Index: common.scm
==================================================================
--- common.scm
+++ common.scm
@@ -408,13 +408,22 @@
 
 (define (common:version-changed?)
   (not (equal? (common:get-last-run-version)
                (common:version-signature))))
 
+
+;; From 1.70 to 1.80, db's are compatible.
+
 (define (common:api-changed?)
-  (not (equal? (substring (->string megatest-version) 0 4)
-               (substring (conc (common:get-last-run-version)) 0 4))))
+  (let* (
+    (megatest-major-version (substring (->string megatest-version) 0 4))
+    (run-major-version (substring (conc (common:get-last-run-version)) 0 4))
+   )
+   (and (not (equal? megatest-major-version "1.80"))
+     (not (equal? megatest-major-version megatest-run-version)))
+  )
+)
 
 ;;======================================================================
 ;; Move me elsewhere ...
 ;; RADT => Why do we meed the version check here, this is called only if version misma
 ;;
@@ -599,14 +608,15 @@
 ;;======================================================================
 ;; Force a megatest cleanup-db if version is changed and skip-version-check not specified
 ;; Do NOT check if not on homehost!
 ;;
 (define (common:exit-on-version-changed)
-  (if (common:on-homehost?)
+  (if (and *toppath*              ;; do nothing if *toppath* not yet provided
+	   (common:on-homehost?))
       (if (common:api-changed?)
 	  (let* ((mtconf (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.config"))
-                 (dbfile  (conc (get-environment-variable "MT_RUN_AREA_HOME") "/megatest.db"))
+                 (dbfile  (conc (get-environment-variable "MT_RUN_AREA_HOME") ".megatest/main.db"))
                  (read-only (not (file-write-access? dbfile)))
                  (dbstruct (db:setup #t))) ;; (db:setup-db *dbstruct-dbs* *toppath* #f))) ;;  #t)))
 	    (debug:print 0 *default-log-port*
 			 "WARNING: Version mismatch!\n"
 			 "   expected: " (common:version-signature) "\n"
@@ -626,14 +636,14 @@
                (common:cleanup-db dbstruct)))
              ((not (common:file-exists? mtconf))
               (debug:print 0 *default-log-port* "   megatest.config does not exist in this area.  Cannot proceed with megatest version migration.")
               (exit 1))
              ((not (common:file-exists? dbfile))
-              (debug:print 0 *default-log-port* "   megatest.db does not exist in this area.  Cannot proceed with megatest version migration.")
+              (debug:print 0 *default-log-port* "   .megatest/main.db does not exist in this area.  Cannot proceed with megatest version migration.")
               (exit 1))
              ((not (eq? (current-user-id)(file-owner mtconf)))
-              (debug:print 0 *default-log-port* "   You do not own megatest.db in this area.  Cannot proceed with megatest version migration.")
+              (debug:print 0 *default-log-port* "   You do not own .megatest/main.db in this area.  Cannot proceed with megatest version migration.")
               (exit 1))
              (read-only
               (debug:print 0 *default-log-port* "   You have read-only access to this area.  Cannot proceed with megatest version migration.")
               (exit 1))
              (else
@@ -980,12 +990,13 @@
 ;;======================================================================
 ;; E X I T   H A N D L I N G
 ;;======================================================================
 
 (define (common:run-sync?)
-    (and (common:on-homehost?)
-	 (args:get-arg "-server")))
+  (and *toppath*               ;; gate if called before *toppath* is set
+       (common:on-homehost?)
+       (args:get-arg "-server")))
 
 (define (common:human-time)
   (time->string (seconds->local-time (current-seconds)) "%Y-%m-%d %H:%M:%S"))
 
 
@@ -1989,10 +2000,17 @@
 		    (host-last-used-set! rec curr-time)
 		    new-best)
 		  (if (null? tal) #f (loop (car tal)(cdr tal) best-host)))))))))
 
 (define (common:wait-for-homehost-load maxnormload msg)
+  (let loop ((start-time (current-seconds))) ;; we saw some instances of this being called before *toppath* was set. This might be an early setup race. This delay should help but it is impossible to test...
+    (if (not *toppath*)
+	(begin
+	  (debug:print 0 *default-log-port* "ERROR: common:wait-for-homehost-load called before *toppath* set.")
+	  (thread-sleep! 30)
+	  (if (< (- (current-seconds) start-time) 300)
+	      (loop start-time)))))
   (let* ((hh-dat (if (common:on-homehost?) ;; if we are on the homehost then pass in #f so the calls are local.
                      #f
                      (server:choose-server *toppath* 'homehost)))
          (hh     (if hh-dat (car hh-dat) #f)))
     (common:wait-for-normalized-load maxnormload msg hh)))

