i18n: git-add "Unstaged changes" message
[gitweb.git] / remote-curl.c
index ebdab3603ec767b42ce04e0de2c3deeb516f2525..04d4813e4183c675b54aba942cd078d8ff632053 100644 (file)
 #include "walker.h"
 #include "http.h"
 #include "exec_cmd.h"
+#include "run-command.h"
+#include "pkt-line.h"
+#include "sideband.h"
 
-static struct ref *get_refs(struct walker *walker, const char *url)
+static struct remote *remote;
+static const char *url; /* always ends with a trailing slash */
+
+struct options {
+       int verbosity;
+       unsigned long depth;
+       unsigned progress : 1,
+               followtags : 1,
+               dry_run : 1,
+               thin : 1;
+};
+static struct options options;
+
+static int set_option(const char *name, const char *value)
+{
+       if (!strcmp(name, "verbosity")) {
+               char *end;
+               int v = strtol(value, &end, 10);
+               if (value == end || *end)
+                       return -1;
+               options.verbosity = v;
+               return 0;
+       }
+       else if (!strcmp(name, "progress")) {
+               if (!strcmp(value, "true"))
+                       options.progress = 1;
+               else if (!strcmp(value, "false"))
+                       options.progress = 0;
+               else
+                       return -1;
+               return 0;
+       }
+       else if (!strcmp(name, "depth")) {
+               char *end;
+               unsigned long v = strtoul(value, &end, 10);
+               if (value == end || *end)
+                       return -1;
+               options.depth = v;
+               return 0;
+       }
+       else if (!strcmp(name, "followtags")) {
+               if (!strcmp(value, "true"))
+                       options.followtags = 1;
+               else if (!strcmp(value, "false"))
+                       options.followtags = 0;
+               else
+                       return -1;
+               return 0;
+       }
+       else if (!strcmp(name, "dry-run")) {
+               if (!strcmp(value, "true"))
+                       options.dry_run = 1;
+               else if (!strcmp(value, "false"))
+                       options.dry_run = 0;
+               else
+                       return -1;
+               return 0;
+       }
+       else {
+               return 1 /* unsupported */;
+       }
+}
+
+struct discovery {
+       const char *service;
+       char *buf_alloc;
+       char *buf;
+       size_t len;
+       unsigned proto_git : 1;
+};
+static struct discovery *last_discovery;
+
+static void free_discovery(struct discovery *d)
+{
+       if (d) {
+               if (d == last_discovery)
+                       last_discovery = NULL;
+               free(d->buf_alloc);
+               free(d);
+       }
+}
+
+static struct discovery* discover_refs(const char *service)
 {
        struct strbuf buffer = STRBUF_INIT;
-       char *data, *start, *mid;
-       char *ref_name;
+       struct discovery *last = last_discovery;
        char *refs_url;
-       int i = 0;
-       int http_ret;
+       int http_ret, is_http = 0, proto_git_candidate = 1;
 
-       struct ref *refs = NULL;
-       struct ref *ref = NULL;
-       struct ref *last_ref = NULL;
+       if (last && !strcmp(service, last->service))
+               return last;
+       free_discovery(last);
 
-       refs_url = xmalloc(strlen(url) + 11);
-       sprintf(refs_url, "%s/info/refs", url);
+       strbuf_addf(&buffer, "%sinfo/refs", url);
+       if (!prefixcmp(url, "http://") || !prefixcmp(url, "https://")) {
+               is_http = 1;
+               if (!strchr(url, '?'))
