Megatest

# make install CSCOPTS='-accumulate-profile -profile-name $(PWD)/profile-ww$(shell date +%V.%u)'
PREFIX=$(PWD)
CSCOPTS= 
INSTALL=install
SRCFILES = common.scm items.scm launch.scm \
           ods.scm runconfig.scm server.scm configf.scm \
           db.scm keys.scm margs.scm megatest-version.scm \
           process.scm runs.scm tasks.scm tests.scm genexample.scm \
	   http-transport.scm nmsg-transport.scm filedb.scm \
           client.scm gutils.scm synchash.scm daemon.scm mt.scm dcommon.scm \
	   tree.scm ezsteps.scm lock-queue.scm sdb.scm \
	   rmt.scm api.scm tdb.scm portlogger.scm

# Eggs to install (straightforward ones)
EGGS=matchable readline apropos base64 regex-literals format regex-case test coops trace csv \
     dot-locking posix-utils posix-extras directory-utils hostinfo tcp-server rpc csv-xml fmt \

    get-steps-data
    login
    testmeta-get-record))

;; These are called by the server on recipt of /api calls
;;    - keep it simple, only return the actual result of the call, i.e. no meta info here
;;
;;    - returns #( flag result )
;;
(define (api:execute-requests dbstruct dat)
  (handle-exceptions
   exn
   (let ((call-chain (get-call-chain)))
     (vector #f (vector exn call-chain dat))) ;; return some stuff for debug if an exception happens
   (if (not (vector? dat))                    ;; it is an error to not receive a vector
       (vector #f #f "remote must be called with a vector")       
       (vector                                   ;; return a vector + the returned data structure
	#t 
	(let ((cmd    (vector-ref dat 0))
	      (params (vector-ref dat 1)))
	  (case (if (symbol? cmd)
		    cmd
		    (string->symbol cmd))
	    ;; SERVERS
	    ((start-server)                 (apply server:kind-run params))
	    ((kill-server)                  (set! *server-run* #f))
	    
	    ;; KEYS
	    ((get-key-val-pairs)            (apply db:get-key-val-pairs dbstruct params))
	    ((get-keys)                     (db:get-keys dbstruct))

	    ;; TESTS
	    ((test-toplevel-num-items)         (apply db:test-toplevel-num-items dbstruct params))
	    ((get-test-info-by-id)	       (apply db:get-test-info-by-id dbstruct params))
	    ((test-get-rundir-from-test-id)    (apply db:test-get-rundir-from-test-id dbstruct params))
	    ((test-set-state-status-by-id)     (apply db:test-set-state-status-by-id dbstruct params))
	    ((get-count-tests-running)         (apply db:get-count-tests-running dbstruct params))
	    ((get-count-tests-running-in-jobgroup) (apply db:get-count-tests-running-in-jobgroup dbstruct params))
	    ((delete-test-records)             (apply db:delete-test-records dbstruct params))
	    ;; ((delete-test-step-records)        (apply db:delete-test-step-records dbstruct params))
	    ((delete-old-deleted-test-records) (apply db:delete-old-deleted-test-records dbstruct params))
	    ((test-set-status-state)           (apply db:test-set-status-state dbstruct params))
	    ((get-previous-test-run-record)    (apply db:get-previous-test-run-record dbstruct params))
	    ((get-matching-previous-test-run-records)(apply db:get-matching-previous-test-run-records dbstruct params))
	    ((test-get-logfile-info)           (apply db:test-get-logfile-info dbstruct params))
	    ((test-get-records-for-index-file)  (apply db:test-get-records-for-index-file dbstruct params))
	    ((get-testinfo-state-status)       (apply db:get-testinfo-state-status dbstruct params))
	    ((test-set-top-process-pid)        (apply db:test-set-top-process-pid dbstruct params))
	    ((test-get-top-process-pid)        (apply db:test-get-top-process-pid dbstruct params))
	    ((test-get-paths-matching-keynames-target-new) (apply db:test-get-paths-matching-keynames-target-new dbstruct params))
	    ((get-prereqs-not-met)             (apply db:get-prereqs-not-met dbstruct params))
	    ((roll-up-pass-fail-counts)        (apply db:roll-up-pass-fail-counts dbstruct params))
	    ((update-fail-pass-counts)         (apply db:general-call dbstruct 'update-pass-fail-counts params))
	    ((get-count-tests-running-for-run-id) (apply db:get-count-tests-running-for-run-id dbstruct params))

	    ;; RUNS
	    ((get-run-info)                 (apply db:get-run-info dbstruct params))
	    ((get-run-status)               (apply db:get-run-status dbstruct params))
	    ((set-run-status)               (apply db:set-run-status dbstruct params))
	    ((register-run)                 (apply db:register-run dbstruct params))
	    ((set-tests-state-status)       (apply db:set-tests-state-status dbstruct params))
	    ((get-tests-for-run)            (apply db:get-tests-for-run dbstruct params))
	    ((get-test-id)                  (apply db:get-test-id dbstruct params))
	    ((get-tests-for-runs-mindata)   (apply db:get-tests-for-runs-mindata dbstruct params))
	    ((get-run-name-from-id)         (apply db:get-run-name-from-id dbstruct params))
	    ((delete-run)                   (apply db:delete-run dbstruct params))
	    ((get-runs)                     (apply db:get-runs dbstruct params))
	    ((get-all-run-ids)              (db:get-all-run-ids dbstruct))
	    ((get-prev-run-ids)             (apply db:get-prev-run-ids dbstruct params))
	    ((get-run-ids-matching-target)  (apply db:get-run-ids-matching-target dbstruct params))
	    ((get-runs-by-patt)             (apply db:get-runs-by-patt dbstruct params))
	    ((lock/unlock-run)              (apply db:lock/unlock-run dbstruct params))
	    ((update-run-event_time)        (apply db:update-run-event_time dbstruct params))
	    ((find-and-mark-incomplete)     (apply db:find-and-mark-incomplete dbstruct params))

	    ;; STEPS
	    ((teststep-set-status!)         (apply db:teststep-set-status! dbstruct params))

	    ;; TEST DATA
	    ((test-data-rollup)             (apply db:test-data-rollup dbstruct params))
	    ((csv->test-data)               (apply db:csv->test-data dbstruct params))
	    ((get-steps-data)               (apply db:get-steps-data dbstruct params))

	    ;; MISC
	    ((login)                        (apply db:login dbstruct params))
	    ((general-call)                 (let ((stmtname   (car params))
						  (run-id     (cadr params))
						  (realparams (cddr params)))
					      (db:with-db dbstruct run-id #t ;; these are all for modifying the db
							  (lambda (db)
							    (db:general-call db stmtname realparams)))))
	    ((sync-inmem->db)               (let ((run-id (car params)))
					      (db:sync-touched dbstruct run-id force-sync: #t)))
	    ((sdb-qry)                      (apply sdb:qry params))
	    ((ping)                         (current-process-id))

	    ;; TESTMETA
	    ((testmeta-get-record)       (apply db:testmeta-get-record dbstruct params))
	    ((testmeta-add-record)       (apply db:testmeta-add-record dbstruct params))
	    ((testmeta-update-field)     (apply db:testmeta-update-field dbstruct params))))))))



;; http-server  send-response
;;                 api:process-request
;;                    db:*
;;
;; NB// Runs on the server as part of the server loop
;;
(define (api:process-request dbstruct $) ;; the $ is the request vars proc
  (let* ((cmd     ($ 'cmd))
	 (paramsj ($ 'params))
	 (params  (db:string->obj paramsj)) ;; (rmt:json-str->dat paramsj))
	 (resdat  (api:execute-requests dbstruct cmd params)) ;; #( flag result )
	 (res     (vector-ref resdat 1)))

    ;; This can be here but needs controls to ensure it doesn't run more than every 4 seconds
    ;; (rmt:dat->json-str
    ;;  (if (or (string? res)
    ;;          (list?   res)
    ;;          (number? res)
    ;;          (boolean? res))

;;   2. We are a run tests, list runs or other interactive process and we must figure out
;;      *transport-type* and *runremote* from the monitor.db
;;
;; client:setup
;;
;; lookup_server, need to remove *runremote* stuff
;;
(define (client:setup run-id #!key (remaining-tries 2) (failed-connects 0))
  (debug:print-info 2 "client:setup remaining-tries=" remaining-tries)
  (let* ((tdbdat (tasks:open-db)))
    (if (<= remaining-tries 0)
	(begin
	  (debug:print 0 "ERROR: failed to start or connect to server for run-id " run-id)
	  (exit 1))























	(let* ((server-dat (tasks:get-server (db:delay-if-busy tdbdat) run-id)))
	  (debug:print-info 4 "client:setup server-dat=" server-dat ", remaining-tries=" remaining-tries)
	  (if server-dat
	      (let* ((iface     (tasks:hostinfo-get-interface server-dat))
		     (hostname  (tasks:hostinfo-get-hostname  server-dat))
		     (port      (tasks:hostinfo-get-port      server-dat))
		     (start-res (case *transport-type*
				  ((http)(http-transport:client-connect iface port))
				  ((nmsg)(nmsg-transport:client-connect hostname port))))
		     (ping-res  (case *transport-type* 
				  ((http)(rmt:login-no-auto-client-setup start-res run-id))
				  ((nmsg)(let ((logininfo (rmt:login-no-auto-client-setup start-res run-id)))
 					   (if logininfo
 					       (car (vector-ref logininfo 1))
 					       #f))))))
		(if (and start-res
			 ping-res)
		    (begin
		      (hash-table-set! *runremote* run-id start-res)
		      (debug:print-info 2 "connected to " (http-transport:server-dat-make-url start-res))
		      start-res)
		    (begin    ;; login failed but have a server record, clean out the record and try again
		      (debug:print-info 0 "client:setup, login failed, will attempt to start server ... start-res=" start-res ", run-id=" run-id ", server-dat=" server-dat)
		      (case *transport-type* 
			((http)(http-transport:close-connections run-id)))
		      (hash-table-delete! *runremote* run-id)
		      (tasks:server-force-clean-run-record (db:delay-if-busy tdbdat)
							   run-id 
							   (tasks:hostinfo-get-interface server-dat)
							   (tasks:hostinfo-get-port      server-dat)
							   " client:setup (server-dat = #t)")