Index: dbfile.scm
==================================================================
--- dbfile.scm
+++ dbfile.scm
@@ -328,11 +328,11 @@
       (current-error-port)
     (lambda ()
       (apply print params))))
     
 (define (dbfile:cautious-open-database fname init-proc sync-mode journal-mode #!optional (tries-left 500))
-  (let* ((busy-file  (conc fname"-journal"))
+  (let* ((busy-file  (conc fname "-journal"))
 	 (delay-time (* (- 51 tries-left) 1.1))
       	 (write-access (file-write-access? fname))
          (dir-access (file-write-access? (pathname-directory fname)))
          (retry      (lambda ()
 		       (thread-sleep! delay-time)
@@ -513,11 +513,12 @@
 	  (db:sync-touched dbstruct runid keys dbinit)
 	  (set! *db-sync-in-progress* #f)
 	  (delete-file* lock-file)
 	  #t)
         (begin
-          (dbfile:print-err "INFO: could not get lock for " from-db-file ", sync likely in progress.")
+	  (if (common:low-noise-print 120 (conc "no lock "from-db-file))
+	      (dbfile:print-err "INFO: could not get lock for " from-db-file ", sync likely in progress."))
 	  #f
 	  ))))
 
 ;; ;; Get a lock from the no-sync-db for the from-db, then delta sync the from-db to the to-db, otherwise return #f
 ;; ;;

Index: megatest-version.scm
==================================================================
--- megatest-version.scm
+++ megatest-version.scm
@@ -18,6 +18,6 @@
 ;; Always use two or four digit decimal
 ;; 1.01, 1.02...1.10,1.11,1.1101 ... 1.99,2.00..
 
 ;; (declare (unit megatest-version))
 
-(define megatest-version 1.7009)
+(define megatest-version 1.8002)

Index: server.scm
==================================================================
--- server.scm
+++ server.scm
@@ -190,46 +190,48 @@
   (let ((server-rx    (regexp "^SERVER STARTED: (\\S+):(\\d+) AT ([\\d\\.]+) server-id: (\\S+) pid: (\\d+)")) ;; SERVER STARTED: host:port AT timesecs server id
         (dbprep-rx    (regexp "^SERVER: dbprep"))
         (dbprep-found 0)
 	(bad-dat      (list #f #f #f #f #f)))
     (handle-exceptions
-	exn
-      (begin
-	(debug:print-info 0 *default-log-port* "Unable to get server info from " logf ", exn=" exn)
-	bad-dat) ;; no idea what went wrong, call it a bad server
-      (with-input-from-file
-	  logf
-	(lambda ()
-	  (let loop ((inl  (read-line))
-		     (lnum 0))
-	    (if (not (eof-object? inl))
-		(let ((mlst (string-match server-rx inl))
-                      (dbprep (string-match dbprep-rx inl)))
-                  (if dbprep (set! dbprep-found 1))
-		  (if (not mlst)
-		      (if (< lnum 500) ;; give up if more than 500 lines of server log read
-			  (loop (read-line)(+ lnum 1))
-			  (begin 
+     exn
+     (begin
+       ;; WARNING: this is potentially dangerous to blanket ignore the errors
+       (if (file-exists? logf)
+	   (debug:print-info 0 *default-log-port* "Unable to get server info from "logf", exn=" exn))
+       bad-dat) ;; no idea what went wrong, call it a bad server
+     (with-input-from-file
+	 logf
+       (lambda ()
+	 (let loop ((inl  (read-line))
+		    (lnum 0))
+	   (if (not (eof-object? inl))
+	       (let ((mlst (string-match server-rx inl))
+		     (dbprep (string-match dbprep-rx inl)))
+		 (if dbprep (set! dbprep-found 1))
+		 (if (not mlst)
+		     (if (< lnum 500) ;; give up if more than 500 lines of server log read
+			 (loop (read-line)(+ lnum 1))
+			 (begin 
                            (debug:print-info 0 *default-log-port* "Unable to get server info from first 500 lines of " logf )
                            bad-dat))
-		      (match mlst
-			((_ host port start server-id pid)
-			 (list host
-			       (string->number port)
-			       (string->number start)
-			       server-id
-			       (string->number pid)))
-			(else
-			 (debug:print 0 *current-log-port* "ERROR: did not recognise SERVER line info "mlst)
-			 bad-dat))))
-                (begin 
-                  (if dbprep-found
-                      (begin
-                         (debug:print-info 2 *default-log-port* "Server is in dbprep at " (common:human-time))
-                         (thread-sleep! 0.5)) ;; was 25 sec but that blocked things from starting?
-                      (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (seconds->time-string (current-seconds))))
-		  bad-dat))))))))
+		     (match mlst
+			    ((_ host port start server-id pid)
+			     (list host
+				   (string->number port)
+				   (string->number start)
+				   server-id
+				   (string->number pid)))
+			    (else
+			     (debug:print 0 *current-log-port* "ERROR: did not recognise SERVER line info "mlst)
+			     bad-dat))))
+	       (begin 
+		 (if dbprep-found
+		     (begin
+		       (debug:print-info 2 *default-log-port* "Server is in dbprep at " (common:human-time))
+		       (thread-sleep! 0.5)) ;; was 25 sec but that blocked things from starting?
+		     (debug:print-info 0 *default-log-port* "Unable to get server info from " logf " at " (seconds->time-string (current-seconds))))
+		 bad-dat))))))))
 
