index : cgit

/*
 * The Atom feed for a repository, which lists recent commits so a reader can
 * follow the project from a feed reader instead of the browsable pages. The
 * response is XML rather than a page, so this file writes its own HTTP headers
 * and never goes through the shared HTML layout.
 */

#define USE_THE_REPOSITORY_VARIABLE

#include "cgit.h"
#include "html.h"
#include "parsing.h"
#include "shared.h"
#include "ui-atom.h"
#include "ui-shared.h"

// Stands in for every byte the feed cannot carry.
#define XML_REPLACEMENT "?"

/*
 * Atom timestamps have to be RFC 3339, so a feed ignores the date-format and
 * local-time settings the browsable pages honour. The zero is the timezone
 * offset, which pins every feed date to UTC.
 */
static const char *feed_date(timestamp_t when)
{
	// The newest commit's date comes straight from git, unclamped, and
	// git's date code dies on one the system cannot hold.
	if (date_overflows(when))
		when = 0;
	return show_date(when, 0, date_mode_from_type(DATE_ISO8601_STRICT));
}

/*
 * The XML counterpart of html_txt. A browser shrugs at a stray control byte
 * or broken UTF-8, but an XML reader must reject the whole feed, so both
 * are replaced instead of passed through.
 */
static void xml_txt(const char *txt)
{
	const unsigned char *p = (const unsigned char *)txt;
	size_t left = txt ? strlen(txt) : 0;
	struct strbuf sb = STRBUF_INIT;

	while (left) {
		unsigned char c = *p;
		size_t seq;

		if (c == '<') {
			strbuf_addstr(&sb, "&lt;");
		} else if (c == '>') {
			strbuf_addstr(&sb, "&gt;");
		} else if (c == '&') {
			strbuf_addstr(&sb, "&amp;");
		} else if (c == '\t' || c == '\n' || c == '\r' || (c >= 0x20 && c < 0x80)) {
			strbuf_addch(&sb, c);
		} else if (c < 0x20) {
			strbuf_addstr(&sb, XML_REPLACEMENT);
		} else if ((seq = cgit_utf8_seq_len(p, left))) {
			// U+FFFE and U+FFFF are valid UTF-8 but not XML.
			if (seq == 3 && p[0] == 0xef && p[1] == 0xbf && p[2] >= 0xbe)
				strbuf_addstr(&sb, XML_REPLACEMENT);
			else
				strbuf_add(&sb, p, seq);
			p += seq - 1;
			left -= seq - 1;
		} else {
			strbuf_addstr(&sb, XML_REPLACEMENT);
		}
		p++;
		left--;
	}
	html_raw(sb.buf, sb.len);
	strbuf_release(&sb);
}

/*
 * cgit keeps an address with the angle brackets it was written with, while
 * Atom wants the bare address.
 */
static void print_email(const char *email)
{
	char *copy = xstrdup(email);
	char *start, *end;

	start = strchr(copy, '<');
	if (start)
		start++;
	else
		start = copy;
	end = strchr(start, '>');
	if (end)
		*end = '\0';

	// An empty element is no address at all, which Atom forbids.
	if (*start) {
		html("<email>");
		xml_txt(start);
		html("</email>\n");
	}
	free(copy);
}

static void print_entry(struct commit *commit, const char *host)
{
	struct commitinfo *info;
	char *hex, *pageurl;

	info = cgit_parse_commit(commit);
	hex = oid_to_hex(&commit->object.oid);
	html("<entry>\n");
	html("<title>");
	xml_txt(info->subject);
	html("</title>\n");
	html("<updated>");
	xml_txt(feed_date(info->committer_date));
	html("</updated>\n");
	html("<author>\n");
	// A person construct must hold a name, so a nameless commit falls
	// back to the address and then to a placeholder. An ident may carry
	// an empty name and an empty address, which count as missing.
	html("<name>");
	if (info->author && *info->author)
		xml_txt(info->author);
	else if (info->author_email && strcmp(info->author_email, "<>"))
		xml_txt(info->author_email);
	else
		html("unknown");
	html("</name>\n");
	if (info->author_email && ctx.cfg.enable_plain_email)
		print_email(info->author_email);
	html("</author>\n");
	html("<published>");
	xml_txt(feed_date(info->author_date));
	html("</published>\n");
	html("<link rel='alternate' type='text/html' href='");
	pageurl = cgit_pageurl(ctx.repo->url, "commit", cgit_fmt("id=%s", hex));
	html_attrf("%s%s%s", cgit_httpscheme(), host, pageurl);
	html("'/>\n");
	free(pageurl);
	html("<id>");
	html_txtf("urn:%s:%s", the_hash_algo->name, hex);
	html("</id>\n");
	html("<content type='text'>");
	xml_txt(info->msg);
	html("</content>\n");
	html("</entry>\n");
	cgit_free_commitinfo(info);
}