		      (server:try-running run-id)
		      (thread-sleep! 5) ;; give server a little time to start up
		      (client:setup run-id remaining-tries: (- remaining-tries 1)))))
	      (begin    ;; no server registered
		(let ((num-available (tasks:num-in-available-state (db:dbdat-get-db tdbdat) run-id)))
		  (debug:print-info 0 "client:setup, no server registered, remaining-tries=" remaining-tries " num-available=" num-available)

		  (if (< num-available 2)

		      (server:try-running run-id))
		  (thread-sleep! 5) ;; give server a little time to start up
		  (client:setup run-id remaining-tries: (- remaining-tries 1)))))))))

;; 	(let ((host-info (hash-table-ref/default *runremote* run-id #f)))
;; 	  (if host-info ;; this is a bit circular. the host-info *is* the start-res FIXME
;; 	      (let* ((iface     (http-transport:server-dat-get-iface host-info))
;; 		     (port      (http-transport:server-dat-get-port  host-info))
;; 		     (start-res (case *transport-type* 
;; 				  ((http)(http-transport:client-connect iface port))
;; 				  ((nmsg)(nmsg-transport:client-connect iface port)) ;; (http-transport:server-dat-get-socket host-info))
;; 				  (else #f)))
;; 		     (ping-res  (case *transport-type*
;; 				  ((http)(rmt:login-no-auto-client-setup start-res run-id))
;; 				  ((nmsg)(let ((logininfo (rmt:login-no-auto-client-setup start-res run-id)))
;; 					   (if logininfo
;; 					       (vector-ref (vector-ref logininfo 1) 1)
;; 					       #f)))
;; 				  (else #f))))
;; 		(if ping-res   ;; sucessful login?
;; 		    (begin
;; 		      (debug:print-info 2 "client:setup, ping is good using host-info=" host-info ", remaining-tries=" remaining-tries)
;; 		      start-res)  ;; return the server info
;; 		    ;; have host info but no ping. shutdown the current connection and try again
;; 		    (begin    ;; login failed
;; 		      (debug:print-info 1 "client:setup, ping is bad for start-res=" start-res " and *runremote*=" host-info)
;; 		      (case *transport-type*
;; 			((http)(http-transport:close-connections run-id)))
;; 		      (hash-table-delete! *runremote* run-id)
;; 		      (if (< remaining-tries 8)
;; 			  (thread-sleep! 5)
;; 			  (thread-sleep! 1))
;; 		      (client:setup run-id remaining-tries: (- remaining-tries 1)))))
;; 	      ;; YUK: rename server-dat here
;; 

;; keep this as a function to ease future 
(define (client:start run-id server-info)
  (http-transport:client-connect (tasks:hostinfo-get-interface server-info)
				 (tasks:hostinfo-get-port server-info)))

;; client:signal-handler

(define *inmemdb*             #f)
(define *task-db*             #f) ;; (vector db path-to-db)
(define *db-access-allowed*   #t) ;; flag to allow access
(define *db-access-mutex*     (make-mutex))

;; SERVER
(define *my-client-signature* #f)
(define *transport-type*    'nmsg)

(define *runremote*         (make-hash-table)) ;; if set up for server communication this will hold <host port>
(define *max-cache-size*    0)
(define *logged-in-clients* (make-hash-table))
(define *client-non-blocking-mode* #f)
(define *server-id*         #f)
(define *server-info*       #f)
(define *time-to-exit*      #f)

  (debug:print-info 2 "starting exit process, finalizing databases.")
  (rmt:print-db-stats)
  (let ((run-ids (hash-table-keys *db-local-sync*)))
    (if (and (not (null? run-ids))
	     (configf:lookup *configdat* "setup" "megatest-db"))
	(db:multi-db-sync run-ids 'new2old)))
  (if *dbstruct-db* (db:close-all *dbstruct-db*))
  (if *inmemdb*     (db:close-all *inmemdb*))
  (if (and *megatest-db*
	   (sqlite3:database? *megatest-db*))
      (begin
	(sqlite3:interrupt! *megatest-db*)
	(sqlite3:finalize! *megatest-db* #t)
	(set! *megatest-db* #f)))
  (if *task-db*     (let ((db (cdr *task-db*)))

	(if (or (not (number? mtime))
		(not (number? stime))
		(> mtime stime)
		force-sync)
	    (begin
	      (db:delay-if-busy rundb)
	      (db:delay-if-busy olddb)

	      (dbr:dbstruct-set-stime! dbstruct (current-milliseconds))
	      (let ((num-synced (db:sync-tables db:sync-tests-only inmem refdb rundb olddb)))
		(mutex-unlock! *http-mutex*)
		num-synced)
	      (begin
		(mutex-unlock! *http-mutex*)
		0))))))

(define (db:close-main dbstruct)

       res))))

;;======================================================================
;; QUEUE UP META, TEST STATUS AND STEPS REMOTE ACCESS
;;======================================================================

;; NOTE: Can remove the regex and base64 encoding for zmq
(define (db:obj->string obj #!key (transport 'http))
  (case transport
    ;; ((fs) obj)
    ((http fs)
     (string-substitute
      (regexp "=") "_"
      (base64:base64-encode 
       (z3:encode-buffer
	(with-output-to-string
	  (lambda ()(serialize obj)))))
      #t))
    ((zmq nmsg)(with-output-to-string (lambda ()(serialize obj))))
    (else obj)))

(define (db:string->obj msg #!key (transport 'http))
  (case transport
    ;; ((fs) msg)
    ((http fs)
     (if (string? msg)
	 (with-input-from-string 
	     (z3:decode-buffer
	      (base64:base64-decode
	       (string-substitute 
		(regexp "_") "=" msg #t)))
	   (lambda ()(deserialize)))
	 (begin
	   (debug:print 0 "ERROR: reception failed. Received " msg " but cannot translate it.")
	   #f))) ;; crude reply for when things go awry
    ((zmq nmsg)(with-input-from-string msg (lambda ()(deserialize))))
    (else msg)))

(define (db:test-set-status-state dbstruct run-id test-id status state msg)
  (let ((dbdat  (db:get-db dbstruct run-id)))
    (if (member state '("LAUNCHED" "REMOTEHOSTSTART"))
	(db:general-call dbdat 'set-test-start-time (list test-id)))
    (if msg

sudo netstat -lptu
sudo netstat -tulpn</code></pre>
</div></div>
</div>
</div>
</div>
<h1 id="_reference">Reference</h1>







<div class="sect1">
<h2 id="_the_testconfig_file">The testconfig File</h2>
<div class="sectionbody">
<div class="sect2">
<h3 id="_setup_section">Setup section</h3>
<div class="sect3">
<h4 id="_header">Header</h4>

<td class="icon">
<div class="title">Note</div>
</td>
<td class="content">There is a trailing space after the --</td>
</tr></table>
</div>
</div>
</div>
</div>
<div class="sect1">
<h2 id="_programming_api">Programming API</h2>
<div class="sectionbody">
<div class="paragraph"><p>These routines can be called from the megatest repl.</p></div>
<div class="tableblock">
<table rules="all"
width="70%"
frame="hsides"
cellspacing="0" cellpadding="4">
<caption class="title">Table 2. API Server Management Calls</caption>
<col width="14%" />
<col width="28%" />
<col width="28%" />
<col width="28%" />
<thead>
<tr>
<th align="center" valign="top">API Call                  </th>
<th align="left" valign="top"> Purpose comments   </th>
<th align="left" valign="top"> Returns                 </th>
<th align="left" valign="top"> Comments</th>
</tr>
</thead>
<tfoot>
<tr>
<td align="center" valign="top"><p class="table">(rmt:login        run-id)</p></td>
<td align="left" valign="top"><p class="table"><code>Verify the the version, testsuite area etc. are correct.</code></p></td>
<td align="left" valign="top"><p class="table"><code>#( #t "successful login" )</code></p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
</tr>
</tfoot>
<tbody>
<tr>
<td align="center" valign="top"><p class="table">(rmt:start-server run-id)</p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
<td align="left" valign="top"><p class="table"><code>#( success/fail   n/a )</code></p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
</tr>
<tr>
<td align="center" valign="top"><p class="table">(rmt:kill-server  run-id)</p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
<td align="left" valign="top"><p class="table"><code>#( success/fail   n/a )</code></p></td>
<td align="left" valign="top"><p class="table"><code>Works only if the server is still reachable</code></p></td>
</tr>
</tbody>
</table>
</div>
<div class="tableblock">
<table rules="all"
width="70%"
frame="hsides"
cellspacing="0" cellpadding="4">
<caption class="title">Table 3. API Keys Related Calls</caption>
<col width="14%" />
<col width="28%" />
<col width="28%" />
<col width="28%" />
<thead>
<tr>
<th align="center" valign="top">API Call                        </th>
<th align="left" valign="top"> Purpose comments   </th>
<th align="left" valign="top"> Returns                 </th>
<th align="left" valign="top"> Comments</th>
</tr>
</thead>
<tfoot>
<tr>
<td align="center" valign="top"><p class="table">(rmt:get-key-val-pairs run-id)</p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
<td align="left" valign="top"><p class="table"><code>#t=success/#f=fail</code></p></td>
<td align="left" valign="top"><p class="table"><code>Works only if the server is still reachable</code></p></td>
</tr>
</tfoot>
<tbody>
<tr>
<td align="center" valign="top"><p class="table">(rmt:get-keys run-id)</p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
<td align="left" valign="top"><p class="table"><code>( key1 key2 &#8230; )</code></p></td>
<td align="left" valign="top"><p class="table"><code></code></p></td>
</tr>
</tbody>
</table>
</div>
<div class="sect2">
<h3 id="_megatest_internals">Megatest Internals</h3>
<div class="imageblock graphviz">
<div class="content">
<img src="server.png" alt="server.png" />
</div>
</div>

</div>
</div>
</div>
<div id="footnotes"><hr /></div>
<div id="footer">
<div id="footer-text">
Version 1.0<br />
Last updated 2014-11-25 21:11:28 MST
</div>
</div>
</body>
</html>

