The travis red is a failed make checksrc:

./vtls/openssl.c:1430:75: warning: badly placed open brace (BRACEPOS)
                                                                           {
                                                                           ^

Fixed "make checksrc" problem.

There's a second place in the file where Curl_cert_hostcheck() is called. Doesn't that require the same procedure?

And regarding the code, I would prefer an approach that uses less #ifdefs in place. Perhaps you can just provide a wrapper function that does the conversions in this style:

#ifdef CURL_DOES_CONVERSIONS
static int hostcheck_convert(const char *match_pattern, const char *hostname)
{
  char *altptr2 = strdup(altptr);
  if(altptr2) {
    if(!Curl_convert_from_network(data, altptr2, altlen))
        rc = Curl_cert_hostcheck(altptr2, hostname);
    else
        rc = failcode;
    free(altptr2);
  }
  return rc;
}

#define Curl_cert_hostcheck(x,y) hostcheck_convert(x,y)
#endif

There's a second place in the file where Curl_cert_hostcheck() is called. Doesn't that require the same procedure?

You are right, I need to check this. If I understand it correctly the second place is triggered when a certificate does not have the subjectAltName extension. I'm not sure at the moment that a conversion is needed there as well, because our previous test environment had a server certificate without subjectAltName extension, and then it worked as expected, albeit with an older version of curl. The current problem on z/OS only surfaced when we created a new server certificate with subjectAltName extension.

And regarding the code, I would prefer an approach that uses less #ifdefs in place. Perhaps you can just provide a wrapper function that does the conversions in this style: ...

I will refactor the code accordingly.

There's a second place in the file where Curl_cert_hostcheck() is called. Doesn't that require the same procedure?

The second place within verifyhost() where Curl_cert_hostcheck()is called does not need to be modified, because there the host name in peer_CN is already converted to host encoding by Curl_convert_from_utf8():

I moved the check against subjectAltName into a separate function. The proposed macro trick to redefine the function Curl_cert_hostcheck() was not possible because the function needs additional parameters.

Thanks, the red travis failure is unrelated.

I took the liberty to modify your function slightly before I merged, please have a look and shout if you think I massacred something.