void cgit_print_atom(char *tip, const char *path, int max_count)
{
	char *host, *fullurl, *repourl;
	// setup_revisions reads a command line, so the first slot is the
	// unused program name and parsing starts at the second.
	const char *argv[] = { NULL, tip, NULL, NULL, NULL };
	struct commit *commit;
	struct rev_info rev;
	struct strbuf idbuf = STRBUF_INIT;
	int argc = 2;
	bool need_updated = true;

	// A leading colon makes git read the path as a pathspec with magic,
	// which can end the request inside git, so the feed refuses it as the
	// log does.
	if (path && path[0] == ':') {
		cgit_print_error_page(400, "Bad Request", "Invalid path");
		return;
	}

	if (ctx.qry.show_all)
		argv[1] = "--all";
	else if (!tip)
		argv[1] = ctx.qry.head;

	if (path) {
		argv[argc++] = "--";
		argv[argc++] = path;
	}

	repo_init_revisions(the_repository, &rev, NULL);
	rev.max_count = max_count;
	setup_revisions(argc, argv, &rev, NULL);
	// A failed setup leaves the walk holding freed commits, so it must
	// not be read from.
	if (prepare_revision_walk(&rev)) {
		cgit_print_error_page(500, "Internal Server Error", "Unable to read the history");
		return;
	}

	// CGI guarantees a server name, so only a bare test run reaches the
	// fallback, which keeps the mandatory feed id and links present.
	host = cgit_hosturl();
	if (!host)
		host = xstrdup("localhost");
	ctx.page.mimetype = "application/atom+xml";
	ctx.page.charset = PAGE_ENCODING;
	cgit_print_http_headers();
	html("<?xml version='1.0' encoding='UTF-8'?>\n");
	html("<feed xmlns='http://www.w3.org/2005/Atom'>\n");
	html("<title>");
	xml_txt(ctx.repo->name);
	if (path) {
		html("/");
		xml_txt(path);
	}
	if (tip && !ctx.qry.show_all) {
		html(", branch ");
		xml_txt(tip);
	}
	html("</title>\n");
	html("<subtitle>");
	xml_txt(ctx.repo->desc);
	html("</subtitle>\n");
	fullurl = cgit_currentfullurl();
	repourl = cgit_repourl(ctx.repo->url);
	strbuf_addf(&idbuf, "%s%s%s", cgit_httpscheme(), host, fullurl);
	html("<id>");
	xml_txt(idbuf.buf);
	html("</id>\n");
	strbuf_release(&idbuf);
	html("<link rel='self' href='");
	html_attrf("%s%s%s", cgit_httpscheme(), host, fullurl);
	html("'/>\n");
	html("<link rel='alternate' type='text/html' href='");
	html_attrf("%s%s%s", cgit_httpscheme(), host, repourl);
	html("'/>\n");
	free(fullurl);
	free(repourl);
	while ((commit = get_revision(&rev))) {
		if (need_updated) {
			html("<updated>");
			xml_txt(feed_date(commit->date));
			html("</updated>\n");
			need_updated = false;
		}
		print_entry(commit, host);
		// release_commit_memory frees the parent list without clearing
		// the pointer to it, so drop the dangling reference here.
		release_commit_memory(the_repository->parsed_objects, commit);
		commit->parents = NULL;
	}
	if (need_updated) {
		// Atom makes a feed level updated mandatory, and an empty
		// feed has no commit to take one from, so the epoch stands in.
		html("<updated>");
		xml_txt(feed_date(0));
		html("</updated>\n");
	}
	html("</feed>\n");
	free(host);
}