A week or two ago, a team mate had an issue with a nested while loop in a BASH script. KSH ran just fine, but when he ran the exact same script under BASH, it had “unexpected results.” The while loop involved piping output to a while loop (to feed the loop), and there were two loops to iterate, both fed in this fashion.
Some of you may have already guessed what the issue was, but I wanted to go into a little detail here, because it is important to understand why two very similar shells behave so very differently sometimes. In this case, it has to do with how BASH deals with an execution chain.
When you go full on Mario (chaining commands together with lots of pipes,) BASH has a unique feature (as compared to KSH) where it stores the exit code for EACH PIECE of the pipeline in an array. This behavior means it is hijacking the pipeline chain, and when you have a loop that relies on receiving output from a pipe, that causes oddities to ensue. What I mean by “hijacking” is that each piece of the pipe is executed in its own subprocess fork. This is how BASH is able to grab each individual exit status to store in the array. Since each piece of the pipe is a “fork” call, the contents of the variable being manipulated isn’t what we expect inside the second loop, and we get output that doesn’t seem to be what we expect, unless we understand that it is forking. KSH doesn’t do this, so there’s nothing trying to inspect the chain, and thus the loops run just fine. After I realized what was going on, I suggested changing the command to not use a pipe to feed the loop, and a workable solution was found that works on both BASH and KSH seamlessly. I don’t recall exactly, but I think I had them change it from an “echo ‘something’ | while … do” to a “while … do < $( echo ‘something’ )” to fix it.
The BASH special array is called PIPESTATUS, and is useful for troubleshooting certain steps in a complicated pipeline, but can cause issues if you don’t know how the behavior affects the pipes in play. In the case of the nested loop, the loss of the next setting of the variable in play was problematic.