/usr/share/doc/scheme48/html/manual-Z-H-4.html

<!doctype html public "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
<html>
<!--

Generated from manual.tex by tex2page, v 20100828
(running on MzScheme 4.2.4, :unix), 
(c) Dorai Sitaram, 
http://evalwhen.com/tex2page/index.html

-->
<head>
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
<title>
The Incomplete Scheme 48 Reference Manual for release 1.9
</title>
<link rel="stylesheet" type="text/css" href="manual-Z-S.css" title=default>
<meta name=robots content="index,follow">
</head>
<body>
<div id=slidecontent>
<div align=right class=navigation>[Go to <span><a href="manual.html">first</a>, <a href="manual-Z-H-3.html">previous</a></span><span>, <a href="manual-Z-H-5.html">next</a></span> page<span>; &nbsp;&nbsp;</span><span><a href="manual-Z-H-1.html#node_toc_start">contents</a></span><span><span>; &nbsp;&nbsp;</span><a href="manual-Z-H-11.html#node_index_start">index</a></span>]</div>
<p></p>
<a name="node_chap_3"></a>
<h1 class=chapter>
<div class=chapterheading><a href="manual-Z-H-1.html#node_toc_node_chap_3">Chapter 3</a></div><br>
<a href="manual-Z-H-1.html#node_toc_node_chap_3">Command processor</a></h1>
<p></p>
<p>
This chapter details Scheme&nbsp;48's command processor, which incorporates
both a read-eval-print loop and an interactive debugger.
At the <tt>&gt;</tt> prompt, you can type either a Scheme form
(expression or definition) or a command beginning with a comma.
In
inspection mode (see section&nbsp;<a href="#node_sec_3.8">3.8</a>)
the prompt changes to <tt>:</tt> and commands
no longer need to be preceded by a comma; input beginning with
a letter or digit is assumed to be a command, not an expression.
In inspection mode the command processor prints out a
menu of selectable components for the current object of interest.</p>
<p>
</p>
<a name="node_sec_3.1"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.1">3.1&nbsp;&nbsp;Current focus value and <tt>##</tt></a></h2>
<p>The command processor keeps track of a current <em>focus value</em>.
This value is normally the last value returned by a command.
If a command returns multiple values the focus object is a list of the
values.
The focus value is not changed if a command returns no values or 
a distinguished `unspecific' value.
Examples of forms that return this unspecific value are definitions,
uses of <tt>set!</tt>, and <tt>(if #f 0)</tt>.
It prints as <tt>#{Unspecific}</tt>.</p>
<p>
The reader used by the command processor reads <tt>##</tt> as a special
expression that evaluates to the current focus object.
</p>
<pre class=verbatim>&gt; (list 'a 'b)
'(a b)
&gt; (car ##)
'a
&gt; (symbol-&gt;string ##)
&quot;a&quot;
&gt; (if #f 0)
#{Unspecific}
&gt; ##
&quot;a&quot;
&gt; 
</pre><p></p>
<p>
</p>
<a name="node_sec_3.2"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.2">3.2&nbsp;&nbsp;Command levels</a></h2>
<p>If an error, keyboard interrupt, or other breakpoint occurs, or the
<tt>,push</tt> command is used, the command
processor invokes a recursive copy of itself, preserving the dynamic state of
the program when the breakpoint occurred.
The recursive invocation creates a new <em>command level</em>.
The command levels form a stack with the current level at the top.
The command prompt indicates the number of stopped levels below the
current one: <tt>&gt;</tt> or <tt>:</tt> for the
base level and <tt><i>n</i>&gt;</tt> or <tt><i>n</i>:</tt> for all other levels,
where <i>n</i> is the command-level nesting depth.
The <tt>levels</tt> setting
described below can be used to disable the automatic pushing of new levels.</p>
<p>
The command processor's evaluation package and the value of the
current focus value are local to each command level.
They are preserved when a new level is pushed and restored when
it is discarded.
The settings of all other settings are shared by all command levels.</p>
<p>
</p>
<dl><dt></dt><dd>
</dd><dt></dt><dd>&lang;eof&rang;<br>
Discards the current command level and resumes running the level down.
&lang;eof&rang; is usually
control-<tt>D</tt> at a Unix shell or control-<tt>C</tt> control-<tt>D</tt> using
the Emacs <tt>cmuscheme48</tt> library.<p>
</p>