Reference
=========







The testconfig File
-------------------

Setup section
~~~~~~~~~~~~~

Header

-----------------
[triggers]
COMPLETED/ xterm -e bash -s -- 
-----------------

NOTE: There is a trailing space after the --

Programming API
---------------

These routines can be called from the megatest repl. 

.API Server Management Calls
[width="70%",cols="^,2m,2m,2m",frame="topbot",options="header,footer"]
|======================
|API Call                  | Purpose comments   | Returns                 | Comments
|(rmt:start-server run-id) |                    | #( success/fail   n/a ) | 
|(rmt:kill-server  run-id) |                    | #( success/fail   n/a ) | Works only if the server is still reachable
|(rmt:login        run-id) | Verify the the version, testsuite area etc. are correct. | #( #t "successful login" ) |
|======================

.API Keys Related Calls
[width="70%",cols="^,2m,2m,2m",frame="topbot",options="header,footer"]
|======================
|API Call                        | Purpose comments   | Returns                 | Comments
|(rmt:get-keys run-id)           |                    | ( key1 key2 ... )       | 
| (rmt:get-key-val-pairs run-id) |                    | #t=success/#f=fail      | Works only if the server is still reachable
|======================


:numbered!:

    (if (vector? server-dat)
	(let ((api-dat (http-transport:server-dat-get-api-uri server-dat)))
	  (close-connection! api-dat)
	  #t)
	#f)))


(define (make-http-transport:server-dat)(make-vector 6))
(define (http-transport:server-dat-get-iface         vec)    (vector-ref  vec 0))
(define (http-transport:server-dat-get-port          vec)    (vector-ref  vec 1))
(define (http-transport:server-dat-get-api-uri       vec)    (vector-ref  vec 2))
(define (http-transport:server-dat-get-api-url       vec)    (vector-ref  vec 3))
(define (http-transport:server-dat-get-api-req       vec)    (vector-ref  vec 4))
(define (http-transport:server-dat-get-last-access   vec)    (vector-ref  vec 5))
(define (http-transport:server-dat-get-socket        vec)    (vector-ref  vec 6))

(define (http-transport:server-dat-make-url vec)
  (if (and (http-transport:server-dat-get-iface vec)
	   (http-transport:server-dat-get-port  vec))
      (conc "http://" 
	    (http-transport:server-dat-get-iface vec)
	    ":"

;; run http-transport:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (http-transport:keep-running server-id run-id)
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown
  ;; This thread waits for the server to come alive
  (debug:print-info 0 "Starting the sync-back, keep alive thread in server for run-id=" run-id)
  (let* ((tdbdat      (tasks:open-db))
	 (server-info (let loop ((start-time (current-seconds))
				 (changed    #t)
				 (last-sdat  "not this"))
                        (let ((sdat #f))
			  (thread-sleep! 0.01)
			  (debug:print-info 0 "Waiting for server alive signature")

      ;; (debug:print 11 "last-access=" last-access ", server-timeout=" server-timeout)
      ;;
      ;; no_traffic, no running tests, if server 0, no running servers
      ;;
      ;; (let ((wait-on-running (configf:lookup *configdat* "server" "wait-on-running"))) ;; wait on running tasks (if not true then exit on time out)
      ;;
      (if (and *server-run*

	       (> (+ last-access server-timeout)
		  (current-seconds)))





	  (begin
	    (debug:print-info 0 "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
	    ;;
	    ;; Consider implementing some smarts here to re-insert the record or kill self is
	    ;; the db indicates so
	    ;;
	    ;; (if (tasks:server-am-i-the-server? tdb run-id)

;; (if (args:get-arg "-host")
;;     (begin
;;       (set! *runremote* (string-split (args:get-arg "-host" ":")))
;;       (client:launch))
;;     (client:launch))

;; ease debugging by loading ~/.dashboardrc
(let ((debugcontrolf (conc (get-environment-variable "HOME") "/.dashboardrc")))
  (if (file-exists? debugcontrolf)
      (load debugcontrolf)))

(define *dbdir* (conc (configf:lookup *configdat* "setup" "linktree") "/.db"))
(define *dbstruct-local*  (make-dbr:dbstruct path:  *dbdir*
					     local: #t))
(define *db-file-path* (db:dbfile-path 0))

;; HACK ALERT: this is a hack, please fix.

;; Copyright 2006-2012, Matthew Welland.
;; 
;;  This program is made available under the GNU GPL version 2.0 or
;;  greater. See the accompanying file COPYING for details.
;; 
;;  This program is distributed WITHOUT ANY WARRANTY; without even the
;;  implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
;;  PURPOSE.

(require-extension (srfi 18) extras tcp s11n)

(use sqlite3 srfi-1 posix regex regex-case srfi-69 hostinfo md5 message-digest)
(import (prefix sqlite3 sqlite3:))

(use nanomsg)

(declare (unit nmsg-transport))

(declare (uses common))
(declare (uses db))
(declare (uses tests))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.
(declare (uses server))

(include "common_records.scm")
(include "db_records.scm")

;; Transition to pub --> sub with pull <-- push
;;
;;   1. client sends request to server via push to the pull port
;;   2. server puts request in queue or processes immediately as appropriate
;;   3. server puts responses from completed requests into pub port 
;;
;; TODO
;;
;; Done Tested
;; [x]  [ ]    1. Add columns pullport pubport to servers table
;; [x]  [ ]    2. Add rm of monitor.db if older than 11/12/2012 
;; [x]  [ ]    3. Add create of pullport and pubport with finding of available ports
;; [x]  [ ]    4. Add client compose of request
;; [x]  [ ]        - name of client: testname/itempath-test_id-hostname 
;; [x]  [ ]        - name of request: callname, params
;; [x]  [ ]        - request key: f(clientname, callname, params)
;; [x]  [ ]    5. Add processing of subscription hits
;; [x]  [ ]        - done when get key 
;; [x]  [ ]        - return results
;; [x]  [ ]    6. Add timeout processing
;; [x]  [ ]        - after 60 seconds
;; [ ]  [ ]            i. check server alive, connect to new if necessary
;; [ ]  [ ]           ii. resend request
;; [ ]  [ ]    7. Turn self ping back on

(define (nmsg-transport:make-server-url hostport #!key (bindall #f))
  (if (not hostport)
      #f
      (conc "tcp://" (if bindall "*" (car hostport)) ":" (cadr hostport))))

(define *server-loop-heart-beat* (current-seconds))
(define *heartbeat-mutex* (make-mutex))

;;======================================================================
;; S E R V E R
;;======================================================================

(define (nmsg-transport:run dbstruct hostn run-id server-id #!key (retrynum 1000))
  (debug:print 2 "Attempting to start the server ...")
  (let* ((start-port      (portlogger:open-run-close portlogger:find-port))
	 (server-thread   (make-thread (lambda ()
					 (nmsg-transport:try-start-server dbstruct run-id start-port server-id))
				       "server thread"))
	 (tdbdat          (tasks:open-db)))
    (thread-start! server-thread)
    (thread-sleep! 0.1)
    (if (nmsg-transport:ping hostn start-port timeout: 2 expected-key: (current-process-id))
	(let ((interface (if (equal? hostn "-")(get-host-name) hostn)))
	  (tasks:server-set-interface-port (db:delay-if-busy tdbdat) server-id interface start-port)
	  (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "dbprep")
	  (set! *server-info* (list hostn start-port)) ;; probably not needed anymore? currently used by keep-running
	  (thread-sleep! 3) ;; give some margin for queries to complete before switching from file based access to server based access
	  ;; (set! *inmemdb*  dbstruct)
	  (tasks:server-set-state! (db:delay-if-busy tdbdat) server-id "running")
	  (thread-start! (make-thread
			  (lambda ()(nmsg-transport:keep-running server-id run-id))
			  "keep running"))
	  (thread-join! server-thread))
	(if (> retrynum 0)
	    (begin
	      (debug:print 0 "WARNING: Failed to connect to server (self) on host " hostn ":" start-port ", trying again.")
	      (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id "failed to start, never received server alive signature")
	      (portlogger:open-run-close portlogger:set-failed start-port)
	      (nmsg-transport:run dbstruct hostn run-id server-id))
	    (begin
	      (debug:print 0 "ERROR: could not find an open port to start server on. Giving up")
	      (exit 1))))))

(define (nmsg-transport:try-start-server dbstruct run-id portnum server-id)
  (let ((repsoc (nn-socket 'rep)))
    (nn-bind repsoc (conc "tcp://*:" portnum))
    (let loop ((msg-in (nn-recv repsoc)))
      (let* ((dat    (db:string->obj msg-in transport: 'nmsg)))
	(debug:print 0 "server, received: " dat)
	(let ((result (api:execute-requests dbstruct dat)))
	  (debug:print 0 "server, sending: " result)
	  (nn-send repsoc (db:obj->string result  transport: 'nmsg)))
	(loop (nn-recv repsoc))))))

;; all routes though here end in exit ...
;;
(define (nmsg-transport:launch run-id)
  (let* ((tdbdat   (tasks:open-db))
	 (dbstruct (db:setup run-id))
	 (hostn    (or (args:get-arg "-server") "-")))
    (set! *run-id*   run-id)
    (set! *inmemdb* dbstruct)
    ;; with nbfake daemonize isn't really needed
    ;;
    ;; (if (args:get-arg "-daemonize")
    ;;     (begin
    ;;       (daemon:ize)
    ;;       (if *alt-log-file* ;; we should re-connect to this port, I think daemon:ize disrupts it
    ;;           (begin
    ;;     	(current-error-port *alt-log-file*)
