index : cgit

/* ui-diff.c: show diff between two blobs
 *
 * Copyright (C) 2006-2014 cgit Development Team <cgit@lists.zx2c4.com>
 *
 * Licensed under GNU General Public License v2
 *   (see LICENSE.txt for full license text)
 */

#define USE_THE_REPOSITORY_VARIABLE

#include "cgit.h"
#include "ui-diff.h"
#include "html.h"
#include "ui-shared.h"
#include "ui-ssdiff.h"

struct fileinfo {
	char status;
	struct object_id old_oid[1];
	struct object_id new_oid[1];
	unsigned short old_mode;
	unsigned short new_mode;
	char *old_path;
	char *new_path;
	unsigned int added;
	unsigned int removed;
	unsigned long old_size;
	unsigned long new_size;
	unsigned int binary:1;
	struct strbuf body;
};

/*
 * The diffstat has to be printed before the file bodies, but both are produced
 * by the same walk over the tree. So each body is rendered into the item it
 * belongs to while that file is already open, and replayed once the stat table
 * above it has been written. Rendering separately meant a second walk with its
 * own rename detection and its own xdiff of every file.
 *
 * A body stops being collected past max-diff-lines, since a file over that is
 * replaced by a link, and collection stops altogether past the budget below,
 * after which the bodies are produced the old way. The budget is a bound on
 * one request, not something to configure.
 *
 * Only a view that caps its bodies collects them. Without max-diff-lines to
 * stop it, a single body grows with the file it came from, which for a
 * side-by-side diff is several times the blob itself, and the budget below
 * cannot help because it is only reached once a body is already complete.
 */
#define CGIT_DIFF_BODY_BUDGET (8 * 1024 * 1024)

/* The revisions being compared, read by ui-ssdiff when it builds line links. */
struct object_id old_rev_oid[1];
struct object_id new_rev_oid[1];

/* One entry per file in the diff, filled by the walk and replayed afterwards. */
static struct fileinfo *items;
static int files, slots;

/* Totals for the diffstat, accumulated across that same walk. */
static int total_adds, total_rems, max_changes;
static int lines_added, lines_removed;

/* How much body text has been collected, and whether collecting is still
 * worthwhile. Once the budget is passed the bodies are produced the old way. */
static size_t body_bytes;
static int bodies_usable;

/* Which renderer the current file's lines go to, and whether the file has
 * already passed max-diff-lines and so stopped being rendered. */
static linediff_fn render_line_fn;
static int render_suppressed;

/* The file being rendered right now, and the path the view is held to. */
static struct diff_filepair *current_filepair;
static const char *current_prefix;

static int use_ssdiff;

/* Caps apply only to whole-commit views. A single-file diff page must
 * always render fully, since it is where the capped views link to. */
static int cap_diffs;
static int item_idx;

/* The bodies have either been replayed or been given up on, so in both cases
 * what they hold is finished with. */
static void release_bodies(void)
{
	int i;

	for (i = 0; i < files; i++)
		strbuf_release(&items[i].body);
}

struct diff_filespec *cgit_get_current_old_file(void)
{
	return current_filepair->one;
}

struct diff_filespec *cgit_get_current_new_file(void)
{
	return current_filepair->two;
}

static void print_fileinfo(struct fileinfo *info)
{
	const char *class;

	switch (info->status) {
	case DIFF_STATUS_ADDED:
		class = "add";
		break;
	case DIFF_STATUS_COPIED:
		class = "cpy";
		break;
	case DIFF_STATUS_DELETED:
		class = "del";
		break;
	case DIFF_STATUS_MODIFIED:
		class = "upd";
		break;
	case DIFF_STATUS_RENAMED:
		class = "mov";
		break;
	case DIFF_STATUS_TYPE_CHANGED:
		class = "typ";
		break;
	case DIFF_STATUS_UNKNOWN:
		class = "unk";
		break;
	case DIFF_STATUS_UNMERGED:
		class = "stg";
		break;
	default:
		die("bug: unhandled diff status %c", info->status);
	}

	html("<tr>");
	html("<td class='mode'>");
	if (is_null_oid(info->new_oid)) {
		cgit_print_filemode(info->old_mode);
	} else {
		cgit_print_filemode(info->new_mode);
	}