</dd><dt></dt><dd><tt>,pop</tt><br>
The same as &lang;eof&rang;.<p>
</p>
</dd><dt></dt><dd><tt>,proceed [<i>exp</i> <tt>...</tt></tt>]<br>
Proceed after an interrupt or error, resuming the next command
level down, delivering the values of <i>exp&nbsp;<tt>...</tt></i> to the continuation.
Interrupt continuations discard any returned values.
<tt>,Pop</tt> and <tt>,proceed</tt> have the same effect after an interrupt
but behave differently after errors.
<tt>,Proceed</tt> restarts the erroneous computation from the point where the
error occurred (although not all errors are proceedable) while
<tt>,pop</tt> (and &lang;eof&rang;) discards it and prompts for
a new command.<p>
</p>
</dd><dt></dt><dd><tt>,push</tt><br>
Pushes a new command level on above the current one.
This is useful if the <tt>levels</tt> setting has been used
to disable the automatic pushing of new levels for errors and interrupts.<p>
</p>
</dd><dt></dt><dd><tt>,reset [<i>number</i>]</tt><br>
Pops down to a given level and restarts that level.
<i>Number</i> defaults to zero, <tt>,reset</tt> restarts the command
processor, discarding all existing levels.<p>
</p>
</dd></dl><p></p>
<p>
Whenever moving to an existing level, either by sending
an &lang;eof&rang;
or by using <tt>,reset</tt> or the other commands listed above,
the command processor runs all of the <tt>dynamic-wind</tt> &ldquo;after&rdquo; thunks
belonging to stopped computations on the discarded level(s).</p>
<p>
</p>
<a name="node_sec_3.3"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.3">3.3&nbsp;&nbsp;Logistical commands</a></h2>
<p></p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>,load <i>filename <tt>...</tt></i></tt><br>
Loads the named Scheme source file(s).
Easier to type than <tt>(load &quot;<i>filename</i>&quot;)</tt> because you don't have to
shift to type the parentheses or quote marks.  
(However, it is still possible to specify a filename as a Scheme
string literal, with quote marks&mdash;you'll need this for filenames
containing whitespace.)
Also, it works in
any package, unlike <tt>(load &quot;<i>filename</i>&quot;)</tt>, which will work only
in packages in which the variable <tt>load</tt> is defined appropriately.<p>
</p>
</dd><dt></dt><dd><tt>,exit [<i>exp</i>]</tt><br>
Exits back out to shell (or executive or whatever invoked Scheme&nbsp;48
in the first place).
<i>Exp</i> should evaluate to an integer.  The
integer is returned to the calling program.
The default value of <i>exp</i> is zero, which, on Unix,
is generally interpreted as success.
</dd></dl><p></p>
<p>
</p>
<a name="node_sec_3.4"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.4">3.4&nbsp;&nbsp;Module commands</a></h2>
<p></p>
<p>
There are many commands related to modules.
Only the most commonly used module commands are described here;
documentation for the
rest can be found in section&nbsp;<a href="manual-Z-H-5.html#node_sec_4.8">4.8</a>.
There is also
a brief description of modules, structures, and packages in section&nbsp;<a href="manual-Z-H-3.html#node_sec_2.6">2.6</a> below.</p>
<p>
</p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>,open <i>structure <tt>...</tt></i></tt><br>
Makes the bindings in the <i>structure</i>s visible in the current package.
The packages associated with the <i>structure</i>s will be loaded if
this has not already been done (the <tt>ask-before-loading</tt> setting
can be used to disable the automatic loading of packages).<p>
</p>
</dd><dt></dt><dd><tt>,config [<i>command</i>]</tt><br>
Executes <i>command</i> in the <tt>config</tt> package, which includes
the module configuration language.
For example, use
<pre class=verbatim>,config ,load <i>filename</i>
</pre><p>
to load a file containing module definitions.
If no <i>command</i> is given, the <tt>config</tt> package becomes the
execution package for future commands.</p>
<p>
</p>
</dd><dt></dt><dd><tt>,user [<i>command</i>]</tt> <br>
This is similar to the <tt>,config</tt>.  It
moves to or executes a command in the user package (which is the
default package when the Scheme&nbsp;48 command processor starts).<p>
</p>
</dd></dl><p></p>
<p>
</p>
<a name="node_sec_3.5"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.5">3.5&nbsp;&nbsp;Debugging commands</a></h2>
