xbps-uhelper fails with redirect using proxy

[jschultz]

For example, fetching https://www.llvm.org/releases/7.0.1/llvm-7.0.1.src.tar.xz using a proxy (in my case squid running at http://172.17.0.1:3128 and a certificate that I have manually added to my system) produces:

looking up 172.17.0.1
connecting to 172.17.0.1:3128
SSL options: 4004
Peer verification enabled
Using CA cert path: /etc/ssl/certs
Verify hostname
TLSv1.2 connection established using ECDHE-RSA-AES256-GCM-SHA384
Certificate subject: /CN=llvm.org
Certificate issuer: /C=AU/ST= /L= /O= /OU= /CN=
requesting https://www.llvm.org/releases/7.0.1/llvm-7.0.1.src.tar.xz
302 redirect to http://releases.llvm.org/7.0.1/llvm-7.0.1.src.tar.xz
looking up 172.17.0.1
connecting to 172.17.0.1:3128
requesting http://releases.llvm.org/7.0.1/llvm-7.0.1.src.tar.xz
https://www.llvm.org/releases/7.0.1/llvm-7.0.1.src.tar.xz: Bad Request

The problem seems to be connected with the fact that an https URL redirects to a http URL. From the proxy log I can see:

1547696915.462 447 172.17.0.1 TCP_MISS/302 671 GET https://www.llvm.org/releases/7.0.1/llvm-7.0.1.src.tar.xz - HIER_DIRECT/54.67.122.174 text/html
1547696915.465 0 172.17.0.1 NONE/400 3937 GET /7.0.1/llvm-7.0.1.src.tar.xz - HIER_NONE/- text/html

I tracked the problem down to code around here in fetch/http.c:

  /* send request */
  if (verbose)
  	fetch_info("requesting %s://%s%s",
  	    url->scheme, host, url->doc);
  if (purl && strcasecmp(URL->scheme, SCHEME_HTTPS) != 0) {
  	http_cmd(conn, "%s %s://%s%s HTTP/1.1\r\n",
  	    op, url->scheme, host, url->doc);
  } else {
  	http_cmd(conn, "%s %s HTTP/1.1\r\n",
  	    op, url->doc);
  }

  if (if_modified_since && url->last_modified > 0)
  	set_if_modified_since(conn, url->last_modified);

  /* virtual host */
  http_cmd(conn, "Host: %s\r\n", host);

  if (strcasecmp(URL->scheme, SCHEME_HTTPS) != 0)
  	send_proxy_headers(conn, purl);

In two places this code uses URL->scheme (the protocol of the original URL) rather than url->scheme (the protocol of the redirected URL. I found that after replacing these two instances the code appears to work in this particular case. But not really understanding these protocols very well I hesitate to make a pull request based on a single case.

I see comments in the code that the whole http_request function needs to be rewritten anyway, so just bringing this issue to the developers' attention.