+                       strbuf_addch(&buffer, '?');
+               else
+                       strbuf_addch(&buffer, '&');
+               strbuf_addf(&buffer, "service=%s", service);
+       }
+       refs_url = strbuf_detach(&buffer, NULL);
 
        http_ret = http_get_strbuf(refs_url, &buffer, HTTP_NO_CACHE);
+
+       /* try again with "plain" url (no ? or & appended) */
+       if (http_ret != HTTP_OK) {
+               free(refs_url);
+               strbuf_reset(&buffer);
+
+               proto_git_candidate = 0;
+               strbuf_addf(&buffer, "%sinfo/refs", url);
+               refs_url = strbuf_detach(&buffer, NULL);
+
+               http_ret = http_get_strbuf(refs_url, &buffer, HTTP_NO_CACHE);
+       }
+
        switch (http_ret) {
        case HTTP_OK:
                break;
        case HTTP_MISSING_TARGET:
                die("%s not found: did you run git update-server-info on the"
                    " server?", refs_url);
+       case HTTP_NOAUTH:
+               die("Authentication failed");
        default:
                http_error(refs_url, http_ret);
                die("HTTP request failed");
        }
 
-       data = buffer.buf;
+       last= xcalloc(1, sizeof(*last_discovery));
+       last->service = service;
+       last->buf_alloc = strbuf_detach(&buffer, &last->len);
+       last->buf = last->buf_alloc;
+
+       if (is_http && proto_git_candidate
+               && 5 <= last->len && last->buf[4] == '#') {
+               /* smart HTTP response; validate that the service
+                * pkt-line matches our request.
+                */
+               struct strbuf exp = STRBUF_INIT;
+
+               if (packet_get_line(&buffer, &last->buf, &last->len) <= 0)
+                       die("%s has invalid packet header", refs_url);
+               if (buffer.len && buffer.buf[buffer.len - 1] == '\n')
+                       strbuf_setlen(&buffer, buffer.len - 1);
+
+               strbuf_addf(&exp, "# service=%s", service);
+               if (strbuf_cmp(&exp, &buffer))
+                       die("invalid server response; got '%s'", buffer.buf);
+               strbuf_release(&exp);
+
+               /* The header can include additional metadata lines, up
+                * until a packet flush marker.  Ignore these now, but
+                * in the future we might start to scan them.
+                */
+               strbuf_reset(&buffer);
+               while (packet_get_line(&buffer, &last->buf, &last->len) > 0)
+                       strbuf_reset(&buffer);
+
+               last->proto_git = 1;
+       }
+
+       free(refs_url);
+       strbuf_release(&buffer);
+       last_discovery = last;
+       return last;
+}
+
+static int write_discovery(int in, int out, void *data)
+{
+       struct discovery *heads = data;
+       int err = 0;
+       if (write_in_full(out, heads->buf, heads->len) != heads->len)
+               err = 1;
+       close(out);
+       return err;
+}
+
+static struct ref *parse_git_refs(struct discovery *heads)
+{
+       struct ref *list = NULL;
+       struct async async;
+
+       memset(&async, 0, sizeof(async));
+       async.proc = write_discovery;
+       async.data = heads;
+       async.out = -1;
+
+       if (start_async(&async))
+               die("cannot start thread to parse advertised refs");
+       get_remote_heads(async.out, &list, 0, NULL, 0, NULL);
+       close(async.out);
+       if (finish_async(&async))
+               die("ref parsing thread failed");
+       return list;
+}
+
+static struct ref *parse_info_refs(struct discovery *heads)
+{
+       char *data, *start, *mid;
+       char *ref_name;
+       int i = 0;
+
+       struct ref *refs = NULL;
+       struct ref *ref = NULL;
+       struct ref *last_ref = NULL;
+
+       data = heads->buf;
        start = NULL;
        mid = data;
-       while (i < buffer.len) {
+       while (i < heads->len) {
                if (!start) {
                        start = &data[i];
                }
@@ -60,10 +244,8 @@ static struct ref *get_refs(struct walker *walker, const char *url)
                i++;
        }
 
-       strbuf_release(&buffer);
-
        ref = alloc_ref("HEAD");
-       if (!walker->fetch_ref(walker, ref) &&
+       if (!http_fetch_ref(url, ref) &&
            !resolve_remote_symref(ref, refs)) {
                ref->next = refs;
                refs = ref;