<p></p>
<p>
</p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>,preview</tt><br>
Somewhat like a backtrace, but because of tail recursion you see
less than you might in debuggers for some other languages.
The stack to display is chosen as follows:
<ol>
<li><p>If the current focus object is a continuation or a thread,
then that continuation or thread's stack is displayed.
</p>
<li><p>Otherwise, if the current command level was initiated because of
a breakpoint in the next level down, then the stack at that
breakpoint is displayed.
</p>
<li><p>Otherwise, there is no stack to display and a message is printed
to that effect.
</p>
</ol><p>
One line is printed out for each continuation on the chosen stack,
going from top to bottom.</p>
<p>
</p>
</dd><dt></dt><dd><tt>,run <i>exp</i></tt><br>
Evaluate <i>exp</i>, printing the result(s) and making them
(or a list of them, if <i>exp</i> returns multiple results)
the new focus object.
The <tt>,run</tt> command is useful in inspection mode (see
section&nbsp;<a href="#node_sec_3.8">3.8</a> below) and when writing
command programs (see section&nbsp;<a href="#node_sec_3.9">3.9</a> below).<p>
</p>
</dd><dt></dt><dd><tt>,trace <i>name</i> <tt>...</tt></tt><br>
Start tracing calls to the named procedure or procedures.
With no arguments, displays all procedures currently traced.
This affects the binding of <i>name</i>, not the behavior of the
procedure that is its current value.  <i>Name</i> is redefined
to be a procedure that prints a message,
calls the original value of <i>name</i>, prints another
message, and finally passes along the value(s) returned by the
original procedure.<p>
</p>
</dd><dt></dt><dd><tt>,untrace <i>name</i> <tt>...</tt></tt><br>
Stop tracing calls to the named procedure or procedures.
With no argument, stop tracing all calls to all procedures.<p>
</p>
</dd><dt></dt><dd><tt>,condition</tt><br>
The <tt>,condition</tt> command displays the condition object
describing the error or interrupt that initiated the current
command level.  The condition object becomes the current focus
value.  This is particularly useful in conjunction with
the inspector.  For example, if a procedure is passed the wrong number of
arguments, do <tt>,condition</tt> followed by
<tt>,inspect</tt> to inspect the
procedure and its arguments.<p>
</p>
</dd><dt></dt><dd><tt>,bound? <i>name</i></tt><br>
Display the binding of <i>name</i>, if there is one, and otherwise
prints `<tt>Not bound</tt>'.<p>
</p>
</dd><dt></dt><dd><tt>,expand <i>form</i></tt>
Show macro expansion of <i>form</i>, if any, fully expanding all 
macros in <i>form</i>.<p>
</p>
</dd><dt></dt><dd><tt>,where <i>procedure</i></tt><br>
Display name of file containing <i>procedure</i>'s source code.
</dd></dl><p></p>
<p>
</p>
<a name="node_sec_3.6"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.6">3.6&nbsp;&nbsp;Profiling commands</a></h2>
<p></p>
<p>
After loading the <tt>profile-commands</tt> structure via
</p>
<pre class=verbatim>,load-package profile-commands  
</pre><p>
Scheme&nbsp;48 provides a basic profiler, with support for flat and call-graph profiles.
In addition to call- and runtimes the heap memory usage is estimated as well.
By adding the byte-code optimizer <tt>profiler-instrumentation</tt> to the definition
of a structure via a clause
</p>
<pre class=verbatim>  (optimize profiler-instrumentation)
</pre><p>
in the structure definition, code will by instrumented with calls to the profiler. The call-times to
instrumented code can then be measured exactly.
Without instrumentation, the profiler is solely based on sampling and therefore has only
limited accuracy.
Any evaluation can be profiled with the following command:</p>
<p>
</p>
<dl><dt></dt><dd><p>
</p>
</dd><dt></dt><dd><tt>,profile <i>command</i></tt><br>
Evaluate <i>command</i> under the profiler, printing the profiling result and making the
return value(s) of <i>command</i> the new focus object.<p>
</p>
</dd></dl><p></p>
<p>
The output of the profiler is explained on the basis of the following piece of senseless code:</p>
<p>
</p>
<pre class=verbatim>(define (c x)
  (if (= x 0)
      0
      (+ 1 (c (- x 1)))))