	if (info->old_mode != info->new_mode &&
	    !is_null_oid(info->old_oid) &&
	    !is_null_oid(info->new_oid)) {
		html("<span class='modechange'>[");
		cgit_print_filemode(info->old_mode);
		html("]</span>");
	}
	htmlf("</td><td class='%s'>", class);
	cgit_diff_link(info->new_path, NULL, NULL, ctx.qry.head, ctx.qry.oid,
		       ctx.qry.oid2, info->new_path);
	if (info->status == DIFF_STATUS_COPIED || info->status == DIFF_STATUS_RENAMED) {
		htmlf(" (%s from ",
		      info->status == DIFF_STATUS_COPIED ? "copied" : "renamed");
		html_txt(info->old_path);
		html(")");
	}
	html("</td><td class='right'>");
	if (info->binary) {
		htmlf("bin</td><td class='graph'>%ld -> %ld bytes",
		      info->old_size, info->new_size);
		return;
	}
	htmlf("%d", info->added + info->removed);
	html("</td><td class='graph'>");
	htmlf("<table summary='file diffstat' width='%d%%'><tr>", (max_changes > 100 ? 100 : max_changes));
	htmlf("<td class='add' style='width: %.1f%%;'/>",
	      info->added * 100.0 / (max_changes ? max_changes : 1));
	htmlf("<td class='rem' style='width: %.1f%%;'/>",
	      info->removed * 100.0 / (max_changes ? max_changes : 1));
	htmlf("<td class='none' style='width: %.1f%%;'/>",
	      (max_changes - info->removed - info->added) * 100.0 / (max_changes ? max_changes : 1));
	html("</tr></table></td></tr>\n");
}

/* Counts every line, and renders it too until the file passes max-diff-lines,
 * at which point the body is going to be replaced by a link anyway. */
static void count_diff_lines(char *line, int len)
{
	if (line && (len > 0)) {
		if (line[0] == '+')
			lines_added++;
		else if (line[0] == '-')
			lines_removed++;
	}
	if (!render_line_fn || render_suppressed)
		return;
	if (cap_diffs && ctx.cfg.max_diff_lines > 0 &&
	    lines_added + lines_removed > ctx.cfg.max_diff_lines) {
		render_suppressed = 1;
		return;
	}
	render_line_fn(line, len);
}

static int show_filepair(struct diff_filepair *pair)
{
	/* Always show if we have no limiting prefix. */
	if (!current_prefix)
		return 1;

	/* Show if either path in the pair begins with the prefix. */
	if (starts_with(pair->one->path, current_prefix) ||
	    starts_with(pair->two->path, current_prefix))
		return 1;

	/* Otherwise we don't want to show this filepair. */
	return 0;
}

static void print_line(char *line, int len);
static void header(const struct object_id *oid1, char *path1, int mode1,
		   const struct object_id *oid2, char *path2, int mode2);

/* Render one file's body into its own buffer, the way filepair_cb would have
 * written it straight out on a second walk. The line count is not known until
 * the diff has run, so the header is collected first and what follows it is
 * decided afterwards. */
static void collect_filepair_body(struct diff_filepair *pair, int idx,
				  int *binary, unsigned long *old_size,
				  unsigned long *new_size)
{
	struct strbuf *body = &items[idx].body;
	linediff_fn line_fn = use_ssdiff ? cgit_ssdiff_line_cb : print_line;
	size_t header_len;

	current_filepair = pair;

	html_capture_begin(body);
	if (use_ssdiff)
		cgit_ssdiff_header_begin();
	header(&pair->one->oid, pair->one->path, pair->one->mode,
	       &pair->two->oid, pair->two->path, pair->two->mode);
	if (use_ssdiff)
		cgit_ssdiff_header_end();
	// Everything from here on is dropped if the file turns out to be over
	// the line budget, so remember where the header ended.
	header_len = body->len;

