index : cgit

/*
 * The blame page, which shows a file next to the commit that last touched
 * each of its lines. git returns blame as runs of neighbouring lines that
 * share a commit, and the page turns every run into one block in each column
 * so the hashes, the line numbers and the striped background all line up with
 * the source. Only a regular file can be blamed, so a path naming a folder is
 * turned away.
 */

#define USE_THE_REPOSITORY_VARIABLE

#include "cgit.h"
#include "filter.h"
#include "html.h"
#include "parsing.h"
#include "shared.h"
#include "ui-blame.h"
#include "ui-shared.h"

// A tab in the rendered source runs on to the next multiple of this. The
// stylesheet leaves tab-size alone, so the measurement has to match what the
// browser does on its own rather than anything cgit picks.
#define TAB_WIDTH 8

enum blame_target {
	TARGET_MISSING,
	TARGET_FILE,
	TARGET_FOLDER,
};

struct walk_tree_context {
	char *rev;
	int match_baselen;
	enum blame_target found;
};

// A commit that touched several separate parts of the file comes back once
// per part, and each repeat would parse the commit again, so the rendered
// detail is cached here and looked up by object id.
static struct string_list suspect_details = STRING_LIST_INIT_DUP;

static void free_suspect_details(void)
{
	struct string_list_item *item;

	for_each_string_list_item(item, &suspect_details)
		free(item->util);
	string_list_clear(&suspect_details, 0);
}

/*
 * The scoreboard keeps a pointer to revs, so the caller owns both and has to
 * keep them alive together.
 */
static void run_blame(struct blame_scoreboard *sb, struct rev_info *revs,
		      const char *path, const char *rev)
{
	struct strvec argv = STRVEC_INIT;
	struct blame_origin *origin;

	strvec_push(&argv, "blame");
	strvec_push(&argv, rev);
	repo_init_revisions(the_repository, revs, NULL);
	revs->diffopt.flags.allow_textconv = 1;
	setup_revisions(argv.nr, argv.v, revs, NULL);
	init_scoreboard(sb);
	sb->revs = revs;
	sb->repo = the_repository;
	sb->path = path;
	setup_scoreboard(sb, &origin);
	origin->suspects = blame_entry_prepend(NULL, 0, sb->num_lines, origin);
	prio_queue_put(&sb->commits, origin->commit);
	blame_origin_decref(origin);
	sb->ent = NULL;
	sb->path = path;
	assign_blame(sb, 0);
	blame_sort_final(sb);
	blame_coalesce(sb);
}

/*
 * A single blame entry can cover the whole file, so the run is handed on in
 * batches rather than built whole.
 */
static void emit_chars(char ch, unsigned long count)
{
	struct strbuf run = STRBUF_INIT;

	while (count) {
		unsigned long n = count < HTML_BATCH ? count : HTML_BATCH;

		strbuf_addchars(&run, ch, n);
		html_raw(run.buf, run.len);
		strbuf_reset(&run);
		count -= n;
	}
	strbuf_release(&run);
}

/*
 * The returned string belongs to suspect_details, not the caller.
 */
static char *suspect_detail(struct blame_origin *suspect)
{
	struct commitinfo *info;
	struct strbuf detail = STRBUF_INIT;
	struct string_list_item *cached;
	char key[GIT_MAX_HEXSZ + 1];

	// A private copy, since oid_to_hex hands back a buffer it reuses and
	// the commit parse below is free to call it again.
	oid_to_hex_r(key, &suspect->commit->object.oid);

	cached = string_list_lookup(&suspect_details, key);
	if (cached)
		return cached->util;

	info = cgit_parse_commit(suspect->commit);

	strbuf_addf(&detail, "author  %s", info->author);
	if (ctx.cfg.enable_plain_email)
		strbuf_addf(&detail, " %s", info->author_email);
	strbuf_addf(&detail, "  %s\n",
		    show_date(info->author_date, info->author_tz,
				    cgit_date_mode(DATE_ISO8601)));

	strbuf_addf(&detail, "committer  %s", info->committer);
	if (ctx.cfg.enable_plain_email)
		strbuf_addf(&detail, " %s", info->committer_email);
	strbuf_addf(&detail, "  %s\n\n",
		    show_date(info->committer_date, info->committer_tz,
				    cgit_date_mode(DATE_ISO8601)));

	strbuf_addstr(&detail, info->subject);

	cgit_free_commitinfo(info);
	cached = string_list_insert(&suspect_details, key);
	cached->util = strbuf_detach(&detail, NULL);
	return cached->util;
}

static void emit_entry_hash(struct blame_entry *ent)
{
	struct blame_origin *suspect = ent->suspect;
	struct object_id *oid = &suspect->commit->object.oid;
	const char *detail = suspect_detail(suspect);

	html("<span class='oid'>");
	cgit_commit_link(repo_find_unique_abbrev(the_repository, oid,
						 DEFAULT_ABBREV),
			 detail, NULL, ctx.qry.head, oid_to_hex(oid),
			 suspect->path);
	html("</span>");

	if (!repo_parse_commit(the_repository, suspect->commit) &&
	    suspect->commit->parents) {
		struct commit *parent = suspect->commit->parents->item;

		html(" ");
		cgit_blame_link("^", "Blame the previous revision", NULL,
				ctx.qry.head, oid_to_hex(&parent->object.oid),
				suspect->path);
	}

