Re: Patches Applied for 21.1.8

Craig> I'm afraid not :-( Hi Craig, I'm so far past my wit's end, that I'm getting serious again! I have revisited my options and settled for minimal change. This is the realistic route to get this into 21.1.8 (no, I havn't given up just yet, I hope you haven't either). This patch is against 21.2-b19 again, but it's smaller. I have reverted all my indentation changes (which make the functions I touched comply with GNU formatting standards). Reviewing each patch hunk I have re-implemented the whole thing. I invert 95/98 PIDs in create_child and keep them that way when converting to Lisp_Int. Everywhere Lisp_Int PIDs are converted back to int, I invert them (when on OS_WIN95) before use. Wherever I call OpenProcess I invert the PID for OS_WIN95. I have proven my patch to be correct by inverting all tests from if (os_subtype == OS_WIN95) to if (os_subtype != OS_WIN95) Now I get these warnings on NT: Fcall_process_internal.child_setup returns PID -306, using PID 306 create_child.CreateProcess abs(PID) -343 > EMACS_INT_MAX (2147483647) Fcall_process_internal.child_setup returns PID -343, using PID 343 create_child.CreateProcess abs(PID) -282 > EMACS_INT_MAX (2147483647) Fcall_process_internal.child_setup returns PID -282, using PID 282 create_child.CreateProcess abs(PID) -280 > EMACS_INT_MAX (2147483647) Fcall_process_internal.child_setup returns PID -280, using PID 280 create_child.CreateProcess abs(PID) -146 > EMACS_INT_MAX (2147483647) Fcall_process_internal.child_setup returns PID -146, using PID 146 create_child.CreateProcess abs(PID) -298 > EMACS_INT_MAX (2147483647) Fcall_process_internal.child_setup returns PID -298, using PID 298 BUT, call-process still works, returning correct exit status and never obtaining invalid process handles. This proves that negative PIDs can be passed to Lisp and back. On Windows 95/98 we will need the opposite, but that requires one bit less. I had to disable a meaningless fork test in src/callproc.c for WINDOWSNT, which I intend to leave in: #ifndef WINDOWSNT if (pid < 0) { if (fd[0] >= 0) close (fd[0]); report_file_error ("Doing fork", Qnil); } #endif Please try this patch and, by all means, report success! Are there any more testers on 95/98 out there? Testing with and without USE_UNION_TYPE is desirable. Please refer to earlier test-cases in this thread and send me their output, as well as any STDERR of XEmacs itself. Thanks in advance! Adrian Craig> That is correct. Patch against 21.2-b19 (to be applied in xemacs/src): cd d:\tmp\21.2\xemacs\src\ cvs diff Compilation started at Sat Oct 23 17:42:11 1999 +0200 (W. Europe Daylight Time) ? call-process-notes ? call-process.patch ? call-process.txt ? cvs-diff ? depend-of-make-src-depend ? Installation ? it ? runxemacs.pdb ? sysdep.c.good ? temacs.bsc ? temacs.ilk ? temacs.map ? temacs.pdb ? waitpid.txt ? xemacs.opt cvs server: Diffing . Index: callproc.c =================================================================== RCS file: /usr/CVSroot/XEmacs/xemacs/src/callproc.c,v retrieving revision 1.29.2.6 diff -u -r1.29.2.6 callproc.c --- callproc.c 1999/05/01 05:13:23 1.29.2.6 +++ callproc.c 1999/10/23 15:51:17 ＠＠ -45,6 +45,7 ＠＠ #define _P_NOWAIT 1 /* from process.h */ #include <windows.h> #include "nt.h" +#include "ntheap.h" #endif #ifdef DOS_NT ＠＠ -86,14 +87,22 ＠＠ static Lisp_Object call_process_kill (Lisp_Object fdpid) { - Lisp_Object fd = Fcar (fdpid); - Lisp_Object