    ;;     	(current-output-port *alt-log-file*)))))
    (if (server:check-if-running run-id)
	(begin
	  (debug:print-info 0 "Server for run-id " run-id " already running")
	  (exit 0)))
    (let loop ((server-id (tasks:server-lock-slot (db:delay-if-busy tdbdat) run-id))
	       (remtries  4))
      (if (not server-id)
	  (if (> remtries 0)
	      (begin
		(thread-sleep! 2)
		(if (not (server:check-if-running run-id))
		    (loop (tasks:server-lock-slot (db:delay-if-busy tdbdat) run-id)
			  (- remtries 1))
		    (begin
		      (debug:print-info 0 "Another server took the slot, exiting")
		      (exit 0))))
	      (begin
		;; since we didn't get the server lock we are going to clean up and bail out
		(debug:print-info 2 "INFO: server pid=" (current-process-id) ", hostname=" (get-host-name) " not starting due to other candidates ahead in start queue")
		(tasks:server-delete-records-for-this-pid (db:delay-if-busy tdbdat) " http-transport:launch")
		))
	  ;; locked in a server id, try to start up
	  (nmsg-transport:run dbstruct hostn run-id server-id))
      (set! *didsomething* #t)
      (exit))))

;;======================================================================
;; S E R V E R   U T I L I T I E S 
;;======================================================================

(define (nmsg-transport:mk-signature)
  (message-digest-string (md5-primitive) 
			 (with-output-to-string
			   (lambda ()
			     (write (list (current-directory)
					  (argv)))))))

;;======================================================================
;; C L I E N T S
;;======================================================================

;; ping the server at host:port
;;   return the open socket if successful (return-socket == #t)
;;   expect the key expected-key returned in payload
;;   send our-key or #f as payload
;;
(define (nmsg-transport:ping hostn port #!key (timeout 3)(return-socket #t)(expected-key #f)(our-key #f)(socket #f))
  ;; send a random number along with pid and check that we get it back
  (let* ((host    (if (or (not hostn)
			  (equal? hostn "-")) ;; use localhost
		      (get-host-name)
		      hostn))
	 (req     (or socket
		      (let ((soc (nn-socket 'req)))
			(nn-connect soc (conc "tcp://" host ":" port))
			soc)))
	 (success #t)
	 (dat     (vector "ping" our-key))
	 (result  (condition-case 
		   (nmsg-transport:client-api-send-receive-raw req dat timeout: timeout)
		   ((timeout)(set! success #f) #f)))
	 (key     (if success 
		      (vector-ref result 1)
		      #f)))
    (debug:print 0 "success=" success ", key=" key ", expected-key=" expected-key ", equal? " (equal? key expected-key))
    (if (and success
	     (or (not expected-key) ;; just getting a reply is good enough then
		 (equal? key expected-key)))
	(if return-socket
	    req
	    (begin
	      (if (not socket)(nn-close req)) ;; don't want a side effect of closing socket if handed it
	      #t))
	(begin
	  (if (not socket)(nn-close req)) ;; failed to ping, close socket as side effect
	  #f))))

;; send data to server, wait max of timeout seconds for a response.
;; return #( success/fail result )
;;
;; for effiency it is easier to do the obj->string and string->obj here.
;;
(define (nmsg-transport:client-api-send-receive-raw socreq indat #!key (enable-send #t)(timeout 25))
  (let* ((success     #f)
	 (result      #f)
	 (keepwaiting #t)
	 (dat         (db:obj->string indat transport: 'nmsg))
	 (send-recv   (make-thread
		       (lambda ()
			 (nn-send socreq dat)
			 (let* ((res (nn-recv socreq)))
			   (set! success #t)
			   (set! result (db:string->obj res transport: 'nmsg))))
		       "send-recv"))
	 (timeout     (make-thread
		       (lambda ()
			 (let loop ((count 0))
			   (thread-sleep! 1)
			   (debug:print-info 1 "send-receive-raw, still waiting after " count " seconds...")
			   (if (and keepwaiting (< count timeout)) ;; yes, this is very aproximate
			       (loop (+ count 1))))
			 (if keepwaiting
			     (begin
			       (print "timeout waiting for ping")
			       (thread-terminate! send-recv))))
		       "timeout")))
    ;; replace with condition-case?
    (handle-exceptions
     exn
     (set! result "timeout")
     (thread-start! timeout)
     (thread-start! send-recv)
     (thread-join! send-recv)
     (if success (thread-terminate! timeout)))
    ;; raise timeout error if timed out
    (if success
	(if (and (vector? result)
		 (vector-ref result 0)) ;; did it fail at the server?
	    result                ;; nope, all good
	    (begin
	      (debug:print 0 "ERROR: error occured at server, info=" (vector-ref result 2))
	      (debug:print 0 " client call chain:")
	      (print-call-chain (current-error-port))
	      (debug:print 0 " server call chain:")
	      (pp (vector-ref result 1) (current-error-port))
	      (signal (vector-ref result 0))))
	(signal (make-composite-condition
		 (make-property-condition 'timeout 'message "nmsg-transport:client-api-send-receive-raw timed out talking to server"))))))

;; run nmsg-transport:keep-running in a parallel thread to monitor that the db is being 
;; used and to shutdown after sometime if it is not.
;;
(define (nmsg-transport:keep-running server-id run-id)
  ;; if none running or if > 20 seconds since 
  ;; server last used then start shutdown
  ;; This thread waits for the server to come alive
  (let* ((server-info (let loop ()
                        (let ((sdat #f))
                          (mutex-lock! *heartbeat-mutex*)
                          (set! sdat *server-info*)
                          (mutex-unlock! *heartbeat-mutex*)
                          (if sdat 
			      (begin
				(debug:print-info 0 "keep-running got sdat=" sdat)
				sdat)
                              (begin
                                (thread-sleep! 0.5)
                                (loop))))))
         (iface       (car server-info))
         (port        (cadr server-info))
         (last-access 0)
	 (tdbdat      (tasks:open-db))
	 (server-timeout (let ((tmo (configf:lookup  *configdat* "server" "timeout")))
			   (if (and (string? tmo)
				    (string->number tmo))
			       (* 60 60 (string->number tmo))
			       ;; (* 3 24 60 60) ;; default to three days
			       (* 60 1)         ;; default to one minute
			       ;; (* 60 60 25)      ;; default to 25 hours
			       ))))
    (print "Keep-running got server pid " server-id ", using iface " iface " and port " port)
    (let loop ((count 0))
      (thread-sleep! 4) ;; no need to do this very often
      ;; NB// sync currently does NOT return queue-length
      (let () ;; (queue-len (cdb:client-call server-info 'sync #t 1)))
      ;; (print "Server running, count is " count)
        (if (< count 1) ;; 3x3 = 9 secs aprox
            (loop (+ count 1)))
        
        (mutex-lock! *heartbeat-mutex*)
        (set! last-access *last-db-access*)
        (mutex-unlock! *heartbeat-mutex*)
	(db:sync-touched *inmemdb* run-id force-sync: #t)
        (if (and *server-run*
	       (> (+ last-access server-timeout)
		  (current-seconds)))
            (begin
              (debug:print-info 0 "Server continuing, seconds since last db access: " (- (current-seconds) last-access))
              (loop 0))
            (begin
              (debug:print-info 0 "Starting to shutdown the server.")
              (set! *time-to-exit* #t)
	      (db:sync-touched *inmemdb* run-id force-sync: #t)
              (tasks:server-delete-record (db:delay-if-busy tdbdat) server-id " http-transport:keep-running")
              (debug:print-info 0 "Server shutdown complete. Exiting")
              (exit)
	      ))))))

;;======================================================================
;; C L I E N T S
;;======================================================================

(define (nmsg-transport:client-connect iface portnum)
  (let* ((reqsoc      (nmsg-transport:ping iface portnum return-socket: #t)))
    (vector iface portnum #f #f #f (current-seconds) reqsoc)))

;; return #( success result )
;;
(define (nmsg-transport:client-api-send-receive run-id connection-info cmd param #!key (remtries 5))
  (mutex-lock! *http-mutex*)
  (let* ((packet  (vector cmd param))
	 (reqsoc  (http-transport:server-dat-get-socket connection-info))
	 (res     (nmsg-transport:client-api-send-receive-raw reqsoc packet)))
;;	 (status  (vector-ref rawres 0))
;;	 (result  (vector-ref rawres 1)))
    (mutex-unlock! *http-mutex*)
    res)) ;; (vector status (if status (db:string->obj result transport: 'nmsg) result))))
	
;;======================================================================
;; J U N K 
;;======================================================================

;; DO NOT USE
;;
(define (nmsg-transport:client-signal-handler signum)
  (handle-exceptions
   exn
   (debug:print " ... exiting ...")
   (let ((th1 (make-thread (lambda ()
			     (if (not *received-response*)
				 (receive-message* *runremote*))) ;; flush out last call if applicable
			   "eat response"))
	 (th2 (make-thread (lambda ()
			     (debug:print 0 "ERROR: Received ^C, attempting clean exit. Please be patient and wait a few seconds before hitting ^C again.")
			     (thread-sleep! 3) ;; give the flush three seconds to do it's stuff
			     (debug:print 0 "       Done.")
			     (exit 4))
			   "exit on ^C timer")))
     (thread-start! th2)
     (thread-start! th1)
     (thread-join! th2))))

