3.6 fails to build on the Hurd

[mochrul]

Because I have no power to reopen bug #355, but I would like to see it be fixed, I created a new ticket for it:

3.6 as of commit 4d90138 fails the test suite on Hurd, with the following message:

FAIL: lib/logproto/tests/test_logproto
======================================

[2014-12-15T14:43:48.462716] Plugin module not found in 'module-path';

  ###########################################################################
  #
  # FAIL: ASSERTION FAILED: LogProtoServer expected status mismatch; actual=2, expected=1
  #
  # Test case: test-text-server.c/test_log_proto_text_server()
  #
  #     test_log_proto_text_server_eol_before_eof()
  #
  ###########################################################################

Full build log available here.

I will be retrying with 3.6.2 (or latest 3.6/master, if different) in a couple of days, to see if the problem has been accidentally fixed since the last snapshot I took.

[mochrul]

With some detour and dead-end, finally, I was able to find what caused this issue.

The problem is caused by this commit: 81e2c83

More precisely, this lines:

  if (GPOINTER_TO_UINT(current_iov->iov_base) < 4096)
    {
      /* error injection */
      errno = GPOINTER_TO_UINT(current_iov->iov_base);
      return -1;
    }

In most of the architectures, this works because the error codes are low. but not in Hurd:

/usr/include/i386-gnu/bits/errno.h:  EIO                            = 0x40000005,       /* Input/output error */

Hopefully, this is the only place where this failed expectation is tried to be exploited.

[gaborznagy]

Fixed with #1914.