emcmodule: retry the tool mmap attach in stat.poll()

src/emc/tooldata/tooldata_mmap.cc

@@ -22,7 +22,9 @@
 #include <unistd.h> // write(2),lseek(2)
 #include <fcntl.h> // open(2)
 #include <sys/mman.h>
+#include <sys/stat.h> // fstat(2)
 #include <string.h>
+#include <errno.h>
 #include "config.h"
 #include <rtapi_mutex.h>
 #include "tooldata.hh"
@@ -185,10 +187,24 @@ int tool_mmap_creator(EMC_TOOL_STAT const * ptr,int random_toolchanger)
 //typ: milltask, guis (emcmodule,emcsh,...), halui
 int tool_mmap_user()
 {
+    static bool warned = false;
+    char why[LINELEN] = {};
     int fd = open(tool_mmap_fname(),
                   TOOL_MMAP_USER_OPEN_FLAGS, TOOL_MMAP_MODE);
+    struct stat st;
 
     if (fd < 0) {
+        snprintf(why, sizeof(why), "open: %s", strerror(errno));
+    } else if (fstat(fd, &st) < 0) {
+        snprintf(why, sizeof(why), "fstat: %s", strerror(errno));
+    } else if ((size_t)st.st_size < (size_t)(TOOL_MMAP_SIZE)) {
+        // The creator truncates the file before extending it: mapping
+        // it while short would SIGBUS on first access.
+        snprintf(why, sizeof(why), "size %zu < %zu",
+                 (size_t)st.st_size, (size_t)(TOOL_MMAP_SIZE));
+    }
+    if (*why) {
+        if (fd >= 0) { close(fd); }
         /*
         ** For the LinuxCNC application, tool_mmap_creator()
         ** should start first and create the mmap file needed here.
@@ -197,7 +213,12 @@ int tool_mmap_user()
         ** continue execution if no mmap file is open.
         ** So print message and return fail indicator.
         */
-        fprintf(stderr,"tool_mmap_user(): tool mmap not available\n");
+        // callers may retry (emcmodule poll()), so warn only once
+        if (!warned) {
+            warned = true;
+            fprintf(stderr,"tool_mmap_user(): tool mmap not available: %s: %s\n",
+                    tool_mmap_fname(), why);
+        }
         tool_mmap_base = (char*)NULL;
         return(-1);
     }

src/emc/usr_intf/axis/extensions/emcmodule.cc

@@ -1006,13 +1006,19 @@ static PyObject *poll(pyStatChannel *s, PyObject * /*o*/) {
       initialized=1;
     }
 #else //}{
-    static bool mmap_available = 1;
-    if (!mmap_available) return NULL;
+    // task creates the mmap file; a GUI may poll before that, so retry
+    // the attach on every poll until it succeeds. Until then
+    // initialized stays 0 and the tool accessors report no data.
+    static bool warned = 0;
     if (!initialized) {
-        initialized=1;
-        if (tool_mmap_user()) {
-          mmap_available = 0;
-          fprintf(stderr,"poll(): continuing without tool mmap data\n");
+        if (tool_mmap_user() == 0) {
+            initialized=1;
+            if (warned) {
+                fprintf(stderr,"poll(): tool mmap available now\n");
+            }
+        } else if (!warned) {
+            warned = 1;
+            fprintf(stderr,"poll(): tool mmap not available yet, retrying\n");
         }
     }
 #endif //}

tests/ui-smoke/_lib/drive.py

@@ -99,10 +99,8 @@ def connect_and_wait_ready(timeout):
     still initialising; recreate the stat object on every iteration so
     a stale invalid buffer does not stick after linuxcncsvr is ready.
 
-    Catch the full Exception hierarchy: in early startup stat.poll()
-    can raise SystemError ('error return without exception set') when
-    the underlying C function reports failure without setting a Python
-    exception. Treat that the same as linuxcnc.error and retry."""
+    Catch the full Exception hierarchy so an unexpected error during
+    startup is retried like linuxcnc.error."""
     deadline = time.monotonic() + timeout
     last_err = None
     while time.monotonic() < deadline: