dan@watson.ibm.COM (Walt Daniels) (10/06/89)
Submitting for a friend (Walt) Is this an IBM problem, or is it a genetic defect? ------------ I'm installing an XIOErrorHandler so that my application can do some reasonable clean-up and shutdown if the connection to the window fails. (I'm running AIX/RT - latest updates to X) My XIOErrorHandler gets control just fine when I run the client on one node and the server on another and I kill the server. But if I run the client and server on the same machine and I kill the server (or cancel the window), the default AIX SIGPIPE signal handler gets control, prints "Broken pipe" and then exits. My handler never gets called. The reason why this happens is that Xlib is writing to a pipe, and if the pipe gets broken, the AIX SIGPIPE handler gets control and never gives Xlib the chance to check the error code from the write and invoke the XIOErrorHandler. I really think that Xlib should do something with SIGPIPE signals other than letting the default action take place. This would allow the XIOErrorHandler (Xlib's default, or user specified) to get invoked. I can get around the problem by saying "signal(SIGPIPE, SIG_IGN);" in my application, but I think this call should be in Xlib. I'm not totally sure why it worked for the remote case. But I would guess that TCP/IP (or whatever the transport mechanism is for remote pipes) is handling SIGPIPE. Karl Schultz Engineering Graphics Products
cat@island.uu.net (Cathleen Greenberg) (10/14/89)
In article <100689.093908.dan@watson.ibm.com> dan@watson.ibm.COM (Walt Daniels) writes: >Submitting for a friend (Walt) >Is this an IBM problem, or is it a genetic defect? >------------ > I really think that Xlib should do something with >SIGPIPE signals other than letting the default action take place. >This would allow the XIOErrorHandler (Xlib's default, or user >specified) to get invoked. To answer the question -- no it is not just IBM -- I have the same problem on a sun 3/60 -- and I agree that Xlib should do something other than hand over control so that the only message is broken pipe. On a similar note -- does anyone else have problems with large Ximages?? When reading (and sometimes writing) TIFF files, I will either get an XIO error (broken pipe) or sometimes the program quits with no message at all. The program handles images better on an NCD X terminal than on a sun 3/60 -- anyone have any ideas?? Anyone else experiencing this phenomenon?? --Cathleen Greenberg ...sun!island!cat -- -------------------------------------------------------------------------------- Cathleen Greenberg work: (415) 491-1000 sun!island!cat ucbcad!island!cat
rws@EXPO.LCS.MIT.EDU (Bob Scheifler) (10/16/89)
Is this an IBM problem, or is it a genetic defect?
It is a genetic defect in Unix, not X.
I can get around the problem by saying
"signal(SIGPIPE, SIG_IGN);" in my application, but I think this call
should be in Xlib.
Sorry, I have to disagree. Signals are a scarce resource, and SIGPIPE
might be used/handled by the application for other reasons. Having Xlib
step on the signal handler would be a significant incompatible change.