>>>>> "Mike" == Mike Alexander <mta(a)arbortext.com&gt; writes: Mike> Here are some patches to make nt_create_process work properly with Mike> the Korn shell that comes with the MKS Toolkit from Mortice-Kern Mike> Systems (and perhaps with other programs). With this patch, running Mike> that shell under XEmacs works ok. Without it any attempt to start a Mike> subshell causes an error message about "Forked child base mismatch". Mike> This is apparently because XEmacs creates a thread in the newly Mike> started process and executes code in that thread before starting the Mike> main thread. I fixed this by deferring this call until after the Mike> main thread has a chance to get started. Hello Mike, this patch of yours breaks (interrupt-process ...) for processes running under a shell. This is why kill-compilation is fails to fulfill its purpose in 21.1.9, 21.1.10, 21.2-b3x. First I verified today that kill-compilation works in 21.1.7. Secondly I verified kill-compilation not to work in 21.2 (beta33) Thirdly I backed out following part of your patch manually (relative to 21.2-b33: Sure enough kill-compilation works again in 21.2-b33! Could you please come up with a patch which makes MKS shell working without breaking kill-compilation. You may use (compile "sleep 600") (interrupt-process "compilation") as a test-case. This can also be used to verify your fix: (start-process "shell" nil "cmd" "/c" "sleep" "5") (message "before: %S" (get-process "shell")) (interrupt-process "shell" t) (message "after: %S" (get-process "shell")) Please note that running above four sexps inside a defun will not work stably. Even intervening (sit-for 1) calls didn't work reliably for me. Please just evaluate them in order with C-x C-e (eval-last-sexp). Best regards, Adrian === cd d:\tmp\21.2\xemacs\src\ === C:\cvs\cvs diff -u process-nt.c Index: process-nt.c =================================================================== RCS file: /usr/CVSroot/XEmacs/xemacs/src/process-nt.c,v retrieving revision 1.11.2.18 diff -u -u -r1.11.2.18 process-nt.c --- process-nt.c 2000/03/29 18:35:34 1.11.2.18 +++ process-nt.c 2000/05/07 10:56:10 ＠＠ -923,17 +923,31 ＠＠ 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 0 + /* + APA: Back-out + Message-Id: <v04220805b45fd4ac8e03(a)[148.59.233.133]&gt; + Date: Tue, 23 Nov 1999 00:36:47 -0500 + To: xemacs-patches(a)xemacs.org + From: Mike Alexander <mta(a)arbortext.com&gt; + Subject: Fixing nt_create_process for MKS Toolkit shell + + 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 (); } +#endif return ((int)pi.dwProcessId); } === Exit status: 1 Mike> There is also a problem with pipes in the shell (in the Mike> shell sense, e.g. "ls | sort"). XEmacs passes the same Mike> handle for stdout and stderr and this confuses the shell Mike> when it creates the pipes to connect the two programs Mike> together. Instead XEmacs needs to pass two separate handles Mike> to the same end of the same pipe. This doesn't affect the Mike> rest of XEmacs since both stderr and stdout still come back Mike> on the same pipe, but it avoids confusing the shell. Mike> This patch has been sent to the xemacs-nt list before, but I'm Mike> sending it to xemacs-patches to make it official. Mike> Mike Alexander Mike> Index: src/process-nt.c Mike> =================================================================== Mike> RCS file: /usr/CVSroot/XEmacs/xemacs/src/process-nt.c,v Mike> retrieving revision 1.11.2.6 Mike> diff -u -r1.11.2.6 process-nt.c Mike> --- process-nt.c 1999/10/24 09:25:03 1.11.2.6 Mike> +++ process-nt.c 1999/11/23 04:35:20 Mike> ＠＠ -53,6 +53,7 ＠＠ Mike> struct nt_process_data Mike> { Mike> HANDLE h_process; Mike> + int need_enable_child_signals; Mike> }; Mike> #define NT_DATA(p) ((struct nt_process_data*)((p)->process_data)) Mike> ＠＠ -421,7 +422,7 ＠＠ Mike> Lisp_Object *argv, int nargv, Mike> Lisp_Object program, Lisp_Object cur_dir) Mike> { Mike> - HANDLE hmyshove, hmyslurp, hprocin, hprocout; Mike> + HANDLE hmyshove, hmyslurp, hprocin, hprocout, hprocerr; Mike> LPTSTR command_line; Mike> BOOL do_io, windowed; Mike> char *proc_env; Mike> ＠＠ -472,6 +473,10 ＠＠ Mike> CreatePipe (&hprocin, &hmyshove, &sa, 0); Mike> CreatePipe (&hmyslurp, &hprocout, &sa, 0); Mike> + /* Duplicate the stdout handle for use as stderr */ Mike> + DuplicateHandle(GetCurrentProcess(), hprocout, Mike> GetCurrentProcess(), &hprocerr, Mike> + 0, TRUE, DUPLICATE_SAME_ACCESS); Mike> + Mike> /* Stupid Win32 allows to create a pipe with *both* ends either Mike> inheritable or not. We need process ends inheritable, and local Mike> ends not inheritable. */ Mike> ＠＠ -599,7 +604,7 ＠＠ Mike> { Mike> si.hStdInput = hprocin; Mike> si.hStdOutput = hprocout; Mike> - si.hStdError = hprocout; Mike> + si.hStdError = hprocerr; Mike> si.dwFlags |= STARTF_USESTDHANDLES; Mike> } Mike> ＠＠ -614,6 +619,7 ＠＠ Mike> /* These just have been inherited; we do not need a copy */ Mike> CloseHandle (hprocin); Mike> CloseHandle (hprocout); Mike> + CloseHandle (hprocerr); Mike> } Mike> /* Handle process creation failure */ Mike> ＠＠ -640,12 +646,18 ＠＠ Mike> CloseHandle (pi.hProcess); Mike> } Mike> - if (!windowed) Mike> - enable_child_signals (pi.hProcess); Mike> - Mike> ResumeThread (pi.hThread); Mike> CloseHandle (pi.hThread); Mike> + /* Remember to enable child signals later if this is not a windowed Mike> + app. Can't do it right now because that screws up the MKS Toolkit Mike> + shell. */ Mike> + if (!windowed) Mike> + { Mike> + NT_DATA(p)->need_enable_child_signals = 10; Mike> + kick_status_notify (); Mike> + } Mike> + Mike> /* Hack to support Windows 95 negative pids */ Mike> return ((int)pi.dwProcessId < 0 Mike> ? -(int)pi.dwProcessId : (int)pi.dwProcessId); Mike> ＠＠ -664,6 +676,18 ＠＠ Mike> nt_update_status_if_terminated (struct Lisp_Process* p) Mike> { Mike> DWORD exit_code; Mike> + Mike> + if (NT_DATA(p)->need_enable_child_signals > 1) Mike> + { Mike> + NT_DATA(p)->need_enable_child_signals -= 1; Mike> + kick_status_notify (); Mike> + } Mike> + else if (NT_DATA(p)->need_enable_child_signals == 1) Mike> + { Mike> + enable_child_signals(NT_DATA(p)->h_process); Mike> + NT_DATA(p)->need_enable_child_signals = 0; Mike> + } Mike> + Mike> if (GetExitCodeProcess (NT_DATA(p)->h_process, &exit_code) Mike> && exit_code != STILL_ACTIVE) Mike> { Mike> ＠＠ -764,6 +788,14 ＠＠ Mike> int current_group, int nomsg) Mike> { Mike> struct Lisp_Process *p = XPROCESS (proc); Mike> + Mike> + /* Enable child signals if necessary. This may lose the first Mike> + but it's better than nothing. */ Mike> + if (NT_DATA(p)->need_enable_child_signals > 0) Mike> + { Mike> + enable_child_signals(NT_DATA(p)->h_process); Mike> + NT_DATA(p)->need_enable_child_signals = 0; Mike> + } Mike> /* Signal error if SIGNO cannot be sent */ Mike> validate_signal_number (signo);