CURLE_WRITE_ERROR shadows CURLE_BAD_CONTENT_ENCODING

[PersDep]

What is going on

curl_multi_info_read returns msg->data.result == CURLE_WRITE_ERROR if received Content-Encoding is not supported.

I expected the following

I expect curl_multi_info_read to return CURLE_BAD_CONTENT_ENCODING.

Code

if(!conn->data->set.http_te_skip && !k->ignorebody) {
    if(!conn->data->set.http_ce_skip && k->writer_stack)
        result = Curl_unencode_write(conn, k->writer_stack, datap, piece);
    else
        result = Curl_client_write(conn, CLIENTWRITE_BODY, datap, piece);

    if(result)
        return CHUNKE_WRITE_ERROR;
}

http_chunks.c 194-202

Curl_unencode_write returns CURLE_BAD_CONTENT_ENCODING and then it is being shadowed by CHUNKE_WRITE_ERROR which later is turning into CURLE_WRITE_ERROR.

curl/libcurl version

curl 7.65.3-DEV (x86_64-pc-linux-gnu) libcurl/7.65.3-DEV OpenSSL/1.1.1 zlib/1.2.11

operating system

Ubuntu 18.04.3 LTS

[bagder]

Do you have an example of how to reproduce this?

[PersDep]

Do you have an example of how to reproduce this?

Here is a test based on tests/libtest/lib597.c

#include "test.h"

#include <limits.h>

#include "testutil.h"
#include "warnless.h"
#include "memdebug.h"

#define TEST_HANG_TIMEOUT 5 * 1000

int test(char *URL)
{
  CURL *easy = NULL;
  CURLM *multi = NULL;
  int res = 0;
  int running;
  int msgs_left;
  CURLMsg *msg;

  start_test_timing();

  global_init(CURL_GLOBAL_ALL);

  easy_init(easy);

  multi_init(multi);

  /* go verbose */
  easy_setopt(easy, CURLOPT_VERBOSE, 1L);

  /* specify target */
  easy_setopt(easy, CURLOPT_URL, URL);

  /* ask for unsupported encoding */
  easy_setopt(easy, CURLOPT_ACCEPT_ENCODING, "br");

  multi_add_handle(multi, easy);

  for(;;) {
    struct timeval interval;
    fd_set fdread;
    fd_set fdwrite;
    fd_set fdexcep;
    long timeout = -99;
    int maxfd = -99;

    multi_perform(multi, &running);

    abort_on_test_timeout();

    if(!running)
      break; /* done */

    FD_ZERO(&fdread);
    FD_ZERO(&fdwrite);
    FD_ZERO(&fdexcep);

    multi_fdset(multi, &fdread, &fdwrite, &fdexcep, &maxfd);

    /* At this point, maxfd is guaranteed to be greater or equal than -1. */

    multi_timeout(multi, &timeout);

    /* At this point, timeout is guaranteed to be greater or equal than
       -1. */

    if(timeout != -1L) {
      int itimeout = (timeout > (long)INT_MAX) ? INT_MAX : (int)timeout;
      interval.tv_sec = itimeout/1000;
      interval.tv_usec = (itimeout%1000)*1000;
    }
    else {
      interval.tv_sec = TEST_HANG_TIMEOUT/1000 + 1;
      interval.tv_usec = 0;
    }

    select_test(maxfd + 1, &fdread, &fdwrite, &fdexcep, &interval);

    abort_on_test_timeout();
  }

  msg = curl_multi_info_read(multi, &msgs_left);
  if(msg)
    res = msg->data.result;
  printf("RES: %d CURLE_WRITE_ERROR: %d\n", res, CURLE_WRITE_ERROR);

  multi_remove_handle(multi, easy);

test_cleanup:

  /* undocumented cleanup sequence - type UA */

  curl_multi_cleanup(multi);
  curl_easy_cleanup(easy);
  curl_global_cleanup();

  return res;
}

You can run it with URL == https://www.programcreek.com/python/example/103281/brotli.compress to see the described behavior. Also curl has to be configured with --without-brotli flag

[bagder]

Thank you, reproduced perfectly. Now I just need to figure out how to deal with this the best way...

[bagder]

Here's a stand alone version to repro: debug-4310.c