@@ -71,17 +253,538 @@ static struct ref *get_refs(struct walker *walker, const char *url)
                free(ref);
        }
 
-       strbuf_release(&buffer);
-       free(refs_url);
        return refs;
 }
 
+static struct ref *get_refs(int for_push)
+{
+       struct discovery *heads;
+
+       if (for_push)
+               heads = discover_refs("git-receive-pack");
+       else
+               heads = discover_refs("git-upload-pack");
+
+       if (heads->proto_git)
+               return parse_git_refs(heads);
+       return parse_info_refs(heads);
+}
+
+static void output_refs(struct ref *refs)
+{
+       struct ref *posn;
+       for (posn = refs; posn; posn = posn->next) {
+               if (posn->symref)
+                       printf("@%s %s\n", posn->symref, posn->name);
+               else
+                       printf("%s %s\n", sha1_to_hex(posn->old_sha1), posn->name);
+       }
+       printf("\n");
+       fflush(stdout);
+       free_refs(refs);
+}
+
+struct rpc_state {
+       const char *service_name;
+       const char **argv;
+       char *service_url;
+       char *hdr_content_type;
+       char *hdr_accept;
+       char *buf;
+       size_t alloc;
+       size_t len;
+       size_t pos;
+       int in;
+       int out;
+       struct strbuf result;
+       unsigned gzip_request : 1;
+       unsigned initial_buffer : 1;
+};
+
+static size_t rpc_out(void *ptr, size_t eltsize,
+               size_t nmemb, void *buffer_)
+{
+       size_t max = eltsize * nmemb;
+       struct rpc_state *rpc = buffer_;
+       size_t avail = rpc->len - rpc->pos;
+
+       if (!avail) {
+               rpc->initial_buffer = 0;
+               avail = packet_read_line(rpc->out, rpc->buf, rpc->alloc);
+               if (!avail)
+                       return 0;
+               rpc->pos = 0;
+               rpc->len = avail;
+       }
+
+       if (max < avail)
+               avail = max;
+       memcpy(ptr, rpc->buf + rpc->pos, avail);
+       rpc->pos += avail;
+       return avail;
+}
+
+#ifndef NO_CURL_IOCTL
+static curlioerr rpc_ioctl(CURL *handle, int cmd, void *clientp)
+{
+       struct rpc_state *rpc = clientp;
+
+       switch (cmd) {
+       case CURLIOCMD_NOP:
+               return CURLIOE_OK;
+
+       case CURLIOCMD_RESTARTREAD:
+               if (rpc->initial_buffer) {
+                       rpc->pos = 0;
+                       return CURLIOE_OK;
+               }
+               fprintf(stderr, "Unable to rewind rpc post data - try increasing http.postBuffer\n");
+               return CURLIOE_FAILRESTART;
+
+       default:
+               return CURLIOE_UNKNOWNCMD;
+       }
+}
+#endif
+
+static size_t rpc_in(const void *ptr, size_t eltsize,
+               size_t nmemb, void *buffer_)
+{
+       size_t size = eltsize * nmemb;
+       struct rpc_state *rpc = buffer_;
+       write_or_die(rpc->in, ptr, size);
+       return size;
+}
+
+static int post_rpc(struct rpc_state *rpc)
+{
+       struct active_request_slot *slot;
+       struct slot_results results;
+       struct curl_slist *headers = NULL;
+       int use_gzip = rpc->gzip_request;
+       char *gzip_body = NULL;
+       int err = 0, large_request = 0;
+
+       /* Try to load the entire request, if we can fit it into the
+        * allocated buffer space we can use HTTP/1.0 and avoid the
+        * chunked encoding mess.
+        */
+       while (1) {
+               size_t left = rpc->alloc - rpc->len;
+               char *buf = rpc->buf + rpc->len;
+               int n;
+
+               if (left < LARGE_PACKET_MAX) {
+                       large_request = 1;
+                       use_gzip = 0;
+                       break;
+               }
+
+               n = packet_read_line(rpc->out, buf, left);
+               if (!n)
+                       break;
+               rpc->len += n;
+       }
+
+       slot = get_active_slot();
+       slot->results = &results;
+
+       curl_easy_setopt(slot->curl, CURLOPT_NOBODY, 0);
+       curl_easy_setopt(slot->curl, CURLOPT_POST, 1);
+       curl_easy_setopt(slot->curl, CURLOPT_URL, rpc->service_url);
+       curl_easy_setopt(slot->curl, CURLOPT_ENCODING, "");
+
+       headers = curl_slist_append(headers, rpc->hdr_content_type);
+       headers = curl_slist_append(headers, rpc->hdr_accept);
+
+       if (large_request) {
+               /* The request body is large and the size cannot be predicted.
+                * We must use chunked encoding to send it.