	if (S_ISGITLINK(pair->one->mode) || S_ISGITLINK(pair->two->mode)) {
		/* The body shows the two Subproject lines, but the stat has
		 * always counted what a diff of the pair produces, so run it
		 * for the count alone. */
		render_line_fn = NULL;
		cgit_diff_files(&pair->one->oid, &pair->two->oid, old_size,
				new_size, binary, 0, ctx.qry.ignorews,
				count_diff_lines);
		if (S_ISGITLINK(pair->one->mode)) {
			char *l = cgit_fmt("-Subproject %s", oid_to_hex(&pair->one->oid));
			line_fn(l, strlen(l) + 1);
		}
		if (S_ISGITLINK(pair->two->mode)) {
			char *l = cgit_fmt("+Subproject %s", oid_to_hex(&pair->two->oid));
			line_fn(l, strlen(l) + 1);
		}
	} else {
		render_line_fn = line_fn;
		render_suppressed = 0;
		if (cgit_diff_files(&pair->one->oid, &pair->two->oid, old_size,
				    new_size, binary, ctx.qry.context,
				    ctx.qry.ignorews, count_diff_lines))
			cgit_print_error("Error running diff");
		render_line_fn = NULL;
		if (*binary) {
			if (use_ssdiff)
				html("<tr><td colspan='4'>Binary files differ</td></tr>");
			else
				html("Binary files differ");
		}
	}
	if (use_ssdiff)
		cgit_ssdiff_footer();

	if (render_suppressed) {
		/* Over the line budget, so the body is a link instead. Setting
		 * the length back would leave the buffer holding everything it
		 * grew to while rendering, which the budget below cannot see
		 * because it only counts what is kept, so rebuild it at the
		 * size actually kept. */
		char *header_text = xmemdupz(body->buf, header_len);

		strbuf_release(body);
		strbuf_attach(body, header_text, header_len, header_len + 1);
		if (use_ssdiff)
			html("<tr><td colspan='4' class='truncated'>");
		else
			html("<div class='truncated'>");
		html("This diff is too large to be rendered inline. ");
		cgit_diff_link("View it on its own page", NULL, NULL,
			       ctx.qry.head, ctx.qry.oid, ctx.qry.oid2,
			       pair->two->path);
		html(".");
		if (use_ssdiff) {
			html("</td></tr>");
			cgit_ssdiff_footer();
		} else
			html("</div>");
	}
	html_capture_end();

	body_bytes += body->len;
	if (body_bytes > CGIT_DIFF_BODY_BUDGET) {
		bodies_usable = 0;
		release_bodies();
	}
}

static void inspect_filepair(struct diff_filepair *pair)
{
	int binary = 0;
	unsigned long old_size = 0;
	unsigned long new_size = 0;

	if (!show_filepair(pair))
		return;

	files++;
	lines_added = 0;
	lines_removed = 0;
	if (files >= slots) {
		if (slots == 0)
			slots = 4;
		else
			slots = slots * 2;
		items = xrealloc(items, slots * sizeof(struct fileinfo));
	}
	memset(&items[files-1], 0, sizeof(items[files-1]));
	strbuf_init(&items[files-1].body, 0);

	if (bodies_usable)
		collect_filepair_body(pair, files - 1, &binary,
				      &old_size, &new_size);
	else
		cgit_diff_files(&pair->one->oid, &pair->two->oid, &old_size,
				&new_size, &binary, 0, ctx.qry.ignorews,
				count_diff_lines);

	items[files-1].status = pair->status;
	oidcpy(items[files-1].old_oid, &pair->one->oid);
	oidcpy(items[files-1].new_oid, &pair->two->oid);
	items[files-1].old_mode = pair->one->mode;
	items[files-1].new_mode = pair->two->mode;
	items[files-1].old_path = xstrdup(pair->one->path);
	items[files-1].new_path = xstrdup(pair->two->path);
	items[files-1].added = lines_added;
	items[files-1].removed = lines_removed;
	items[files-1].old_size = old_size;
	items[files-1].new_size = new_size;
	items[files-1].binary = binary;
	if (lines_added + lines_removed > max_changes)
		max_changes = lines_added + lines_removed;
	total_adds += lines_added;
	total_rems += lines_removed;
}

static void cgit_print_diffstat(const struct object_id *old_oid,
				const struct object_id *new_oid,
				const char *prefix)
{
	int i;

	html("<div class='diffstat-header'>");
	cgit_diff_link("Diffstat", NULL, NULL, ctx.qry.head, ctx.qry.oid,
		       ctx.qry.oid2, NULL);
	if (prefix) {
		html(" (limited to '");
		html_txt(prefix);
		html("')");
	}
	html("</div>");
	html("<table summary='diffstat' class='diffstat'>");
	max_changes = 0;
	cgit_diff_tree(old_oid, new_oid, inspect_filepair, prefix,
		       ctx.qry.ignorews);
	for (i = 0; i<files; i++)
		print_fileinfo(&items[i]);
	html("</table>");
	html("<div class='diffstat-summary'>");
	htmlf("%d files changed, %d insertions, %d deletions",
	      files, total_adds, total_rems);
	html("</div>");
}

