Re: "Forked child base mismatch" using MKS Toolkit shell

>>>> "Andy" == Andy Piper <andyp(a)beasys.com&gt; writes:

> A few weeks ago I sent a message to the list asking if anyone knew what was > causing the MKS Toolkit Korn shell to fail when used with the XEmacs shell > command. If you recall any attempt to invoke a sub shell produced the > message "Forked child base mismatch" and pipes didn't work (they appeared > to cause the shell to hang). I think I've fixed both problems and the > shell seems to be working ok now. > > The first problem was happening because nt-create-process creates the > process suspended and then uses run-in-other-process to issue a call to > SetConsoleCtrlHandler in the new process before it lets it run. Apparently > this call is doing something (probably allocating memory) which confuses > the MKS shell. I fixed this by deferring the call to SetConsoleCtrlHandler > until after the shell has had a chance to initialize. The only way I could > figure out to do this without changing core XEmacs was to use the > update_status_if_terminated call back which is a bit of a kludge. If > someone knows a better way, let me know. > > The other problem turned out to be caused by the fact that XEmacs was > passing the same handle for both StdOutput and StdError. When the shell > created the pipe between the two commands for something like "ps | sort" it > got confused because it wasn't expecting two copies of the same > handle. The net result was that it lost it's connection to StdError so no > further prompts appeared and I thought the shell had hung. This was easy > to fix by simply making StdError a copy of the StdOutput handle instead of > passing the same handle for both. > > Here is the patch that fixes both problems I'd appreciate it if someone > would give it a try and see if it breaks anything. I've tried this with > the current CVS source as of last night. > > > Mike Alexander Arbortext, Inc. > mta(a)arbortext.com +1-734-327-6803 > > --- process-nt.c 1999/09/11 08:57:25 1.11.2.5 > +++ process-nt.c 1999/10/19 08:24:07 > ＠＠ -53,6 +53,7 ＠＠ > struct nt_process_data > { > HANDLE h_process; > + int need_enable_child_signals; > }; > > #define NT_DATA(p) ((struct nt_process_data*)((p)->process_data)) > ＠＠ -421,7 +422,7 ＠＠ > Lisp_Object *argv, int nargv, > Lisp_Object program, Lisp_Object cur_dir) > { > - HANDLE hmyshove, hmyslurp, hprocin, hprocout; > + HANDLE hmyshove, hmyslurp, hprocin, hprocout, hprocerr; > LPTSTR command_line; > BOOL do_io, windowed; > char *proc_env; > ＠＠ -472,6 +473,10 ＠＠ > CreatePipe (&hprocin, &hmyshove, &sa, 0); > CreatePipe (&hmyslurp, &hprocout, &sa, 0); > > + /* Duplicate the stdout handle for use as stderr */ > + DuplicateHandle(GetCurrentProcess(), hprocout, GetCurrentProcess(), > &hprocerr, > + 0, TRUE, DUPLICATE_SAME_ACCESS); > + > /* Stupid Win32 allows to create a pipe with *both* ends either > inheritable or not. We need process ends inheritable, and local > ends not inheritable. */ > ＠＠ -599,7 +604,7 ＠＠ > { > si.hStdInput = hprocin; > si.hStdOutput = hprocout; > - si.hStdError = hprocout; > + si.hStdError = hprocerr; > si.dwFlags |= STARTF_USESTDHANDLES; > } > > ＠＠ -614,6 +619,7 ＠＠ > /* These just have been inherited; we do not need a copy */ > CloseHandle (hprocin); > CloseHandle (hprocout); > + CloseHandle (hprocerr); > } > > /* Handle process creation failure */ > ＠＠ -640,12 +646,18 ＠＠ > CloseHandle (pi.hProcess); > } > > - if (!windowed) > - enable_child_signals (pi.hProcess); > - > ResumeThread (pi.hThread); > CloseHandle (pi.hThread); > > + /* Remember to enable child signals later if this is not a windowed > + app. Can't do it right now because that screws up the MKS Toolkit > + shell. */ > + if (!windowed) > + { > + NT_DATA(p)->need_enable_child_signals = 10; > + kick_status_notify (); > + } > + > /* Hack to support Windows 95 negative pids */ > return ((int)pi.dwProcessId < 0 > ? -(int)pi.dwProcessId : (int)pi.dwProcessId); > ＠＠ -664,6 +676,18 ＠＠ > nt_update_status_if_terminated (struct Lisp_Process* p) > { > DWORD exit_code; > + > + if (NT_DATA(p)->need_enable_child_signals > 1) > + { > + NT_DATA(p)->need_enable_child_signals -= 1; > + kick_status_notify (); > + } > + else if (NT_DATA(p)->need_enable_child_signals == 1) > + { > + enable_child_signals(NT_DATA(p)->h_process); > + NT_DATA(p)->need_enable_child_signals = 0; > + } > + > if (GetExitCodeProcess (NT_DATA(p)->h_process, &exit_code) > && exit_code != STILL_ACTIVE) > { > ＠＠ -764,6 +788,14 ＠＠ > int current_group, int nomsg) > { > struct Lisp_Process *p = XPROCESS (proc); > + > + /* Enable child signals if necessary. This may lose the first > + but it's better than nothing. */ > + if (NT_DATA(p)->need_enable_child_signals > 0) > + { > + enable_child_signals(NT_DATA(p)->h_process); > + NT_DATA(p)->need_enable_child_signals = 0; > + } > > /* Signal error if SIGNO cannot be sent */ > validate_signal_number (signo); > > >