+                */
+               headers = curl_slist_append(headers, "Expect: 100-continue");
+               headers = curl_slist_append(headers, "Transfer-Encoding: chunked");
+               rpc->initial_buffer = 1;
+               curl_easy_setopt(slot->curl, CURLOPT_READFUNCTION, rpc_out);
+               curl_easy_setopt(slot->curl, CURLOPT_INFILE, rpc);
+#ifndef NO_CURL_IOCTL
+               curl_easy_setopt(slot->curl, CURLOPT_IOCTLFUNCTION, rpc_ioctl);
+               curl_easy_setopt(slot->curl, CURLOPT_IOCTLDATA, rpc);
+#endif
+               if (options.verbosity > 1) {
+                       fprintf(stderr, "POST %s (chunked)\n", rpc->service_name);
+                       fflush(stderr);
+               }
+
+       } else if (use_gzip && 1024 < rpc->len) {
+               /* The client backend isn't giving us compressed data so
+                * we can try to deflate it ourselves, this may save on.
+                * the transfer time.
+                */
+               size_t size;
+               z_stream stream;
+               int ret;
+
+               memset(&stream, 0, sizeof(stream));
+               ret = deflateInit2(&stream, Z_BEST_COMPRESSION,
+                               Z_DEFLATED, (15 + 16),
+                               8, Z_DEFAULT_STRATEGY);
+               if (ret != Z_OK)
+                       die("cannot deflate request; zlib init error %d", ret);
+               size = deflateBound(&stream, rpc->len);
+               gzip_body = xmalloc(size);
+
+               stream.next_in = (unsigned char *)rpc->buf;
+               stream.avail_in = rpc->len;
+               stream.next_out = (unsigned char *)gzip_body;
+               stream.avail_out = size;
+
+               ret = deflate(&stream, Z_FINISH);
+               if (ret != Z_STREAM_END)
+                       die("cannot deflate request; zlib deflate error %d", ret);
+
+               ret = deflateEnd(&stream);
+               if (ret != Z_OK)
+                       die("cannot deflate request; zlib end error %d", ret);
+
+               size = stream.total_out;
+
+               headers = curl_slist_append(headers, "Content-Encoding: gzip");
+               curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDS, gzip_body);
+               curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDSIZE, size);
+
+               if (options.verbosity > 1) {
+                       fprintf(stderr, "POST %s (gzip %lu to %lu bytes)\n",
+                               rpc->service_name,
+                               (unsigned long)rpc->len, (unsigned long)size);
+                       fflush(stderr);
+               }
+       } else {
+               /* We know the complete request size in advance, use the
+                * more normal Content-Length approach.
+                */
+               curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDS, rpc->buf);
+               curl_easy_setopt(slot->curl, CURLOPT_POSTFIELDSIZE, rpc->len);
+               if (options.verbosity > 1) {
+                       fprintf(stderr, "POST %s (%lu bytes)\n",
+                               rpc->service_name, (unsigned long)rpc->len);
+                       fflush(stderr);
+               }
+       }
+
+       curl_easy_setopt(slot->curl, CURLOPT_HTTPHEADER, headers);
+       curl_easy_setopt(slot->curl, CURLOPT_WRITEFUNCTION, rpc_in);
+       curl_easy_setopt(slot->curl, CURLOPT_FILE, rpc);
+
+       slot->curl_result = curl_easy_perform(slot->curl);
+       finish_active_slot(slot);
+
+       if (results.curl_result != CURLE_OK) {
+               err |= error("RPC failed; result=%d, HTTP code = %ld",
+                       results.curl_result, results.http_code);
+       }
+
+       curl_slist_free_all(headers);
+       free(gzip_body);
+       return err;
+}
+
+static int rpc_service(struct rpc_state *rpc, struct discovery *heads)
+{
+       const char *svc = rpc->service_name;
+       struct strbuf buf = STRBUF_INIT;
+       struct child_process client;
+       int err = 0;
+
+       memset(&client, 0, sizeof(client));
+       client.in = -1;
+       client.out = -1;
+       client.git_cmd = 1;
+       client.argv = rpc->argv;
+       if (start_command(&client))