pid = Fcdr (fdpid); + Lisp_Object lfd = Fcar (fdpid); + Lisp_Object lpid = Fcdr (fdpid); + int fd = XINT (lfd); + int pid = XINT (lpid); - if (!NILP (fd)) - close (XINT (fd)); +#ifdef WINDOWSNT + if (os_subtype == OS_WIN95) + { + pid = -pid; + } +#endif + if (!NILP (lfd)) + close (fd); - if (!NILP (pid)) - EMACS_KILLPG (XINT (pid), SIGKILL); + if (!NILP (lpid)) + EMACS_KILLPG (pid, SIGKILL); synch_process_alive = 0; return Qnil; ＠＠ -105,6 +114,12 ＠＠ int fd = XINT (Fcar (fdpid)); int pid = XINT (Fcdr (fdpid)); +#ifdef WINDOWSNT + if (os_subtype == OS_WIN95) + { + pid = -pid; + } +#endif if (!call_process_exited && EMACS_KILLPG (pid, SIGINT) == 0) { ＠＠ -114,7 +129,11 ＠＠ /* #### "c-G" -- need non-consing Single-key-description */ message ("Waiting for process to die...(type C-g again to kill it instantly)"); +#ifdef WINDOWSNT + wait_for_termination (OpenProcess(PROCESS_ALL_ACCESS, 0, pid)); +#else wait_for_termination (pid); +#endif /* "Discard" the unwind protect. */ XCAR (fdpid) = Qnil; ＠＠ -170,6 +189,9 ＠＠ Lisp_Object infile, buffer, current_dir, display, path; int fd[2]; int filefd; +#ifdef WINDOWSNT + HANDLE pHandle; +#endif int pid; char buf[16384]; char *bufptr = buf; ＠＠ -359,6 +381,20 ＠＠ #ifdef WINDOWSNT pid = child_setup (filefd, fd1, fd_error, new_argv, (char *) XSTRING_DATA (current_dir)); + /* OpenProcess() as soon after child_setup as possible. It's too + late once the process terminated. */ + if (os_subtype == OS_WIN95) + { + stderr_out ("Fcall_process_internal.child_setup returns PID %d, using PID %d\n", pid, -pid); + /* create_child() in ntproc.c hacks PIDs for OS_WIN95 to fit + into Lisp_Int, i.e. negates them. Hence we have to use -pid + to get at the real PID for OS_WIN95. */ + pHandle = OpenProcess(PROCESS_ALL_ACCESS, 0, -pid); + } + else + { + pHandle = OpenProcess(PROCESS_ALL_ACCESS, 0, pid); + } #else /* not WINDOWSNT */ pid = fork (); ＠＠ -394,12 +430,14 ＠＠ if (!NILP (fork_error)) signal_error (Qfile_error, fork_error); +#ifndef WINDOWSNT if (pid < 0) { if (fd[0] >= 0) close (fd[0]); report_file_error ("Doing fork", Qnil); } +#endif if (INTP (buffer)) { ＠＠ -507,7 +545,11 ＠＠ QUIT; /* Wait for it to terminate, unless it already has. */ +#ifdef WINDOWSNT + wait_for_termination (pHandle); +#else wait_for_termination (pid); +#endif /* Don't kill any children that the subprocess may have left behind when exiting. */ Index: ntproc.c =================================================================== RCS file: /usr/CVSroot/XEmacs/xemacs/src/ntproc.c,v retrieving revision 1.14.2.10 diff -u -r1.14.2.10 ntproc.c --- ntproc.c 1999/10/10 12:40:13 1.14.2.10 +++ ntproc.c 1999/10/23 15:51:28 ＠＠ -446,15 +446,19 ＠＠ cp->procinfo.hProcess=NULL; /* Hack for Windows 95, which assigns large (ie negative) pids */ - if (cp->pid < 0) - cp->pid = -cp->pid; +#ifdef WINDOWSNT + if (os_subtype == OS_WIN95) + { + cp->pid = -cp->pid; + } +#endif /* pid must fit in a Lisp_Int */ -#ifdef USE_UNION_TYPE - cp->pid = (cp->pid & ((1U << VALBITS) - 1)); -#else - cp->pid = (cp->pid & VALMASK); -#endif + + if (cp->pid > EMACS_INT_MAX) { + stderr_out ("create_child.CreateProcess abs(PID) %d > EMACS_INT_MAX (%d)\n", + cp->pid, EMACS_INT_MAX); + } *pPid = cp->pid; Index: sysdep.c =================================================================== RCS file: /usr/CVSroot/XEmacs/xemacs/src/sysdep.c,v retrieving revision 1.32.2.7 diff -u -r1.32.2.7 sysdep.c --- sysdep.c 1999/07/05 05:56:44 1.32.2.7 +++ sysdep.c 1999/10/23 15:51:45 ＠＠ -234,8 +234,11 ＠＠ #endif /* NO_SUBPROCESSES */ -void -wait_for_termination (int pid) +#ifdef WINDOWSNT +void wait_for_termination (HANDLE pid) +#else +void wait_for_termination (int pid) +#endif { /* #### With the new improved SIGCHLD handling stuff, there is much less danger of race conditions and some of the comments below ＠＠ -345,6 +348,47 ＠＠ Since implementations may add their own error indicators on top, we ignore it by default. */ +#elif defined (WINDOWSNT) + int ret = 0, status = 0; + if (pid == NULL) + { + stderr_out ("wait_for_termination: pid == NULL, GetLastError () = %d, (int)pid = %d\n", GetLastError (), (int)pid); + return; + } + do + { + QUIT; + ret = WaitForSingleObject(pid, 100); + } + while (ret == WAIT_TIMEOUT); + if (ret == WAIT_FAILED) + { + stderr_out ("wait_for_termination.WaitForSingleObject returns %d (WAIT_FAILED) GetLastError () %d for pid %d\n", ret, GetLastError (), (int)pid); + } + if (ret == WAIT_ABANDONED) + { + stderr_out ("wait_for_termination.WaitForSingleObject returns %d (WAIT_ABANDONED) GetLastError () %d for pid %d\n", ret, GetLastError (), (int)pid); + } + if (ret == WAIT_OBJECT_0) + { + ret = GetExitCodeProcess(pid, &status); + if (ret) + { + synch_process_alive = 0; + synch_process_retcode = status; + } + else + { + /* GetExitCodeProcess() didn't return a valid exit status, + nothing to do. APA */ + stderr_out ("wait_for_termination.GetExitCodeProcess status %d GetLastError () %d for pid %d\n", status, GetLastError (), (int)pid); + } + } + if (!CloseHandle(pid)) + { + stderr_out ("wait_for_termination.CloseHandle GetLastError () %d for pid %d\n", + GetLastError (), (int)pid); + } #elif defined (EMACS_BLOCK_SIGNAL) && !defined (BROKEN_WAIT_FOR_SIGNAL) && defined (SIGCHLD) while (1) { ＠＠ -376,7 +420,7 ＠＠ Try defining BROKEN_WAIT_FOR_SIGNAL. */ EMACS_WAIT_FOR_SIGNAL (SIGCHLD); } -#else /* not HAVE_WAITPID and (not EMACS_BLOCK_SIGNAL or BROKEN_WAIT_FOR_SIGNAL) */ +#else /* not HAVE_WAITPID and not WINDOWSNT and (not EMACS_BLOCK_SIGNAL or BROKEN_WAIT_FOR_SIGNAL) */ /* This approach is kind of cheesy but is guaranteed(?!) to work for all systems. */ while (1) ＠＠ -578,7 +622,11 ＠＠ static void sys_subshell (void) { +#ifdef WINDOWSNT + HANDLE pid; +#else int pid; +#endif struct save_signal saved_handlers[5]; Lisp_Object dir; unsigned char *str = 0; ＠＠ -617,7 +665,7 ＠＠ xyzzy: #ifdef WINDOWSNT - pid = -1; + pid = NULL; #else /* not WINDOWSNT */ pid = fork (); ＠＠ -651,7 +699,7 ＠＠ #ifdef WINDOWSNT /* Waits for process completion */ pid = _spawnlp (_P_WAIT, sh, sh, NULL); - if (pid == -1) + if (pid == NULL) write (1, "Can't execute subshell", 22); #else /* not WINDOWSNT */ Index: sysdep.h =================================================================== RCS file: /usr/CVSroot/XEmacs/xemacs/src/sysdep.h,v retrieving revision 1.11.2.1 diff -u -r1.11.2.1 sysdep.h --- sysdep.h 1998/12/05 16:56:31 1.11.2.1 +++ sysdep.h 1999/10/23 15:51:46 ＠＠ -48,7 +48,12 ＠＠ /* Wait for subprocess with process id `pid' to terminate and make sure it will get eliminated (not remain forever as a zombie) */ +#ifdef WINDOWSNT +#include <windows.h> +void wait_for_termination (HANDLE pid); +#else void wait_for_termination (int pid); +#endif /* flush any pending output * (may flush input as well; it does not matter the way we use it) cvs server: Diffing m cvs server: Diffing s Compilation exited abnormally with code 1 at Sat Oct 23 17:43:17