(use json format)

(declare (unit rmt))
(declare (uses api))
(declare (uses tdb))
(declare (uses http-transport))
(declare (uses nmsg-transport))

;;
;; THESE ARE ALL CALLED ON THE CLIENT SIDE!!!
;;

;; ;; For debugging add the following to ~/.megatestrc
;;

	cinfo
	;; NB// can cache the answer for server running for 10 seconds ...
	;;  ;; (and (not (rmt:write-frequency-over-limit? cmd run-id))
	(if (tasks:server-running-or-starting? (db:delay-if-busy (tasks:open-db)) run-id)
	    (client:setup run-id)
	    #f))))




(define (rmt:send-receive cmd rid params #!key (attemptnum 1)) ;; start attemptnum at 1 so the modulo below works as expected
  ;; clean out old connections
  (mutex-lock! *db-multi-sync-mutex*)
  ;; (let ((expire-time (- (current-seconds) 60)))
  ;;   (for-each 
  ;;    (lambda (run-id)
  ;;      (let ((connection (hash-table-ref/default *runremote* run-id #f)))
  ;;        (if (and connection 
  ;;       	  (< (http-transport:server-dat-get-last-access connection) expire-time))
  ;;            (begin
  ;;              (debug:print-info 0 "Discarding connection to server for run-id " run-id ", too long between accesses")
  ;;              ;; SHOULD CLOSE THE CONNECTION HERE
  ;;              (hash-table-delete! *runremote* run-id)))))
  ;;    (hash-table-keys *runremote*)))
  (mutex-unlock! *db-multi-sync-mutex*)
  (let* ((run-id          (if rid rid 0))
	 (connection-info (rmt:get-connection-info run-id)))

    ;; the nmsg method does the encoding under the hood (the http method should be changed to do this also)
    (if connection-info
	;; use the server if have connection info
	(let* ((dat     (case *transport-type*
			  ((http)(http-transport:client-api-send-receive run-id connection-info cmd params))
			  ((nmsg)(condition-case
				  (nmsg-transport:client-api-send-receive run-id connection-info cmd params)
				  ((timeout)(vector #f "timeout talking to server"))))
			  (else  (exit))))
	       (success (if (and dat (vector? dat)) (vector-ref dat 0) #f))
	       (res     (if (and dat (vector? dat)) (vector-ref dat 1) #f)))
	  (http-transport:server-dat-update-last-access connection-info)
	  (if success
	      (case *transport-type* 
		((http) res) ;; (db:string->obj res))
		((nmsg) res)) ;; (vector-ref res 1)))








	      (begin ;; let ((new-connection-info (client:setup run-id)))
		(debug:print 0 "WARNING: Communication failed, trying call to rmt:send-receive again.")
		;; (case *transport-type*
		;;   ((nmsg)(nn-close (http-transport:server-dat-get-socket connection-info))))
		(hash-table-delete! *runremote* run-id) ;; don't keep using the same connection
		(if (eq? (modulo attemptnum 5) 0)
		    (tasks:kill-server-run-id run-id tag: "api-send-receive-failed"))
		(tasks:start-and-wait-for-server (tasks:open-db) run-id 15)
		;; (nmsg-transport:client-api-send-receive run-id connection-info cmd param remtries: (- remtries 1))))))

		;; no longer killing the server in http-transport:client-api-send-receive
		;; may kill it here but what are the criteria?
		;; start with three calls then kill server
		;; (if (eq? attemptnum 3)(tasks:kill-server-run-id run-id))
		;; (thread-sleep! 2)
		(rmt:send-receive cmd run-id params attemptnum: (+ attemptnum 1)))))
	;; no connection info? try to start a server
	(if (and (< attemptnum 15)
		 (tasks:need-server run-id))
	    (begin
	      (hash-table-delete! *runremote* run-id)
	      (tasks:start-and-wait-for-server (db:delay-if-busy (tasks:open-db)) run-id 10)
	      (client:setup run-id)
	      (thread-sleep! (random 5)) ;; give some time to settle and minimize collison?
	      (rmt:send-receive cmd rid params attemptnum: (+ attemptnum 1)))
	    (begin
	      (debug:print 0 "ERROR: Communication failed!")
	      (exit)
	      ;; (rmt:open-qry-close-locally cmd run-id params))))
	      )))))

(define (rmt:update-db-stats run-id rawcmd params duration)
  (mutex-lock! *db-stats-mutex*)
  (handle-exceptions
   exn
   (begin
     (debug:print 0 "WARNING: stats collection failed in update-db-stats")

			     (let* ((dbdir (conc    (configf:lookup *configdat* "setup" "linktree") "/.db"))
				    (db (make-dbr:dbstruct path:  dbdir local: #t)))
			       (set! *dbstruct-db* db)
			       db)))
	 (db-file-path   (db:dbfile-path 0)))
    ;; (read-only      (not (file-read-access? db-file-path)))
    (let* ((start         (current-milliseconds))
	   (resdat        (api:execute-requests dbstruct-local (symbol->string cmd) params))
	   (res           (vector-ref resdat 1))
	   (duration      (- (current-milliseconds) start)))
      (rmt:update-db-stats run-id cmd params duration)
      ;; mark this run as dirty if this was a write
      (if (not (member cmd api:read-only-queries))
	  (let ((start-time (current-seconds)))
	    (mutex-lock! *db-multi-sync-mutex*)
	    ;; (if (not (hash-table-ref/default *db-local-sync* run-id #f))

;;  M I S C
;;======================================================================

(define (rmt:login run-id)
  (rmt:send-receive 'login run-id (list *toppath* megatest-version run-id *my-client-signature*)))

;; This login does no retries under the hood - it acts a bit like a ping.
;; Deprecated for nmsg-transport.
;;
(define (rmt:login-no-auto-client-setup connection-info run-id)
  (case *transport-type*
    ((http)(rmt:send-receive-no-auto-client-setup connection-info 'login run-id (list *toppath* megatest-version run-id *my-client-signature*)))
    ((nmsg)(nmsg-transport:client-api-send-receive run-id connection-info 'login (list *toppath* megatest-version run-id *my-client-signature*)))))

;; hand off a call to one of the db:queries statements
;; added run-id to make looking up the correct db possible 
;;
(define (rmt:general-call stmtname run-id . params)
  (rmt:send-receive 'general-call run-id (append (list stmtname run-id) params)))

(define (rmt:sync-inmem->db run-id)

	     (newtal      (append tal (list hed)))
	     (regfull     (>= (length reg) reglen))
	     (num-running (rmt:get-count-tests-running-for-run-id run-id)))

	;; every couple minutes verify the server is there for this run
	(if (and (common:low-noise-print 60 "try start server"  run-id)
		 (tasks:need-server run-id))
	    (tasks:start-and-wait-for-server tdbdat run-id 10)) ;; NOTE: delay and wait is done under the hood
	
	(if (> num-running 0)
	  (set! last-time-some-running (current-seconds)))

      (if (> (current-seconds)(+ last-time-some-running 240))
	  (hash-table-set! *max-tries-hash* tfullname (+ (hash-table-ref/default *max-tries-hash* tfullname 0) 1)))
	;; (debug:print 0 "max-tries-hash: " (hash-table->alist *max-tries-hash*))

(declare (unit server))

(declare (uses common))
(declare (uses db))
(declare (uses tasks)) ;; tasks are where stuff is maintained about what is running.
(declare (uses synchash))
(declare (uses http-transport))
(declare (uses nmsg-transport))
(declare (uses launch))
;; (declare (uses zmq-transport))
(declare (uses daemon))

(include "common_records.scm")
(include "db_records.scm")

;;

;; all routes though here end in exit ...
;;
;; start_server
;;
(define (server:launch run-id)
  (case *transport-type*
    ((http)(http-transport:launch run-id))
    ((nmsg)(nmsg-transport:launch run-id))
    (else (debug:print 0 "ERROR: unknown server type " *transport-type*))))

;;======================================================================
;; Q U E U E   M A N A G E M E N T
;;======================================================================

;; We don't want to flush the queue if it was just flushed
(define *server:last-write-flush* (current-milliseconds))

	       (trycount 0))
    (if server
	;; note: client:start will set *runremote*. this needs to be changed
	;;       also, client:start will login to the server, also need to change that.
	;;
	;; client:start returns #t if login was successful.
	;;
	(let ((res (case *transport-type*
		     ((http)(server:ping-server run-id 
						(tasks:hostinfo-get-interface server)
						(tasks:hostinfo-get-port      server)))
		     ((nmsg)(nmsg-transport:ping (tasks:hostinfo-get-interface server)
						 (tasks:hostinfo-get-port      server)
						 timeout: 2)))))
	  ;; if the server didn't respond we must remove the record
	  (if res
	      #t
	      (begin
		(debug:print-info 0 "server at " server " not responding, removing record")
		(tasks:server-force-clean-running-records-for-run-id (db:delay-if-busy tdbdat) run-id 
				" server:check-if-running")

   (get-host-name)            ;; hostname
   -1                         ;; port
   -1                         ;; pubport
   (random 1000)              ;; priority (used a tiebreaker on get-available)
   "available"                ;; state
   (common:version-signature) ;; mt_version
   -1                         ;; interface
   (conc *transport-type*)    ;; transport
   run-id
   ))

(define (tasks:num-in-available-state mdb run-id)
  (let ((res 0))
    (sqlite3:for-each-row
     (lambda (num-in-queue)

(use nanomsg srfi-18 sqlite3 numbers)

(define resp (nn-socket 'rep))

(use zmq posix numbers)

(define cname "Bob")
(define runtime 10)
(let ((args (argv)))
  (if (< (length args) 3)
      (begin
	(print "Usage: mockupclient clientname runtime")
	(exit))
      (begin
	(set! cname (cadr args))
	(set! runtime (string->number (caddr args))))))
      
;; (define start-delay (/ (random 100) 9))
;; (define runtime     (+ 1 (/ (random 200) 2)))