/*
 * print a single line returned from xdiff
 */
static void print_line(char *line, int len)
{
	const char *class = "ctx";
	char c = line[len-1];

	if (line[0] == '+')
		class = "add";
	else if (line[0] == '-')
		class = "del";
	else if (line[0] == '@')
		class = "hunk";

	htmlf("<div class='%s'>", class);
	line[len-1] = '\0';
	html_txt(line);
	html("</div>");
	line[len-1] = c;
}

static void header(const struct object_id *oid1, char *path1, int mode1,
		   const struct object_id *oid2, char *path2, int mode2)
{
	char *abbrev1, *abbrev2;
	int subproject;

	subproject = (S_ISGITLINK(mode1) || S_ISGITLINK(mode2));
	html("<div class='head'>");
	html("diff --git a/");
	html_txt(path1);
	html(" b/");
	html_txt(path2);

	if (mode1 == 0)
		htmlf("<br/>new file mode %.6o", mode2);

	if (mode2 == 0)
		htmlf("<br/>deleted file mode %.6o", mode1);

	if (!subproject) {
		abbrev1 = xstrdup(repo_find_unique_abbrev(the_repository, oid1, DEFAULT_ABBREV));
		abbrev2 = xstrdup(repo_find_unique_abbrev(the_repository, oid2, DEFAULT_ABBREV));
		htmlf("<br/>index %s..%s", abbrev1, abbrev2);
		free(abbrev1);
		free(abbrev2);
		if (mode1 != 0 && mode2 != 0) {
			htmlf(" %.6o", mode1);
			if (mode2 != mode1)
				htmlf("..%.6o", mode2);
		}
		if (is_null_oid(oid1)) {
			path1 = "dev/null";
			html("<br/>--- /");
		} else
			html("<br/>--- a/");
		if (mode1 != 0)
			cgit_tree_link(path1, NULL, NULL, ctx.qry.head,
				       oid_to_hex(old_rev_oid), path1);
		else
			html_txt(path1);
		if (is_null_oid(oid2)) {
			path2 = "dev/null";
			html("<br/>+++ /");
		} else
			html("<br/>+++ b/");
		if (mode2 != 0)
			cgit_tree_link(path2, NULL, NULL, ctx.qry.head,
				       oid_to_hex(new_rev_oid), path2);
		else
			html_txt(path2);
	}
	html("</div>");
}

static void filepair_cb(struct diff_filepair *pair)
{
	unsigned long old_size = 0;
	unsigned long new_size = 0;
	int binary = 0;
	int idx;
	linediff_fn print_line_fn = print_line;

	if (!show_filepair(pair))
		return;

	idx = item_idx++;
	current_filepair = pair;
	if (use_ssdiff) {
		cgit_ssdiff_header_begin();
		print_line_fn = cgit_ssdiff_line_cb;
	}
	header(&pair->one->oid, pair->one->path, pair->one->mode,
	       &pair->two->oid, pair->two->path, pair->two->mode);
	if (use_ssdiff)
		cgit_ssdiff_header_end();