(define (b x)
  (let ((y (- x 1)))
    (if (&gt; y 0)
        (begin
          (c y)
          (a y)
          (+ 1 (a y)))
        0)))

(define (a x)
  (let ((y (- x 1)))
    (if (&gt; y 0)
        (begin
          (c y)
          (b y)
          (+ 1 (b y)))
        0)))

(define (main x)
  (+ 1 (a x)))
</pre><p></p>
<p>
The profiler produces the following output:</p>
<p>
</p>
<pre class=verbatim>&gt; ,profile (main 21)

** Samples:        60 (approx. one per 40ms)
** Interrupt time: 50ms
** Real run time:  2430ms
** Total memory:   22k
** GC runs:        0

** Flat result (times in ms): 

   time  cumu  self   mem  calls  ms/call  name
 61.66%  1498  1498    2k  37+51       17  c in &quot;x.scm&quot;
  25.0%  2430   608   10k    436        6  b in &quot;x.scm&quot;
 13.33%  2430   324   10k    442        5  a in &quot;x.scm&quot;
   0.0%  2430     0    0k      1     2430  main in &quot;x.scm&quot;

** Tree result (times in ms): 

i    time self child  mem    calls  name
0  100.0%    0  2430   0k        0  &lt;profiler&gt; [0]
             0  2430   0k      1/1     main in &quot;x.scm&quot; [1]
=========================================================
                           441/442     b in &quot;x.scm&quot; &lt;cycle 0&gt; [3]
           324   891  10k    1/442     main in &quot;x.scm&quot; [1]
1  100.0%  324   891  10k      442  a in &quot;x.scm&quot; &lt;cycle 0&gt; [2]
           891     0   1k    22/37     c in &quot;x.scm&quot; [4]
                           436/436     b in &quot;x.scm&quot; &lt;cycle 0&gt; [3]
=========================================================
                           436/436     a in &quot;x.scm&quot; &lt;cycle 0&gt; [2]
2  100.0%  608   608  10k      436  b in &quot;x.scm&quot; &lt;cycle 0&gt; [3]
           608     0   1k    15/37     c in &quot;x.scm&quot; [4]
                           441/442     a in &quot;x.scm&quot; &lt;cycle 0&gt; [2]
=========================================================
             0  2430   0k      1/1     &lt;profiler&gt; [0]
3  100.0%    0  2430   0k        1  main in &quot;x.scm&quot; [1]
           324   891  10k    1/442     a in &quot;x.scm&quot; &lt;cycle 0&gt; [2]
=========================================================
           608     0   1k    15/37     b in &quot;x.scm&quot; &lt;cycle 0&gt; [3]
           891     0   1k    22/37     a in &quot;x.scm&quot; &lt;cycle 0&gt; [2]
4  61.66% 1498     0   2k    37+51  c in &quot;x.scm&quot; [4]
=========================================================
           932  1498  20k      1/1     main in &quot;x.scm&quot; [1]
0  100.0%  932  1498  20k    1+877  &lt;cycle 0 as a whole&gt;
           324   891  10k      441     a in &quot;x.scm&quot; &lt;cycle 0&gt; [2]
           608   608  10k      436     b in &quot;x.scm&quot; &lt;cycle 0&gt; [3]
             0     0   2k    37/37     c in &quot;x.scm&quot; [4]