+               exit(1);
+       if (heads)
+               write_or_die(client.in, heads->buf, heads->len);
+
+       rpc->alloc = http_post_buffer;
+       rpc->buf = xmalloc(rpc->alloc);
+       rpc->in = client.in;
+       rpc->out = client.out;
+       strbuf_init(&rpc->result, 0);
+
+       strbuf_addf(&buf, "%s%s", url, svc);
+       rpc->service_url = strbuf_detach(&buf, NULL);
+
+       strbuf_addf(&buf, "Content-Type: application/x-%s-request", svc);
+       rpc->hdr_content_type = strbuf_detach(&buf, NULL);
+
+       strbuf_addf(&buf, "Accept: application/x-%s-result", svc);
+       rpc->hdr_accept = strbuf_detach(&buf, NULL);
+
+       while (!err) {
+               int n = packet_read_line(rpc->out, rpc->buf, rpc->alloc);
+               if (!n)
+                       break;
+               rpc->pos = 0;
+               rpc->len = n;
+               err |= post_rpc(rpc);
+       }
+
+       close(client.in);
+       client.in = -1;
+       strbuf_read(&rpc->result, client.out, 0);
+
+       close(client.out);
+       client.out = -1;
+
+       err |= finish_command(&client);
+       free(rpc->service_url);
+       free(rpc->hdr_content_type);
+       free(rpc->hdr_accept);
+       free(rpc->buf);
+       strbuf_release(&buf);
+       return err;
+}
+
+static int fetch_dumb(int nr_heads, struct ref **to_fetch)
+{
+       struct walker *walker;
+       char **targets = xmalloc(nr_heads * sizeof(char*));
+       int ret, i;
+
+       if (options.depth)
+               die("dumb http transport does not support --depth");
+       for (i = 0; i < nr_heads; i++)
+               targets[i] = xstrdup(sha1_to_hex(to_fetch[i]->old_sha1));
+
+       walker = get_http_walker(url);
+       walker->get_all = 1;
+       walker->get_tree = 1;
+       walker->get_history = 1;
+       walker->get_verbosely = options.verbosity >= 3;
+       walker->get_recover = 0;
+       ret = walker_fetch(walker, nr_heads, targets, NULL, NULL);
+       walker_free(walker);
+
+       for (i = 0; i < nr_heads; i++)
+               free(targets[i]);
+       free(targets);
+
+       return ret ? error("Fetch failed.") : 0;
+}
+
+static int fetch_git(struct discovery *heads,
+       int nr_heads, struct ref **to_fetch)
+{
+       struct rpc_state rpc;
+       char *depth_arg = NULL;
+       const char **argv;
+       int argc = 0, i, err;
+
+       argv = xmalloc((15 + nr_heads) * sizeof(char*));
+       argv[argc++] = "fetch-pack";
+       argv[argc++] = "--stateless-rpc";
+       argv[argc++] = "--lock-pack";
+       if (options.followtags)
+               argv[argc++] = "--include-tag";
+       if (options.thin)
+               argv[argc++] = "--thin";
+       if (options.verbosity >= 3) {
+               argv[argc++] = "-v";
+               argv[argc++] = "-v";
+       }
+       if (!options.progress)
+               argv[argc++] = "--no-progress";
+       if (options.depth) {
+               struct strbuf buf = STRBUF_INIT;
+               strbuf_addf(&buf, "--depth=%lu", options.depth);
+               depth_arg = strbuf_detach(&buf, NULL);
+               argv[argc++] = depth_arg;
+       }
+       argv[argc++] = url;
+       for (i = 0; i < nr_heads; i++) {
+               struct ref *ref = to_fetch[i];
+               if (!ref->name || !*ref->name)
+                       die("cannot fetch by sha1 over smart http");
+               argv[argc++] = ref->name;
+       }
+       argv[argc++] = NULL;
+
+       memset(&rpc, 0, sizeof(rpc));
+       rpc.service_name = "git-upload-pack",
+       rpc.argv = argv;
+       rpc.gzip_request = 1;
+
+       err = rpc_service(&rpc, heads);
+       if (rpc.result.len)
+               safe_write(1, rpc.result.buf, rpc.result.len);
+       strbuf_release(&rpc.result);
+       free(argv);
+       free(depth_arg);
+       return err;
+}
+
+static int fetch(int nr_heads, struct ref **to_fetch)
+{
+       struct discovery *d = discover_refs("git-upload-pack");
+       if (d->proto_git)
+               return fetch_git(d, nr_heads, to_fetch);
+       else
+               return fetch_dumb(nr_heads, to_fetch);
+}
+
+static void parse_fetch(struct strbuf *buf)
+{
+       struct ref **to_fetch = NULL;
+       struct ref *list_head = NULL;
+       struct ref **list = &list_head;