	/* The diffstat pass counted this file's lines. Over the budget,
	 * print the header and hand off to the single-file diff page
	 * instead of rendering the whole thing inline. */
	if (cap_diffs && ctx.cfg.max_diff_lines > 0 && idx < files &&
	    !items[idx].binary && items[idx].added + items[idx].removed >
				  (unsigned int)ctx.cfg.max_diff_lines) {
		if (use_ssdiff)
			html("<tr><td colspan='4' class='truncated'>");
		else
			html("<div class='truncated'>");
		html("This diff is too large to be rendered inline. ");
		cgit_diff_link("View it on its own page", NULL, NULL,
			       ctx.qry.head, ctx.qry.oid, ctx.qry.oid2,
			       items[idx].new_path);
		html(".");
		if (use_ssdiff) {
			html("</td></tr>");
			cgit_ssdiff_footer();
		} else
			html("</div>");
		return;
	}
	if (S_ISGITLINK(pair->one->mode) || S_ISGITLINK(pair->two->mode)) {
		if (S_ISGITLINK(pair->one->mode)) {
			char *l = cgit_fmt("-Subproject %s", oid_to_hex(&pair->one->oid));
			print_line_fn(l, strlen(l) + 1);
		}
		if (S_ISGITLINK(pair->two->mode)) {
			char *l = cgit_fmt("+Subproject %s", oid_to_hex(&pair->two->oid));
			print_line_fn(l, strlen(l) + 1);
		}
		if (use_ssdiff)
			cgit_ssdiff_footer();
		return;
	}
	if (cgit_diff_files(&pair->one->oid, &pair->two->oid, &old_size,
			    &new_size, &binary, ctx.qry.context,
			    ctx.qry.ignorews, print_line_fn))
		cgit_print_error("Error running diff");
	if (binary) {
		if (use_ssdiff)
			html("<tr><td colspan='4'>Binary files differ</td></tr>");
		else
			html("Binary files differ");
	}
	if (use_ssdiff)
		cgit_ssdiff_footer();
}

void cgit_print_diff_ctrls(void)
{
	int i, curr;

	html("<div class='cgit-panel'>");
	html("<b>diff options</b>");
	html("<form method='get'>");
	cgit_add_hidden_formfields(1, 0, ctx.qry.page);
	html("<table>");
	html("<tr><td colspan='2'/></tr>");
	html("<tr>");
	html("<td class='label'>context:</td>");
	html("<td class='ctrl'>");
	html("<select name='context' data-autosubmit='1'>");
	curr = ctx.qry.context;
	if (!curr)
		curr = 3;
	for (i = 1; i <= 10; i++)
		html_intoption(i, cgit_fmt("%d", i), curr);
	for (i = 15; i <= 40; i += 5)
		html_intoption(i, cgit_fmt("%d", i), curr);
	html("</select>");
	html("</td>");
	html("</tr><tr>");
	html("<td class='label'>space:</td>");
	html("<td class='ctrl'>");
	html("<select name='ignorews' data-autosubmit='1'>");
	html_intoption(0, "include", ctx.qry.ignorews);
	html_intoption(1, "ignore", ctx.qry.ignorews);
	html("</select>");
	html("</td>");
	html("</tr><tr>");
	html("<td class='label'>mode:</td>");
	html("<td class='ctrl'>");
	html("<select name='dt' data-autosubmit='1'>");
	curr = ctx.qry.has_difftype ? ctx.qry.difftype : ctx.cfg.difftype;
	html_intoption(0, "unified", curr);
	html_intoption(1, "ssdiff", curr);
	html_intoption(2, "stat only", curr);
	html("</select></td></tr>");
	html("<tr><td/><td class='ctrl'>");
	html("<noscript><input type='submit' value='reload'/></noscript>");
	html("</td></tr></table>");
	html("</form>");
	html("</div>");
}

void cgit_print_diff(const char *new_rev, const char *old_rev,
		     const char *prefix, int show_ctrls, int raw)
{
	struct commit *commit, *commit2;
	const struct object_id *old_tree_oid, *new_tree_oid;
	diff_type difftype;

	// A path-limited diff is naturally bounded, so the size caps only
	// apply to a whole-commit diff. Decide that from the caller's prefix
	// before the follow logic rewrites it to "", otherwise follow=1
	// silently disables the caps.
	cap_diffs = !prefix;

	// If "follow" is set then the diff machinery needs to examine the
	// entire commit to detect renames so we must limit the paths in our
	// own callbacks and not pass the prefix to the diff machinery.
	if (ctx.qry.follow && ctx.repo->enable_follow_links) {
		current_prefix = prefix;
		prefix = "";
	} else {
		current_prefix = NULL;
	}

	if (!new_rev)
		new_rev = ctx.qry.head;
	if (repo_get_oid(the_repository, new_rev, new_rev_oid)) {
		cgit_print_error_page(404, "Not found",
			"Bad object name: %s", new_rev);
		return;
	}
	commit = lookup_commit_reference(the_repository, new_rev_oid);
	if (!commit || repo_parse_commit(the_repository, commit)) {
		cgit_print_error_page(404, "Not found",
			"Bad commit: %s", oid_to_hex(new_rev_oid));
		return;
	}
	new_tree_oid = get_commit_tree_oid(commit);

