Incorrect communication to client proc

[rhc54]

Strictly in the v2.x series. The daemon immediately attempts to send a message to its lowest-ranked child. The child process is required to initiate communications and will connect back to the daemon when ready. The unexpected message results in a SIGPIPE as the child hasn't yet setup a connection and causes the pipe to fail when two simultaneous connections are in progress.

If we are not running under a debugger, then the daemon traps the SIGPIPE and ignores it. However, when under a debugger, the debugger traps the SIGPIPE and reacts to it, causing the entire procedure to potentially collapse. Unfortunately, this specifically impacts a particular debugger vendor.

[rhc54]

@dirk-schubert-arm (Dirk Schubert, DDT developer) I have shifted this problem from #3660

[ghost]

Sorry for the late reply, I was on holiday. @xavier1arm has applied your fix in #4630 and can confirm it works. Thanks @rhc54. Is there anything I can help with to move #4630 forward. I feel not familiar enough with the code base to do the review, but on the other hand it's just a small and local change.

[rhc54]

Thanks for the update. I think we are stuck until @jsquyres returns from vacation, but the release isn't going anywhere until then anyway.

[hppritcha]

merged fix #4630 into v2.x. closing this issue.

[jsquyres]

@rhc54 Is this a request that I test this (on v2.1.x) under DDT on my cluster?

If so, I can confirm that with the head of each of the following branches as of 3 Jan 2018:

1. 2.0.x: DDT attachment works
2. 2.1.x: DDT attachment works
3. 3.0.x: DDT attachment works
4. 3.1.x: DDT attachment works

[jsquyres]

I neglected to mention -- I tested with DDT Allinea Forge v7.0.2

[rhc54]

Sorry - wasn't necessary as Allinea as already confirmed it works 😢