(print "Starting client " cname " with runtime " runtime)

(include "mockupclientlib.scm")

(set! endtime (+ (current-seconds) runtime))

;; first ping the server to ensure we have a connection
(if (server-ping cname 5)
    (print "SUCCESS: Client " cname " connected to server")
    (begin
      (print "ERROR: Client " cname " failed ping of server, exiting")
      (exit)))

(let loop ()
  (let ((x (random 15))
	(varname (list-ref (list "hello" "goodbye" "saluton" "kiaorana")(random 4))))
    (case x
      ;; ((1)(dbaccess cname 'sync "nodat"    #f))
      ((2 3 4 5)(dbaccess cname 'set varname (random 999)))
      ((6 7 8 9 10)(print cname ": Get \"" varname "\" " (dbaccess cname 'get varname #f)))
      (else
       (thread-sleep! 0.011)))
    (if (< (current-seconds) endtime)
	(loop))))

(print "Client " cname " all done!!")

(define reqs (nn-socket 'req))

(connect-socket reqs "tcp://localhost:6563")

(thread-sleep! 0.2)

(define (server-ping cname timeout)
  (let ((msg     (conc cname ":ping:" timeout))
	(maxtime (+ (current-seconds) timeout)))
    (print "pinging server from " cname " with timeout " timeout)
    (let loop ((res   #f))
      (if (< maxtime (current-seconds))
	  #f ;; failed to ping
	  (if (equal? res "Got ping")
	      #t
	      (begin
		(print "Ping received from server " res)
		(send-message push msg)
		(thread-sleep! 0.1)
		(loop (receive-message sub non-blocking: #t))))))))
  
(define (dbaccess cname cmd var val #!key (numtries 20))
  (let* ((msg (conc cname ":" cmd ":" (if val (conc var " " val) var)))
	 (res #f)
	 (mtx1 (make-mutex))
	 (do-access (lambda ()
		      (let ((tmpres #f))
			(print "Sending msg: " msg)
			(send-message push msg)
			(print "Message " msg " sent")
			(print "Client " cname " waiting for response to " msg)
			(print "Client " cname " received address " (receive-message* sub))
			(set! tmpres (receive-message* sub))
			(mutex-lock! mtx1)
			(set! res tmpres)
			(mutex-unlock! mtx1))))
	 (th1 (make-thread do-access "do access"))
	 (th2 (make-thread (lambda ()
			     (let ((result #f))
			       (mutex-lock! mtx1)
			       (set! result res)
			       (mutex-unlock! mtx1)
			       (thread-sleep! 5)
			       (if (not result)
				   (if (> numtries 0)
				       (begin
					 (print "WARNING: access timed out for " cname ", trying again. Trys remaining=" numtries)
					 (dbaccess cname cmd var val numtries: (- numtries 1)))
				       (begin
					 (print "ERROR: dbaccess timed out. Exiting")
					 (exit)))))
			     "timeout thread"))))
    (thread-start! th1)
    (thread-start! th2)
    (thread-join! th1)
    (if res (print "SUCCESS: received " res " with " numtries " remaining possible attempts"))
    res))

;; pub/sub with envelope address
;; Note that if you don't insert a sleep, the server will crash with SIGPIPE as soon
;; as a client disconnects.  Also a remaining client may receive tons of
;; messages afterward.

(use nanomsg srfi-18 sqlite3 numbers)

(define resp (nn-socket 'rep))
(define cname "server")
(define total-db-accesses 0)
(define start-time (current-seconds))

(nn-bind resp  "tcp://*:6563")

(thread-sleep! 0.2)

(define (open-db)
  (let* ((dbpath    "mockup.db")
	 (dbexists  (file-exists? dbpath))
	 (db        (open-database dbpath)) ;; (never-give-up-open-db dbpath))
	 (handler   (make-busy-timeout 10)))
    (set-busy-handler! db handler)
    (if (not dbexists)
	(for-each
	 (lambda (stmt)
	   (execute db stmt))
	 (list
	  "PRAGMA SYNCHRONOUS=0;"
	  "CREATE TABLE clients (id INTEGER PRIMARY KEY,name TEXT,num_accesses INTEGER DEFAULT 0);"
	  "CREATE TABLE vars    (var TEXT,val TEXT,CONSTRAINT vars_constraint UNIQUE (var));")))
    db))

(define cid-cache (make-hash-table))

(define (get-client-id db cname)
  (let ((cid (hash-table-ref/default cid-cache cname #f)))
    (if cid 
	cid
	(begin
	  (execute db "INSERT OR REPLACE INTO clients (name) VALUES(?);" cname)
	  (for-each-row 
	   (lambda (id)
	     (set! cid id))
	   db
	   "SELECT id FROM clients WHERE name=?;" cname)
	  (hash-table-set! cid-cache cname cid)
	  (set! total-db-accesses (+ total-db-accesses 2))
	  cid))))

(define (count-client db cname)
  (let ((cid (get-client-id db cname)))
    (execute db "UPDATE clients SET num_accesses=num_accesses+1 WHERE id=?;" cid)
    (set! total-db-accesses (+ total-db-accesses 1))
    ))

(define db (open-db))
;; (define queuelst '())
;; (define mx1 (make-mutex))

(define max-queue-len 0)

(define (process-queue queuelst)
  (let ((queuelen (length queuelst)))
    (if (> queuelen max-queue-len)
	(set! max-queue-len queuelen))
    (for-each
     (lambda (item)
       (let ((cname (vector-ref item 1))
	     (clcmd (vector-ref item 2))
	     (cdata (vector-ref item 3)))
	 (send-message pub cname send-more: #t)
	 (send-message pub (case clcmd
			     ((sync)
			      (conc queuelen))
			     ((set)
			      (set! total-db-accesses (+ total-db-accesses 1))
			      (apply execute db "INSERT OR REPLACE INTO vars (var,val) VALUES (?,?);" (string-split cdata))
			      "ok")
			     ((get)
			      (set! total-db-accesses (+ total-db-accesses 1))
			      (let ((res "noval"))
				(for-each-row
				 (lambda (val)
				   (set! res val))
				 db 
				 "SELECT val FROM vars WHERE var=?;" cdata)
				res))
			     (else (conc "unk cmd: " clcmd))))))
     queuelst)))

;; SERVER THREAD
(define th1 (make-thread 
	     (lambda ()
	       (let ((last-run 0)) ;; current-seconds when run last
		 (let loop ((queuelst '()))
		   (let* ((indat (receive-message* pull))
			  (parts (string-split indat ":"))
			  (cname (car parts))                   ;; client name
			  (clcmd (string->symbol (cadr parts))) ;; client cmd
			  (cdata (caddr parts))                 ;; client data
			  (svect (vector (current-seconds) cname clcmd cdata))) ;; record for the queue
		     ;; (print "Server received message: " indat)
		     (count-client db cname)
		     (case clcmd
		       ((ping)
			(print "Got ping from " cname)
			(send-message pub cname send-more: #t)
			(send-message pub "Got ping")
			(loop queuelst))
		       ((sync) ;; just process the queue
			(print "Got sync from " cname)
			(process-queue (cons svect queuelst))
			(loop '()))
		       ((get)
			(process-queue (cons svect queuelst))
			(loop '()))
		       (else
			(loop (cons svect queuelst))))))))
	     "server thread"))

(include "mockupclientlib.scm")

;; SYNC THREAD
;; send a sync to the pull port
(define th2 (make-thread
	     (lambda ()
	       (let ((last-action-time (current-seconds)))
		 (let loop ()
		   (thread-sleep! 5)
		   (let ((queuelen (string->number (dbaccess "server" 'sync "nada" #f)))
			 (last-action-delta #f))
		     (if (> queuelen 1)(set! last-action-time (current-seconds)))
		     (set! last-action-delta (- (current-seconds) last-action-time))
		     (print "Server: Got queuelen=" queuelen ", last-action-delta=" last-action-delta)
		     (if (< last-action-delta 60)
			 (loop)
			 (print "Server exiting, 25 seconds since last access"))))))
	     "sync thread"))

(thread-start! th1)
(thread-start! th2)
(thread-join! th2)

(let* ((run-time       (- (current-seconds) start-time))
       (queries/second (/  total-db-accesses run-time)))
  (print "Server exited! Total db accesses=" total-db-accesses " in " run-time " seconds for " queries/second " queries/second with max queue length of: " max-queue-len))

;; watch nanomsg's pipeline load-balancer in action.
(use nanomsg)

(define push (nn-socket 'push))
(define pull1 (nn-socket 'pull))
(define pull2 (nn-socket 'pull))

(nn-bind    push  "inproc://test")
(nn-connect pull1 "inproc://test")
(nn-connect pull2 "inproc://test")

(nn-send push "a")
(nn-send push "b")
(nn-send push "c")
(nn-send push "d")

(define ((th sock))
  (print (current-thread) ": " (nn-recv sock))
  (print (current-thread) ": " (nn-recv sock))
  (print (current-thread) " is done"))

(thread-start! (th pull1))
(thread-start! (th pull2))

(thread-sleep! 1)

;; watch nanomsg's pipeline load-balancer in action.
(use nanomsg)

(define req   (nn-socket 'req))

(nn-connect req  "tcp://localhost:22022")

;; (with-output-to-string (lambda ()(serialize obj)))
(define (client-send-receive soc msg)
  (nn-send soc msg)
  (nn-recv soc))

(define ((talk-to-server soc))
  (let loop ((cnt 20))
    (let ((name (list-ref '("Matt" "Tom" "Bob" "Jill" "James" "Jane")(random 6))))
      (print "Sending " name)
      (print (client-send-receive req name))
      (if (> cnt 0)(loop (- cnt 1)))))
  (print (client-send-receive req "quit"))
  (nn-close req)
  (exit))

;; (thread-start! (lambda ()
;; 		 (thread-sleep! 20)
;; 		 (print "Give up on waiting for the server")
;; 		 (nn-close req)
;; 		 (exit)))

(thread-join! (thread-start! (talk-to-server req)))

;; watch nanomsg's pipeline load-balancer in action.
(use nanomsg)

;; (use trace)
;; (trace nn-bind nn-socket nn-assert nn-recv nn-send thread-terminate! nn-close )

(define port  22022)
(define host  "127.0.0.1")

(define rep   (nn-socket 'rep))

(print "connecting, got: " (nn-bind    rep  (conc "tcp://" "*" ":" port)))

(define (server soc)
  (print "server starting")
  (let loop ((msg-in (nn-recv soc)))
    (print "server received: " msg-in)
    (cond
     ((equal? msg-in "quit")
      (nn-send soc "Ok, quitting"))
     ((and (>= (string-length msg-in) 4)
	   (equal? (substring msg-in 0 4) "ping"))
      (nn-send soc (conc (current-process-id)))