+       int alloc_heads = 0, nr_heads = 0;
+
+       do {
+               if (!prefixcmp(buf->buf, "fetch ")) {
+                       char *p = buf->buf + strlen("fetch ");
+                       char *name;
+                       struct ref *ref;
+                       unsigned char old_sha1[20];
+
+                       if (strlen(p) < 40 || get_sha1_hex(p, old_sha1))
+                               die("protocol error: expected sha/ref, got %s'", p);
+                       if (p[40] == ' ')
+                               name = p + 41;
+                       else if (!p[40])
+                               name = "";
+                       else
+                               die("protocol error: expected sha/ref, got %s'", p);
+
+                       ref = alloc_ref(name);
+                       hashcpy(ref->old_sha1, old_sha1);
+
+                       *list = ref;
+                       list = &ref->next;
+
+                       ALLOC_GROW(to_fetch, nr_heads + 1, alloc_heads);
+                       to_fetch[nr_heads++] = ref;
+               }
+               else
+                       die("http transport does not support %s", buf->buf);
+
+               strbuf_reset(buf);
+               if (strbuf_getline(buf, stdin, '\n') == EOF)
+                       return;
+               if (!*buf->buf)
+                       break;
+       } while (1);
+
+       if (fetch(nr_heads, to_fetch))
+               exit(128); /* error already reported */
+       free_refs(list_head);
+       free(to_fetch);
+
+       printf("\n");
+       fflush(stdout);
+       strbuf_reset(buf);
+}
+
+static int push_dav(int nr_spec, char **specs)
+{
+       const char **argv = xmalloc((10 + nr_spec) * sizeof(char*));
+       int argc = 0, i;
+
+       argv[argc++] = "http-push";
+       argv[argc++] = "--helper-status";
+       if (options.dry_run)
+               argv[argc++] = "--dry-run";
+       if (options.verbosity > 1)
+               argv[argc++] = "--verbose";
+       argv[argc++] = url;
+       for (i = 0; i < nr_spec; i++)
+               argv[argc++] = specs[i];
+       argv[argc++] = NULL;
+
+       if (run_command_v_opt(argv, RUN_GIT_CMD))
+               die("git-%s failed", argv[0]);
+       free(argv);
+       return 0;
+}
+
+static int push_git(struct discovery *heads, int nr_spec, char **specs)
+{
+       struct rpc_state rpc;
+       const char **argv;
+       int argc = 0, i, err;
+
+       argv = xmalloc((10 + nr_spec) * sizeof(char*));
+       argv[argc++] = "send-pack";
+       argv[argc++] = "--stateless-rpc";
+       argv[argc++] = "--helper-status";
+       if (options.thin)
+               argv[argc++] = "--thin";
+       if (options.dry_run)
+               argv[argc++] = "--dry-run";
+       if (options.verbosity > 1)
+               argv[argc++] = "--verbose";
+       argv[argc++] = url;
+       for (i = 0; i < nr_spec; i++)
+               argv[argc++] = specs[i];
+       argv[argc++] = NULL;
+
+       memset(&rpc, 0, sizeof(rpc));
+       rpc.service_name = "git-receive-pack",
+       rpc.argv = argv;
+
+       err = rpc_service(&rpc, heads);
+       if (rpc.result.len)
+               safe_write(1, rpc.result.buf, rpc.result.len);
+       strbuf_release(&rpc.result);
+       free(argv);
+       return err;
+}
+
+static int push(int nr_spec, char **specs)
+{
+       struct discovery *heads = discover_refs("git-receive-pack");
+       int ret;
+
+       if (heads->proto_git)
+               ret = push_git(heads, nr_spec, specs);
+       else
+               ret = push_dav(nr_spec, specs);
+       free_discovery(heads);
+       return ret;
+}
+
+static void parse_push(struct strbuf *buf)
+{
+       char **specs = NULL;
+       int alloc_spec = 0, nr_spec = 0, i;
+
+       do {
+               if (!prefixcmp(buf->buf, "push ")) {
+                       ALLOC_GROW(specs, nr_spec + 1, alloc_spec);
+                       specs[nr_spec++] = xstrdup(buf->buf + 5);
+               }
+               else
+                       die("http transport does not support %s", buf->buf);
+
+               strbuf_reset(buf);
+               if (strbuf_getline(buf, stdin, '\n') == EOF)
+                       return;
+               if (!*buf->buf)