	// The stripes only line up across the columns if each column gives an
	// entry the same height, so pad out to the lines the entry covers.
	emit_chars('\n', ent->num_lines);
}

static void emit_hashes(struct blame_scoreboard *sb)
{
	struct blame_entry *ent;

	html("<td class='hashes'>");
	for (ent = sb->ent; ent; ent = ent->next) {
		html("<div class='alt'><pre>");
		emit_entry_hash(ent);
		html("</pre></div>");
	}
	html("</td>\n");
}

static void emit_entry_linenumbers(struct blame_entry *ent)
{
	const char *numberfmt = "<a id='n%1$d' href='#n%1$d'>%1$d</a>\n";
	struct strbuf numbers = STRBUF_INIT;
	int lineno = ent->lno;

	while (lineno < ent->lno + ent->num_lines) {
		strbuf_addf(&numbers, numberfmt, ++lineno);
		if (numbers.len >= HTML_BATCH) {
			html_raw(numbers.buf, numbers.len);
			strbuf_reset(&numbers);
		}
	}
	html_raw(numbers.buf, numbers.len);
	strbuf_release(&numbers);
}

static void emit_linenumbers(struct blame_scoreboard *sb)
{
	struct blame_entry *ent;

	html("<td class='linenumbers'>");
	for (ent = sb->ent; ent; ent = ent->next) {
		html("<div class='alt'><pre>");
		emit_entry_linenumbers(ent);
		html("</pre></div>");
	}
	html("</td>\n");
}

static size_t line_width(struct blame_scoreboard *sb, int line)
{
	const char *pos = blame_nth_line(sb, line);
	const char *end = blame_nth_line(sb, line + 1);
	size_t width = 0;

	while (pos < end) {
		width++;
		if (*pos++ == '\t')
			width = (width + TAB_WIDTH - 1) & ~(TAB_WIDTH - 1);
	}
	return width;
}

/*
 * The stylesheet takes the source pre out of flow and positions it over these
 * blocks, so nothing else gives the cell a size and each block has to be
 * padded to the height and the width of the lines it stands behind.
 */
static void emit_entry_background(struct blame_scoreboard *sb,
				  struct blame_entry *ent)
{
	size_t widest = 2;
	int line;

	for (line = ent->lno; line < ent->lno + ent->num_lines; line++) {
		size_t width = line_width(sb, line);

		if (width > widest)
			widest = width;
	}

	emit_chars('\n', ent->num_lines);
	emit_chars(' ', widest - 1);
}

/*
 * Frees each entry on the way past, so this has to be the last pass over the
 * scoreboard's entries.
 */
static void emit_line_backgrounds(struct blame_scoreboard *sb)
{
	struct blame_entry *ent = sb->ent;

	html("<div>");
	while (ent) {
		struct blame_entry *next = ent->next;

		html("<div class='alt'><pre>");
		emit_entry_background(sb, ent);
		html("</pre></div>");
		free(ent);
		ent = next;
	}
	html("</div>");
}

static void print_blame_page(const struct object_id *oid, const char *path,
			     const char *filename, const char *rev)
{
	enum object_type type;
	char *buf;
	unsigned long size;
	struct rev_info revs;
	struct blame_scoreboard sb;

	type = odb_read_object_info(the_repository->objects, oid, &size);
	if (type == OBJ_BAD) {
		cgit_print_error_page(404, "Not Found", "Bad object name: %s",
				      oid_to_hex(oid));
		return;
	}
	if (ctx.cfg.max_blob_size && size / 1024 > (unsigned long)ctx.cfg.max_blob_size) {
		cgit_print_error_page(413, "Content Too Large",
			"blob size (%luKB) exceeds display size limit (%dKB)",
			size / 1024, ctx.cfg.max_blob_size);
		return;
	}

	buf = odb_read_object(the_repository->objects, oid, &type, &size);
	if (!buf) {
		cgit_print_error_page(500, "Internal Server Error",
			"Error reading object %s", oid_to_hex(oid));
		return;
	}

	run_blame(&sb, &revs, path, rev);

	cgit_set_title_from_path(path);

	cgit_print_layout_start();
	htmlf("blob: %s (", oid_to_hex(oid));
	cgit_plain_link("plain", NULL, NULL, ctx.qry.head, rev, path);
	html(") (");
	cgit_tree_link("tree", NULL, NULL, ctx.qry.head, rev, path);
	html(")\n");