      (loop (nn-recv soc)))
     ;;((and (>= (string-length msg-in)
     (else
      (let ((this-task (random 15)))
	(thread-sleep! this-task)
	(nn-send soc (conc "hello " msg-in " this task took " this-task " seconds to complete"))
	(loop (nn-recv soc)))))))

(define (ping-self host port #!key (return-socket #t))
  ;; send a random number along with pid and check that we get it back
  (let* ((req     (nn-socket 'req))
	 (key     "ping")
	 (success #f)
	 (keepwaiting #t)
	 (ping    (make-thread
		   (lambda ()
		     (print "ping: sending string \"" key "\", expecting " (current-process-id))
		     (nn-send req key)
		     (let ((result  (nn-recv req)))
		       (if (equal? (conc (current-process-id)) result)
			   (begin
			     (print "ping, success: received \"" result "\"")
			     (set! success #t))
			   (begin
			     (print "ping, failed: received key \"" result "\"")
			     (set! keepwaiting #f)
			     (set! success #f)))))
		   "ping"))
	 (timeout (make-thread (lambda ()
				 (let loop ((count 0))
				   (thread-sleep! 1)
				   (print "still waiting after count seconds...")
				   (if (and keepwaiting (< count 10))
				       (loop (+ count 1))))
				 (if keepwaiting
				     (begin
				       (print "timeout waiting for ping")
				       (thread-terminate! ping))))
			       "timeout")))
    (nn-connect req (conc "tcp://" host ":" port))
    (handle-exceptions
     exn
     (begin
       (print-call-chain)
       (print 0 " message: " ((condition-property-accessor 'exn 'message) exn))
       (print "exn=" (condition->list exn))
       (print "ping failed to connect to " host ":" port))
     (thread-start! timeout)
     (thread-start! ping)
     (thread-join! ping)
     (if success (thread-terminate! timeout)))
    (if return-socket
	(if success req #f)
	(begin
	  (nn-close req)
	  success))))

(let ((server-thread (make-thread (lambda ()(server rep)) "server")))
  (thread-start! server-thread)
  ;; (thread-sleep! 1)
  (if (ping-self host port)
      (begin
	(thread-join! server-thread)
	(nn-close rep))
      (print "ping failed")))

(exit)

;; watch nanomsg's pipeline load-balancer in action.
(use nanomsg)

(define req   (nn-socket 'req))
(define rep   (nn-socket 'rep))

(nn-bind    rep  "inproc://test")
(nn-connect req  "inproc://test")

(define (client-send-receive soc msg)
  (nn-send soc msg)
  (nn-recv soc))

(define ((server soc))
  (let loop ((msg-in (nn-recv soc)))
    (if (not (equal? msg-in "quit"))
	(begin
	  (nn-send soc (conc "hello " msg-in))
	  (loop (nn-recv soc))))))

(thread-start! (server rep))

(print (client-send-receive req "Matt"))
(print (client-send-receive req "Tom"))

;; (client-send-receive req "quit")

(nn-close req)
(nn-close rep)
(exit)

stopserver :
	cd ..;make -j && make install
	cd fullrun;$(MEGATEST) -stop-server 0

repl :
	cd ..;make -j && make install
	cd fullrun;$(MEGATEST) -:b -repl

test0 : cleanprep
	cd simplerun ; $(MEGATEST) -server - -debug $(DEBUG)

test1 : cleanprep

test2 : fullprep

# Some simple checks for bootstrapping and run loop logic 

test9 : minsetup test9a test9b test9c test9d test9e

test9a :
	@echo Run super-simple mintest e, no waitons.
	cd mintest;$(DASHBOARD)&
	cd mintest;$(MEGATEST) -preclean -runtests e -target $(VER) -runname $(shell date +%H.%M.%S) -debug $(DEBUG)

test9b :
	@echo Run simple mintest d with one waiton c
	cd mintest;$(MEGATEST) -preclean -runtests d -target $(VER) -runname `date +%H.%M.%S` -debug $(DEBUG)

test9c :
	@echo Run mintest a with full waiton chain a -> b -> c -> d -> e
	cd mintest;$(MEGATEST) -preclean -runtests a -target $(VER) -runname `date +%H.%M.%S` -debug $(DEBUG)

test9d :
	@echo Run an itemized test with no items
	cd mintest;$(MEGATEST) -preclean -runtests g -target $(VER) -runname `date +%H.%M.%S` -debug $(DEBUG)

test9e :
	@echo Run mintest a1 with full waiton chain with d1fail: a1 -> b1 -> c1 -> d1fail -> e1
	cd mintest;$(MEGATEST) -preclean -runtests a1 -target $(VER) -runname `date +%H.%M.%S` -debug $(DEBUG)

test10 :
	@echo Run a bunch of different targets simultaneously
	(cd fullrun;$(MEGATEST) -server - ;sleep 2)&
	for targ in mint/btrfs/mintdir sunos/sshfs/loc; do \
	   (cd fullrun;$(MEGATEST) -preclean -runtests priority_10_waiton_1 -target $$targ :runname $(RUNNAME) &); done
	for sys in ubuntu suse redhat debian;do \

# If the server can't be started on this port it will try the next port until
# it succeeds
port 8080

# This server will keep running this number of hours after last access. 
# Three minutes is 0.05 hours
# timeout 0.025
timeout 0.061

# Server is required - slower but more resistant to Sqlite issues.
required yes

# Start server when average query takes longer than this
# server-query-threshold 55500
server-query-threshold -1


# daemonize yes
# hostname #{scheme (get-host-name)}

## disks are:
## name host:/path/to/area
## -or-

[fields]
X TEXT

[setup]
max_concurrent_jobs 50
linktree #{getenv MT_RUN_AREA_HOME}/linktree
transport http

[server]
port 8090

[jobtools]
useshell yes

(test "setup for run" #t (begin (launch:setup-for-run)
 				(string? (getenv "MT_RUN_AREA_HOME"))))

;; NON Server tests go here

(test #f #f (db:dbdat-get-path *db*))
(test #f #f (db:get-run-name-from-id *db* run-id))
;; (test #f '("SYSTEM" "RELEASE") (rmt:get-keys))

;; (exit)

;; Server tests go here 
(for-each (lambda (run-id)
(test #f #f (tasks:server-running-or-starting? (db:delay-if-busy (tasks:open-db)) run-id))
(server:kind-run run-id)
(test "did server start within 20 seconds?"
      #t
      (let loop ((remtries 20)
		 (running (tasks:server-running-or-starting? (db:delay-if-busy
							      (tasks:open-db))
							     run-id)))
	(if running 
	    (> running 0)
	    (if (> remtries 0)
		(begin
		  (thread-sleep! 1)
		  (loop (- remtries 1)
			(tasks:server-running-or-starting? (db:delay-if-busy
							    (tasks:open-db))
							   run-id)))))))

(test "did server become available" #t
      (let loop ((remtries 10)
		 (res      (tasks:get-server (db:delay-if-busy (tasks:open-db)) run-id)))
	(if res
	    (vector? res)
	    (begin
	      (if (> remtries 0)
		  (begin
		    (thread-sleep! 1.1)
		    (loop (- remtries 1)(tasks:get-server (db:delay-if-busy (tasks:open-db)) run-id)))
		  res)))))
)
(list 0 1))
(define user    (current-user-name))
(define runname "mytestrun")
(define keys    (rmt:get-keys))
(define runinfo #f)
(define keyvals '(("SYSTEM" "abc")("RELEASE" "def")))
(define header  (vector "SYSTEM" "RELEASE" "id" "runname" "state" "status" "owner" "event_time"))

      (let ((run-info (rmt:get-run-info run-id)))
	(db:get-value-by-header (db:get-rows run-info)
				(db:get-header run-info)
				"runname")))

      ;; (vector header (vector "abc" "def" 1 "mytestrun" "new" "n/a" "matt" 1416280640.0))

(for-each (lambda (run-id)
;; test killing server
;;
(tasks:kill-server-run-id run-id)

(test #f #f (tasks:server-running-or-starting? (db:delay-if-busy (tasks:open-db)) run-id))
)
(list 0 1))
;; (test #f #f (client:setup run-id))

;; (set! *transport-type* 'http)
;; 
;; (test "setup for run" #t (begin (launch:setup-for-run)
;; 				(string? (getenv "MT_RUN_AREA_HOME"))))
;; 

$(PREFIX)/bin/sqlite3 : sqlite-autoconf-$(SQLITE3_VERSION)/config.log
	cd sqlite-autoconf-$(SQLITE3_VERSION);./configure --prefix=$(PREFIX);make;make install

$(PREFIX)/lib/sqlite3.so : $(PREFIX)/bin/sqlite3
	CSC_OPTIONS="-I$(PREFIX)/include -L$(PREFIX)/lib" $(CHICKEN_INSTALL) $(PROX) sqlite3

#======================================================================
# N  A N O M S G
#======================================================================

nanomsg-0.5-beta.tar.gz :
	wget http://download.nanomsg.org/nanomsg-0.5-beta.tar.gz

nanomsg-0.5-beta/COPYING : nanomsg-0.5-beta.tar.gz
	tar xfvz nanomsg-0.5-beta.tar.gz

$(PREFIX)/bin/nanocat : nanomsg-0.5-beta/COPYING
	cd nanomsg-0.5-beta;./configure --prefix=$(PREFIX);make;make install

$(PREFIX)/lib/nanomsg.so : $(PREFIX)/bin/nanocat
	CSC_OPTIONS="-I$(PREFIX)/include -L$(PREFIX)/lib" $(CHICKEN_INSTALL) $(PROX) nanomsg

#======================================================================
# M A T T S   U T I L S
#======================================================================

opensrc.fossil :
	fossil clone http://www.kiatoa.com/fossils/opensrc opensrc.fossil

#!/mfs/pkgs/chicken/4.8.0.5/bin/csi -nbq

;; Coming soon (right?) Usage: plot-code file1.scm,file2.scm "fun1,fun2,x*" *.scm > plot.dot
;; Usage: plot-code file1.scm,file2.scm *.scm > plot.dot
;;        dot -Tpdf plot.dot > plot.pdf
;; first param is comma separated list of files to include in the map, use - to do all
;; second param is list of regexs for functions to include in the map
;; third param is list of files to scan

(use regex srfi-69 srfi-13)

(define targs #f) 
(define files (cddddr (argv)))

(let ((targdat (cadddr (argv))))
  (if (equal? targdat "-")
      (set! targs files)
      (set! targs (string-split targdat ","))))

(define filedat-defns (make-hash-table))
(define filedat-usages (make-hash-table))

(define defn-rx (regexp "^\\s*\\(define\\s+\\(([^\\s\\)]+).*"))
(define all-regexs (make-hash-table))

(define all-fns '())

;; for the se

(define (print-err . data)
  (with-output-to-port (current-error-port)
    (lambda ()
      (apply print data))))

(print-err "Making graph for files: " (string-intersperse targs ", "))

		    (loop (car tal)
			  (cdr tal)
