Prevent potential hang with scripts (take 3)

[cgull]

@juikim last tried to fix this in #1843. However, we're still having problems with hangs from pkg when background processes remain and hold PKG_MSGFD open.

When I looked at the code in pkg_run_scripts(), I found several problems:

• getline() is used to read input, this will hang if input is not terminated with LF.
• The code handles the ordering of testing process termination and input availability incorrectly. If a script starts a daemon that holds PKG_MSGFD open and exits, the code will fall into getline() and wait forever.
• The line/linecap variables allocated the input line buffer only once at the start of the loop; longer lines in input after that wouldn't have been completely read by getline().

I reworked the code to check waitpid(), then check for input, then read any input, then continue or terminate the loop. Now, it will wait and read any input until the child exits, then it will read any available input but not wait. Once the child has terminated and we know that from waitpid(), any leftover output from the child will be buffered in kernel and immediately available; once we have read that we can exit the loop. It also removes use of stdio and line blocking for the input reads.

One minor point: I looked at pkg_emit_msg() and the event code and I saw no obvious requirement that the message needs to be terminated by an LF. Is this correct?

I also added a test to verify this issue and the fix.

[bapt]

not there was a small style issue, but I'll fix after the merge, thank you very much!

[cgull]

Thanks for taking this! When do you expect to branch/release 1.16?

[bapt]

asap, I am working on more test cases for triggers