	if (old_rev) {
		if (repo_get_oid(the_repository, old_rev, old_rev_oid)) {
			cgit_print_error_page(404, "Not found",
				"Bad object name: %s", old_rev);
			return;
		}
	} else if (commit->parents && commit->parents->item) {
		oidcpy(old_rev_oid, &commit->parents->item->object.oid);
	} else {
		oidclr(old_rev_oid, the_repository->hash_algo);
	}

	if (!is_null_oid(old_rev_oid)) {
		commit2 = lookup_commit_reference(the_repository, old_rev_oid);
		if (!commit2 || repo_parse_commit(the_repository, commit2)) {
			cgit_print_error_page(404, "Not found",
				"Bad commit: %s", oid_to_hex(old_rev_oid));
			return;
		}
		old_tree_oid = get_commit_tree_oid(commit2);
	} else {
		old_tree_oid = NULL;
	}

	if (raw) {
		struct diff_options diffopt;

		repo_diff_setup(the_repository, &diffopt);
		diffopt.output_format = DIFF_FORMAT_PATCH;
		diffopt.flags.recursive = 1;
		diff_setup_done(&diffopt);

		ctx.page.mimetype = "text/plain";
		cgit_print_http_headers();
		if (old_tree_oid) {
			diff_tree_oid(old_tree_oid, new_tree_oid, "",
				       &diffopt);
		} else {
			diff_root_tree_oid(new_tree_oid, "", &diffopt);
		}
		diffcore_std(&diffopt);
		diff_flush(&diffopt);

		return;
	}

	difftype = ctx.qry.has_difftype ? ctx.qry.difftype : ctx.cfg.difftype;
	use_ssdiff = difftype == DIFF_SSDIFF;
	/* A stat-only view never shows a body, and only a capped view bounds
	 * the size of the ones it does show. */
	bodies_usable = difftype != DIFF_STATONLY && cap_diffs;

	if (show_ctrls) {
		cgit_print_layout_start();
		cgit_print_diff_ctrls();
	}

	/*
	 * Clicking on a link to a file in the diff stat should show a diff
	 * of the file, showing the diff stat limited to a single file is
	 * pretty useless.  All links from this point on will be to
	 * individual files, so we simply reset the difftype in the query
	 * here to avoid propagating DIFF_STATONLY to the individual files.
	 */
	if (difftype == DIFF_STATONLY)
		ctx.qry.difftype = ctx.cfg.difftype;

	cgit_print_diffstat(old_rev_oid, new_rev_oid, prefix);

	if (difftype == DIFF_STATONLY) {
		if (show_ctrls)
			cgit_print_layout_end();
		return;
	}

	/* A commit touching more files than max-diff-files stops at the
	 * stat above, where every file links to its own diff page. */
	if (cap_diffs && ctx.cfg.max_diff_files > 0 &&
	    files > ctx.cfg.max_diff_files) {
		html("<div class='truncated'>");
		html("This diff is too large to be rendered inline. "
		     "Follow a file above, or the ");
		cgit_patch_link("patch", NULL, NULL, NULL, ctx.qry.oid, NULL);
		html(" link for the whole commit.");
		html("</div>");
		if (show_ctrls)
			cgit_print_layout_end();
		return;
	}

	if (use_ssdiff) {
		html("<table summary='ssdiff' class='ssdiff'>");
	} else {
		html("<table summary='diff' class='diff'>");
		html("<tr><td>");
	}
	if (bodies_usable) {
		int i;
		for (i = 0; i < files; i++)
			html_raw(items[i].body.buf, items[i].body.len);
		release_bodies();
	} else {
		/* The bodies outgrew the budget, so they are produced the way
		 * they were before, by walking the tree again. */
		item_idx = 0;
		cgit_diff_tree(old_rev_oid, new_rev_oid, filepair_cb, prefix,
			       ctx.qry.ignorews);
	}
	if (!use_ssdiff)
		html("</td></tr>");
	html("</table>");

	if (show_ctrls)
		cgit_print_layout_end();
}