Change write() and read() function to send() and recv()

[DK-DARKmatter]

Replace the read() and write() function with send() and recv() in main.c, the flag is set to 0.

[DK-DARKmatter]

My bad. I just want to use the CI function, and show what kind of error will happen when we replace those function. I did not want to change current program, but I think if we want to improve the test coverage we should figure out the bugs first.

[npmccallum]

I asked Ke to improve our test coverage. Before he made larger changes, he was just going to swap the IO functions and ensure that everything still worked. It didn't. For some reason, clients fail when using send()/recv() instead of read()/write(). We need to find out why.

[puiterwijk]

@yingxiongraomingzk The problem with your patch is that your code now sends recv() and send() on stdio (buffers 0, 1, 2). That is invalid, as these functions are only allowed for sockets. (stdin, stdout and stderr are non-socket file descriptors.)

The following, brute-force, patch made the tests pass for me:

diff --git a/bin/main.c b/bin/main.c
index 166a4a3..78e2538 100644
--- a/bin/main.c
+++ b/bin/main.c
@@ -254,7 +254,10 @@ on_conn(options_t *opts, int con, int in, int out, const struct addrinfo *ai)
       if (!pfds[i].revents)
         continue;

-      ret = read(pfds[i].fd, buffer, sizeof(buffer));
+      if (pfds[i].fd <= 2)
+        ret = read(pfds[i].fd, buffer, sizeof(buffer));
+      else
+        ret = recv(pfds[i].fd, buffer, sizeof(buffer), 0);
       if (ret <= 0) {
         if (pfds[i].revents != POLLHUP &&
             (errno == EAGAIN || errno == EWOULDBLOCK))
diff --git a/bin/non.c b/bin/non.c
index c20db0c..34bbf48 100644
--- a/bin/non.c
+++ b/bin/non.c
@@ -98,7 +98,10 @@ non_write(int fd, void *buf, size_t len)
   uint8_t *b = buf;
   ssize_t ret;

-  ret = write(fd, buf, len);
+  if (fd <= 2)
+    ret = write(fd, buf, len);
+  else
+    ret = send(fd, buf, len, 0);
   if (ret < 0) {
     if (errno != EAGAIN)
       return ret;

[npmccallum]

This PR is superseded by #31 .