+                       break;
+       } while (1);
+
+       if (push(nr_spec, specs))
+               exit(128); /* error already reported */
+       for (i = 0; i < nr_spec; i++)
+               free(specs[i]);
+       free(specs);
+
+       printf("\n");
+       fflush(stdout);
+}
+
 int main(int argc, const char **argv)
 {
-       struct remote *remote;
        struct strbuf buf = STRBUF_INIT;
-       const char *url;
-       struct walker *walker = NULL;
        int nongit;
 
        git_extract_argv0_path(argv[0]);
@@ -91,48 +794,60 @@ int main(int argc, const char **argv)
                return 1;
        }
 
+       options.verbosity = 1;
+       options.progress = !!isatty(2);
+       options.thin = 1;
+
        remote = remote_get(argv[1]);
 
        if (argc > 2) {
-               url = argv[2];
+               end_url_with_slash(&buf, argv[2]);
        } else {
-               url = remote->url[0];
+               end_url_with_slash(&buf, remote->url[0]);
        }
 
+       url = strbuf_detach(&buf, NULL);
+
+       http_init(remote);
+
        do {
                if (strbuf_getline(&buf, stdin, '\n') == EOF)
                        break;
                if (!prefixcmp(buf.buf, "fetch ")) {
-                       char *obj = buf.buf + strlen("fetch ");
                        if (nongit)
                                die("Fetch attempted without a local repo");
-                       if (!walker)
-                               walker = get_http_walker(url, remote);
-                       walker->get_all = 1;
-                       walker->get_tree = 1;
-                       walker->get_history = 1;
-                       walker->get_verbosely = 0;
-                       walker->get_recover = 0;
-                       if (walker_fetch(walker, 1, &obj, NULL, NULL))
-                               die("Fetch failed.");
-                       printf("\n");
-                       fflush(stdout);
-               } else if (!strcmp(buf.buf, "list")) {
-                       struct ref *refs;
-                       struct ref *posn;
-                       if (!walker)
-                               walker = get_http_walker(url, remote);
-                       refs = get_refs(walker, url);
-                       for (posn = refs; posn; posn = posn->next) {
-                               if (posn->symref)
-                                       printf("@%s %s\n", posn->symref, posn->name);
-                               else
-                                       printf("%s %s\n", sha1_to_hex(posn->old_sha1), posn->name);
-                       }
-                       printf("\n");
+                       parse_fetch(&buf);
+
+               } else if (!strcmp(buf.buf, "list") || !prefixcmp(buf.buf, "list ")) {
+                       int for_push = !!strstr(buf.buf + 4, "for-push");
+                       output_refs(get_refs(for_push));
+
+               } else if (!prefixcmp(buf.buf, "push ")) {
+                       parse_push(&buf);
+
+               } else if (!prefixcmp(buf.buf, "option ")) {
+                       char *name = buf.buf + strlen("option ");
+                       char *value = strchr(name, ' ');
+                       int result;
+
+                       if (value)
+                               *value++ = '\0';
+                       else
+                               value = "true";
+
+                       result = set_option(name, value);
+                       if (!result)
+                               printf("ok\n");
+                       else if (result < 0)
+                               printf("error invalid value\n");
+                       else
+                               printf("unsupported\n");
                        fflush(stdout);
+
                } else if (!strcmp(buf.buf, "capabilities")) {
                        printf("fetch\n");
+                       printf("option\n");
+                       printf("push\n");
                        printf("\n");
                        fflush(stdout);
                } else {
@@ -140,5 +855,8 @@ int main(int argc, const char **argv)
                }
                strbuf_reset(&buf);
        } while (1);
+
+       http_cleanup();
+
        return 0;
 }