	// A NUL past the window buffer_is_binary sniffs would end html_txt
	// early while the hash and line number columns still cover the whole
	// file, so a blob holding one anywhere is treated as binary too.
	if (buffer_is_binary(buf, size) || memchr(buf, 0, size)) {
		struct blame_entry *ent = sb.ent, *next;

		// The scoreboard is normally consumed by the column passes,
		// so this early exit has to release it itself.
		for (; ent; ent = next) {
			next = ent->next;
			free(ent);
		}
		free((void *)sb.final_buf);
		html("<div class='error'>blob is binary.</div>");
		goto cleanup;
	}
	html("<table class='blame blob'>\n<tr>\n");

	emit_hashes(&sb);

	if (ctx.cfg.enable_tree_linenumbers)
		emit_linenumbers(&sb);

	html("<td class='lines'><div>");
	emit_line_backgrounds(&sb);

	free((void *)sb.final_buf);

	html("<pre><code>");
	if (ctx.repo->source_filter) {
		char *filter_arg = xstrdup(filename);
		cgit_open_filter(ctx.repo->source_filter, filter_arg);
		html_raw(buf, size);
		cgit_close_filter(ctx.repo->source_filter);
		free(filter_arg);
	} else {
		html_txt(buf);
	}
	html("</code></pre>");

	html("</div></td>\n");

	html("</tr>\n</table>\n");

cleanup:
	cgit_print_layout_end();
	free_suspect_details();
	free(buf);
}

static int walk_tree(const struct object_id *oid, struct strbuf *base,
		     const char *pathname, unsigned mode, void *data)
{
	struct walk_tree_context *walk = data;

	// match_baselen is -1 when no path was given, which no length equals.
	if (walk->match_baselen >= 0 &&
	    base->len == (size_t)walk->match_baselen) {
		if (S_ISREG(mode)) {
			struct strbuf fullpath = STRBUF_INIT;

			strbuf_addbuf(&fullpath, base);
			strbuf_addstr(&fullpath, pathname);
			print_blame_page(oid, fullpath.buf, pathname,
					 walk->rev);
			strbuf_release(&fullpath);
			walk->found = TARGET_FILE;
		} else if (S_ISDIR(mode)) {
			walk->found = TARGET_FOLDER;
		}
	} else if (base->len < INT_MAX
			&& (int)base->len > walk->match_baselen) {
		walk->found = TARGET_FOLDER;
	} else if (S_ISDIR(mode)) {
		return READ_TREE_RECURSIVE;
	}
	return 0;
}

static int basedir_len(const char *path)
{
	const char *slash = strrchr(path, '/');

	if (slash)
		return slash - path + 1;
	return 0;
}

void cgit_print_blame(void)
{
	const char *rev = ctx.qry.oid;
	struct object_id oid;
	struct commit *commit;
	int path_len = ctx.qry.path ? strlen(ctx.qry.path) : 0;
	// nowildcard_len matches len so git treats the path as literal rather
	// than as a glob, which is what a request naming one file means.
	struct pathspec_item path_items = {
		.match = ctx.qry.path,
		.len = path_len,
		.nowildcard_len = path_len
	};
	struct pathspec paths = {
		.nr = 1,
		.items = &path_items
	};
	struct walk_tree_context walk = {
		.found = TARGET_MISSING
	};

	if (!rev)
		rev = ctx.qry.head;

	if (repo_get_oid(the_repository, rev, &oid)) {
		cgit_print_error_page(404, "Not Found",
			"Invalid revision name: %s", rev);
		return;
	}
	commit = lookup_commit_reference(the_repository, &oid);
	if (!commit || repo_parse_commit(the_repository, commit)) {
		cgit_print_error_page(404, "Not Found",
			"Invalid commit reference: %s", rev);
		return;
	}

	walk.rev = xstrdup(rev);
	walk.match_baselen = path_items.match ?
			     basedir_len(path_items.match) : -1;

	read_tree(the_repository, repo_get_commit_tree(the_repository, commit),
		  &paths, walk_tree, &walk);
	if (walk.found == TARGET_MISSING)
		cgit_print_error_page(404, "Not Found", "Not found");
	else if (walk.found == TARGET_FOLDER)
		cgit_print_error_page(404, "Not Found",
			"Blame is not available for folders.");

	free(walk.rev);
}