_GROW(buffer->orig, buffer->orig_nr + 1,
@@ -404,6 +423,7 @@ static void diff_words_fill(struct diff_words_buffer *buffer, mmfile_t *out)
 		buffer->orig_nr++;
 
 		/* store one word */
+		ALLOC_GROW(out->ptr, out->size + j - i + 1, alloc);
 		memcpy(out->ptr + out->size, buffer->text.ptr + i, j - i);
 		out->ptr[out->size + j - i] = '\n';
 		out->size += j - i + 1;
@@ -432,8 +452,8 @@ static void diff_words_show(struct diff_words_data *diff_words)
 
 	memset(&xpp, 0, sizeof(xpp));
 	memset(&xecfg, 0, sizeof(xecfg));
-	diff_words_fill(&diff_words->minus, &minus);
-	diff_words_fill(&diff_words->plus, &plus);
+	diff_words_fill(&diff_words->minus, &minus, diff_words->word_regex);
+	diff_words_fill(&diff_words->plus, &plus, diff_words->word_regex);
 	xpp.flags = XDF_NEED_MINIMAL;
 	xecfg.ctxlen = 0;
 	xdi_diff_outf(&minus, &plus, fn_out_diff_words_aux, diff_words,
@@ -473,6 +493,7 @@ static void free_diff_words_data(struct emit_callback *ecbdata)
 		free (ecbdata->diff_words->minus.orig);
 		free (ecbdata->diff_words->plus.text.ptr);
 		free (ecbdata->diff_words->plus.orig);
+		free(ecbdata->diff_words->word_regex);
 		free(ecbdata->diff_words);
 		ecbdata->diff_words = NULL;
 	}
@@ -1495,6 +1516,14 @@ static void builtin_diff(const char *name_a,
 			ecbdata.diff_words =
 				xcalloc(1, sizeof(struct diff_words_data));
 			ecbdata.diff_words->file = o->file;
+			if (o->word_regex) {
+				ecbdata.diff_words->word_regex = (regex_t *)
+					xmalloc(sizeof(regex_t));
+				if (regcomp(ecbdata.diff_words->word_regex,
+						o->word_regex, REG_EXTENDED))
+					die ("Invalid regular expression: %s",
+							o->word_regex);
+			}
 		}
 		xdi_diff_outf(&mf1, &mf2, fn_out_consume, &ecbdata,
 			      &xpp, &xecfg, &ecb);
@@ -2510,6 +2539,10 @@ int diff_opt_parse(struct diff_options *options, const char **av, int ac)
 		DIFF_OPT_CLR(options, COLOR_DIFF);
 	else if (!strcmp(arg, "--color-words"))
 		options->flags |= DIFF_OPT_COLOR_DIFF | DIFF_OPT_COLOR_DIFF_WORDS;
+	else if (!prefixcmp(arg, "--color-words=")) {
+		options->flags |= DIFF_OPT_COLOR_DIFF | DIFF_OPT_COLOR_DIFF_WORDS;
+		options->word_regex = arg + 14;
+	}
 	else if (!strcmp(arg, "--exit-code"))
 		DIFF_OPT_SET(options, EXIT_WITH_STATUS);
 	else if (!strcmp(arg, "--quiet"))
diff --git a/diff.h b/diff.h
index 4d5a327..23cd90c 100644
--- a/diff.h
+++ b/diff.h
@@ -98,6 +98,7 @@ struct diff_options {
 
 	int stat_width;
 	int stat_name_width;
+	const char *word_regex;
 
 	/* this is set by diffcore for DIFF_FORMAT_PATCH */
 	int found_changes;

diff --git a/diff.c b/diff.c
index d6bba72..676eb79 100644
--- a/diff.c
+++ b/diff.c
@@ -381,8 +381,10 @@ static int find_word_boundary(mmfile_t *buffer, int i, regex_t *word_regex)
 
 	if (word_regex) {
 		regmatch_t match[1];
-		if (!regexec(word_regex, buffer->ptr + i, 1, match, 0))
-			i += match[0].rm_eo;
+		if (!regexec(word_regex, buffer->ptr + i, 1, match, 0)) {
+			char *p = memchr(buffer->ptr + i, '\n', match[0].rm_eo);
+			i = p ? p - buffer->ptr : match[0].rm_eo + i;
+		}
 	}
 	else
 		while (i < buffer->size && !isspace(buffer->ptr[i]))