 ;; ;; get a list of servers from the log files, with all relevant data
 ;; ;; ( mod-time host port start-time pid )
 ;; ;;
 ;; (define (server:get-list areapath #!key (limit #f))
@@ -419,10 +421,11 @@
 
 ;; oldest server alive determines host then choose random of youngest
 ;; five servers on that host
 ;;
 (define (server:get-servers-info areapath)
+  (assert *toppath* "FATAL: server:get-servers-info called before *toppath* has been set.")
   (let* ((servinfodir (conc *toppath*"/.servinfo")))
     (if (not (file-exists? servinfodir))
 	(create-directory servinfodir))
     (let* ((allfiles    (glob (conc servinfodir"/*")))
 	   (res         (make-hash-table)))
@@ -468,13 +471,13 @@
 	       (oldest-dat (hash-table-ref serversdat oldest))
 	       (host       (list-ref oldest-dat 0))
 	       (all-valid  (filter (lambda (x)
 				     (equal? host (list-ref (hash-table-ref serversdat x) 0)))
 				   by-time-asc))
-	       (best-five  (lambda ()
-			     (if (> (length all-valid) 5)
-				 (take all-valid 5)
+	       (best-ten  (lambda ()
+			     (if (> (length all-valid) 10)
+				 (take all-valid 10)
 				 all-valid)))
 	       (names->dats (lambda (names)
 			      (map (lambda (x)
 				     (hash-table-ref serversdat x))
 				   names)))
@@ -488,15 +491,15 @@
 	     (print "oldest: "oldest-dat", selected host: "host", all-valid: "all-valid)
 	     (print "youngest: "(hash-table-ref serversdat (car all-valid))))
 	    ((home)     host)
 	    ((homehost) (cons host (am-home?))) ;; shut up old code
 	    ((home?)    (am-home?))
-	    ((best-five)(names->dats (best-five)))
+	    ((best-ten)(names->dats (best-ten)))
 	    ((all-valid)(names->dats all-valid))
-	    ((best)     (let* ((best-five (best-five))
-			       (len       (length best-five)))
-			  (hash-table-ref serversdat (list-ref best-five (random len)))))
+	    ((best)     (let* ((best-ten (best-ten))
+			       (len       (length best-ten)))
+			  (hash-table-ref serversdat (list-ref best-ten (random len)))))
 	    ((count)(length all-valid))
 	    (else
 	     (debug:print 0 *default-log-port* "ERROR: invalid command "mode)
 	     #f)))
 	(begin
@@ -516,11 +519,11 @@
 ;;
 (define (server:kind-run areapath)
   ;; look for $MT_RUN_AREA_HOME/logs/server-start-last
   ;; and wait for it to be at least <server idletime> seconds old
   ;; (server:wait-for-server-start-last-flag areapath)
-  (if (< (server:choose-server areapath 'count) 10)
+  (if (< (server:choose-server areapath 'count) 20)
       (server:run areapath))
   #;(if (not (server:check-if-running areapath)) ;; why try if there is already a server running?
       (let* ((lock-file    (conc areapath "/logs/server-start.lock")))
 	(let* ((start-flag (conc areapath "/logs/server-start-last")))
 	  (common:simple-file-lock-and-wait lock-file expire-time: 25)
@@ -555,11 +558,11 @@
 
 ;; no longer care if multiple servers are started by accident. older servers will drop off in time.
 ;;
 (define (server:check-if-running areapath) ;;  #!key (numservers "2"))
   (let* ((ns            (server:get-num-servers)) ;; get the setting the for maximum number of servers allowed
-	 (servers       (server:choose-server areapath 'best-five))) ;; (server:get-best (server:get-list areapath))))
+	 (servers       (server:choose-server areapath 'best-ten))) ;; (server:get-best (server:get-list areapath))))
     (if (or (and servers
 		 (null? servers))
 	    (not servers))
 	    ;; (and (list? servers)
 	    ;;	 (< (length servers) (+ 1 (random ns))))) ;; somewhere between 1 and numservers