			  newres)))
	      (if (null? tal)
		  res
		  (loop (car tal)(cdr tal) res)))))
      '()))

;; (define mm-header #<<MMHEADER
;; <map version="freeplane 1.2.0">
;; <!--To view this file, download free mind mapping software Freeplane from http://freeplane.sourceforge.net -->
;; 
;; MMHEADER
;; 
;; (define (add-node text)
;;   <node TEXT="homenode" ID="ID_1723255651" CREATED="1283093380553" MODIFIED="1417113442955"><hook NAME="MapStyle">
;; )
;; 
;;  minimal mindmap file
;;  <map version="freeplane 1.2.0">
;;   <!--To view this file, download free mind mapping software Freeplane from http://freeplane.sourceforge.net -->
;;   <node TEXT="homenode" ID="ID_1723255651" CREATED="1283093380553" MODIFIED="1417113442955">
;;   <node TEXT="node1" POSITION="right" ID="ID_1810107939" CREATED="1417113473476" MODIFIED="1417113480425">
;;   <node TEXT="node2" ID="ID_68133256" CREATED="1417113482134" MODIFIED="1417113484466"/>
;;   <node TEXT="node3" ID="ID_1572284821" CREATED="1417113487785" MODIFIED="1417113491589"/>
;;   </node>
;;   </node>
;;   </map>

;; Gather the usages
(print "digraph G {")
(define curr-cluster-num 0)
(define function-calls '())

(for-each
 (lambda (fname)

;;;; trace.import.scm - GENERATED BY CHICKEN 4.9.0.1 -*- Scheme -*-

(eval '(import
         scheme
         chicken
         csi
         advice
         extras
         ports
         data-structures
         (except srfi-1 break)
         miscmacros))
(##sys#register-compiled-module
  'trace
  (list)
  '((breakpoint . trace#breakpoint)
    (trace . trace#trace)
    (untrace . trace#untrace)
    (break . trace#break)
    (unbreak . trace#unbreak)
    (trace-output-port . trace#trace-output-port)
    (continue . trace#continue)
    (c . trace#c)
    (traced? . trace#traced?)
    (trace-module . trace#trace-module)
    (untrace-module . trace#untrace-module)
    (trace-verbose . trace#trace-verbose)
    (trace/untrace . trace#trace/untrace))
  (list)
  (list))

;; END OF FILE

;;;; trace.meta -*- Scheme -*-


((category tools)
 (synopsis "tracing and breakpoints")
 (author "felix winkelmann")
 (license "public domain")
 (needs advice				; don't we all?
	miscmacros)
 (files "tests/run.scm" "trace.meta" "trace.release-info" "trace.scm" "trace.setup") )

;;;; trace.scm


(module trace (breakpoint 
	       trace untrace
	       break unbreak
	       trace-output-port
	       continue c 
	       traced?
	       trace-module untrace-module
	       trace-verbose
	       trace/untrace)
	       
(import scheme chicken csi)

(use advice extras ports data-structures)
(require-library srfi-1)
(import (except srfi-1 break) miscmacros)


(define *last-breakpoint* #f)
(define *traced-procedures* '())
(define *broken-procedures* '())
(define *trace-indent-level* 0)

(define trace-output-port (make-parameter (current-output-port)))
(define trace-verbose (make-parameter #t))

(define (break-entry name args)
  ;; Does _not_ unwind!
  (##sys#call-with-current-continuation
   (lambda (c)
     (let ((exn (##sys#make-structure
		 'condition
		 '(exn breakpoint)
		 (list '(exn . message) "*** breakpoint ***"
		       '(exn . arguments) (list (cons name args))
		       '(exn . location) name
		       '(exn . continuation) c) ) ) )
       (set! *last-breakpoint* exn)
       (signal exn) ) ) ) )

(define (break-resume exn)
  (let ((a (member '(exn . continuation) (##sys#slot exn 2))))
    (if a
	((cadr a) (void))
	(error "condition has no continuation" exn) ) ) )

(define (breakpoint #!optional (name 'breakpoint))
  (break-entry name '()) )

(define (trace-indent)
  (let ((port (trace-output-port)))
    (do ((i (fxmin 3 *trace-indent-level*) (fx- i 1)))
	((fx<= i 0))
      (write-char #\space port) )
    (fprintf port "[~a] " *trace-indent-level*) ) )

(define (traced-procedure-entry name args)
  (let ((port (trace-output-port)))
    (trace-indent)
    (set! *trace-indent-level* (fx+ 1 *trace-indent-level*))
    (write (cons name args) port)
    (write ", Called from: " port)
    (write (conc (car (reverse (get-call-chain)))))
    (write-char #\newline port)
    (flush-output port) ) )

(define (traced-procedure-exit name results)
  (let ((port (trace-output-port)))
    (set! *trace-indent-level* (fx- *trace-indent-level* 1))
    (trace-indent)
    (fprintf port "~a -> " name)
    (if results
	(for-each
	 (lambda (x)
	   (write x port)
	   (write-char #\space port) )
	 results)
	(display "(escaping)" port))
    (write-char #\newline port)
    (flush-output port) ) )

(define (procedure-name proc)
  (cond ((procedure-information proc) =>
	 (lambda (info)
	   (if (pair? info) (car info) info) ) )
	(else '<unknown>)) )

(define (do-trace procs)
  (for-each
   (lambda (s)
     (ensure procedure? s)
     (cond ((traced? s)
	    (warning "procedure already traced" s) )
	   (else
	    (let ((name (procedure-name s)))
	      (when (trace-verbose)
		(fprintf (current-error-port) "; tracing ~a~%" name))
	      (set! *traced-procedures* (cons (cons s name) *traced-procedures*))
	      (advise 
	       'around s
	       (lambda (next args)
		 (let ((results #f))
		   (dynamic-wind
		       (cut traced-procedure-entry name args)
		       (lambda () 
			 (call-with-values (cut apply next args)
			   (lambda rs
			     (set! results rs)
			     (apply values rs))))
		       (cut traced-procedure-exit name results))))
	       '*trace*)))))
   procs) )

(define (do-untrace-all)
  (define (unadvise* p)
    (ignore-errors (unadvise p '*trace*)))
  (for-each
   (lambda (proc)
     (let ((proc (car proc)))
       (when (trace-verbose)
	 (fprintf (current-error-port) "; untracing ~a~%" (procedure-name proc))
	 (unadvise* proc))))
   *traced-procedures*)
  (set! *traced-procedures* '()))

(define (do-untrace procs)
  (for-each
   (lambda (s)
     (ensure procedure? s)
     (let ((p (assq s *traced-procedures*)) 
	   (name (procedure-name s)))
       (cond ((not p) (warning "procedure not traced" name))
	     (else
	      (when (trace-verbose)
		(fprintf (current-error-port) "; untracing ~a~%" name))
	      (ignore-errors (unadvise s '*trace*))
	      (set! *traced-procedures* 
		(delete 
		 p *traced-procedures* 
		 eq?))))))
   procs) )

(define (do-break procs)
  (for-each
   (lambda (s)
     (let ((name (procedure-name s)))
       (ensure procedure? s)
       (cond ((assq s *broken-procedures*)
	      (warning "procedure already has break-point" name))
	     (else
	      (when (trace-verbose)
		(fprintf (current-error-port) "; setting break-point in ~a~%" name))
	      (set! *broken-procedures* (cons (cons s name) *broken-procedures*))
	      (advise 
	       'before s
	       (lambda (args)
		 (break-entry name args) )
	       '*break*) ) )))
   procs) )

(define (do-unbreak procs)
  (for-each
   (lambda (s)
     (ensure procedure? s)
     (let ((p (assq s *broken-procedures*)) 
	   (name (procedure-name s)))
       (cond ((not p) (warning "procedure has no breakpoint" name))
	     (else
	      (when (trace-verbose)
		(fprintf (current-error-port) "; removing break-point in ~a~%" name))
	      (ignore-errors (unadvise s '*break*))
	      (set! *broken-procedures* (delete p *broken-procedures* eq?) ) ) ) ) )
   procs) )

(define (do-unbreak-all)
  (for-each
   (lambda (bp)
     (ignore-errors (unadvise (car bp) '*break*)))
   *broken-procedures*)
  (set! *broken-procedures* '())
  (void))

(define (trace . procs)
  (cond ((null? procs)
	 (when (pair? *traced-procedures*)
	   (printf "Traced:~%~%")
	   (for-each (lambda (p) (printf "  ~a~%" (cdr p))) *traced-procedures*)) )
	(else
	 (do-trace procs) ) ) )

(define (untrace . procs)
  (cond ((null? procs) (do-untrace-all))
	(else (do-untrace procs)))
  (void))

(define (break . procs)
  (cond ((null? procs)
	 (when (pair? *broken-procedures*)
	   (printf "Breakpoints:~%~%")
	   (for-each (lambda (p) (printf "  ~a~%" (cdr p))) *broken-procedures*)) )
	(else
	 (do-break procs) ) ) )

(define (unbreak . procs)
  (cond ((null? procs) (do-unbreak-all))
	(else (do-unbreak procs))))

(define (continue #!optional (bp *last-breakpoint*))
  (cond (*last-breakpoint*
	 (let ((exn *last-breakpoint*))
	   (set! *last-breakpoint* #f)
	   (break-resume exn) ) )
	(else (display "no breakpoint pending\n") ) ) )

(define c continue)

(define (traced? proc)
  (assq proc *traced-procedures*))

(define (trace/untrace . procs)
  (for-each
   (lambda (proc)
     ((if (traced? proc) do-untrace do-trace) (list proc)))
   procs))

(define (walk-module mname proc)
  (let* ((m (##sys#find-module mname))
	 (exps (nth-value 1 (##sys#module-exports m))))
    (for-each
     (lambda (exp)
       (let* ((realname (cdr exp))
	      (prim (get realname '##core#primitive)))
	 (if prim
	     (warning "export is a core-library primitive - not traced" (car exp))
	     (when (##sys#symbol-has-toplevel-binding? realname)
	       (let ((val (##sys#slot realname 0)))
		 (when (procedure? val)
		   (proc val)))))))
     exps)))

(define (trace-module . mnames)
  (for-each
   (lambda (mname)
     (walk-module mname trace))
   mnames))

(define (untrace-module . mnames)
  (for-each
   (lambda (mname)
     (walk-module 
      mname 
      (lambda (proc)
	(when (traced? proc)
	  (do-untrace (list proc))))))
   mnames))

)

;;;; trace.setup -*- Scheme -*-


(compile -s trace.scm -O3 -d1 -j trace)
(compile -s trace.import.scm -O3 -d0)

(install-extension
 'trace
 '("trace.so" "trace.import.so"))