</pre><p></p>
<p>
The formatting of the output is based on the output of
GNU gprof <a name="node_call_footnote_Temp_3"></a><sup><small><a href="#node_footnote_Temp_3">1</a></small></sup>.</p>
<p>
First, general information about the profiling is shown:
</p>
<ul>
<li><p><tt>samples</tt>: total number of samples taken and the
average time between two samples
</p>
<li><p><tt>interrupt time</tt>: theoretical interrupt time, the time between
two samples
</p>
<li><p><tt>real run time</tt>: the total run time of the evaluation, simple measurement
from the beginning to the end of the evaluation
</p>
<li><p><tt>total memory</tt>: the total amount of memory used by the program, here in kilobytes
</p>
<li><p><tt>GC runs</tt>: the number of times the garbage collector was running while evaluating
</p>
</ul><p></p>
<p>
The flat profile gives overall statistics for each procedure in the program:
</p>
<ul>
<li><p><tt>time</tt>:
percentage of time the procedure has in the program (based on &ldquo;<tt>self</tt>&rdquo;)
</p>
<li><p><tt>cumu</tt>:
total cumulative time the procedure was running
</p>
<li><p><tt>self</tt>:
total time the procedure was running itself (without children)
</p>
<li><p><tt>mem</tt>:
approximated memory usage of the procedure
</p>
<li><p><tt>calls</tt>:
the number of non-recursive calls to the procedure. Recursive calls are displayed after an
optionally appended &quot;+&quot; to former number (see procedure <tt>c</tt> in example output above).
</p>
<li><p><tt>ms/call</tt>:
estimate of the time per call (with children), based on non-recursive calls and &ldquo;<tt>cumu</tt>&rdquo;.
</p>
<li><p><tt>name</tt>:
the name of the procedure and the file it is defined in
</p>
</ul><p></p>
<p>
The tree result displays profiling information depending on the caller of a particular procedure.
As in gprof output, for each procedure there is an <i>entry</i>. Entries are separated by lines of <tt>=</tt>s.
Each entry displays the callers (indented), the procedure itself and the procedure called (indented).
The meaning of the fields depends on the type of the entry, for example:</p>
<p>
</p>
<pre class=verbatim> i    time self  child  mem    calls   name
            324    891  10k    1/442      main [1]
 1  100.0%  324    891  10k      442   a &lt;cycle 0&gt; [2]
            891      0   1k    22/37      c [4]
</pre><p></p>
<p>
Every procedure has a unique index number <tt>i</tt> (here &ldquo;1&rdquo;), based on the position in the
call-graph. It is used to quickly find the corresponding entry. The number is appended to
the procedure name (in brackets).
The meaning of the &ldquo;cycle&rdquo; suffix is explained below.</p>
<p>
Callers (here <tt>main</tt>):
</p>
<ul>
<li><p><tt>self</tt>:
time spent directly in <tt>a</tt> when called by <tt>main</tt>
</p>
<li><p><tt>child</tt>:
time spent in children of <tt>a</tt> when called by <tt>main</tt>
</p>
<li><p><tt>mem</tt>:
memory usage of <tt>a</tt> when called by <tt>main</tt>
</p>
<li><p><tt>calls</tt>:
two numbers:
number of calls to <tt>a</tt> from <tt>main</tt>
and
total number of non-recursive calls to <tt>a</tt>
</p>
</ul><p></p>
<p>
Primary line (here <tt>a</tt>):
</p>
<ul>
<li><p><tt>time</tt>:
total percentage of time <tt>a</tt> was running
</p>
<li><p><tt>self</tt>:
total time spent directly in <tt>a</tt>
</p>
<li><p><tt>child</tt>:
total time spent in children of <tt>a</tt>
</p>
<li><p><tt>mem</tt>:
total memory usage of <tt>a</tt>
</p>
<li><p><tt>calls</tt>:
the total number of non-recursive calls to <tt>a</tt>. Recursive calls are
displayed after an optional <tt>+</tt>.
</p>
</ul><p></p>
<p>
Called (here <tt>c</tt>):
</p>
<ul>
<li><p><tt>self</tt>:
time spent directly in <tt>c</tt> when called by <tt>a</tt>
</p>
<li><p><tt>child</tt>:
time spent in children of <tt>c</tt> when called by <tt>a</tt>
</p>
<li><p><tt>mem</tt>:
memory usage of <tt>c</tt> when called by <tt>a</tt>
</p>
<li><p><tt>calls</tt>:
two numbers:
number of calls to <tt>c</tt> from <tt>a</tt>
and
total number of non-recursive calls to <tt>c</tt>
</p>
</ul><p></p>
<p>
A special treatment is needed for mutual-recursive procedures. In the example,
<tt>a</tt> calls <tt>b</tt> and <tt>b</tt> calls <tt>a</tt>. This is called a <i>cycle</i>.
If <tt>a</tt> would call another procedure <tt>d</tt> and <tt>d</tt> would call <tt>a</tt>, all three
procedures form a cycle.</p>
<p>
Cycles have a separate entry in the call graph.
This entry consists of the callers into the cycle, the primary line of the cycle,
the member procedures of the cycle and the external procedures called by the cycle:</p>
<p>
</p>
<pre class=verbatim>            932   1498  20k      1/1      main [1]
 0  100.0%  932   1498  20k    1+877   &lt;cycle 0 as a whole&gt;
            324    891  10k      441      a &lt;cycle 0&gt; [2]
            608    608  10k      436      b &lt;cycle 0&gt; [3]
              0      0   2k    37/37      c [4]
</pre><p></p>
<p>
Callers (here <tt>main</tt>):
</p>
<ul>
<li><p>the meaning of the fields are the same
as above. The member procedures of the cycle are seen as a whole.
</p>
</ul><p></p>
<p>
Primary line:
</p>
<ul>
<li><p><tt>time</tt>:
total percentage of time any procedure of the cycle was running
</p>
<li><p><tt>self</tt>:
total time spent directly in a procedure of the cycle
</p>
<li><p><tt>child</tt>:
total time spent in external children of <tt>a</tt>
</p>
<li><p><tt>mem</tt>:
sum of memory usage of all cycle member procedures
</p>
<li><p><tt>calls</tt>:
the total number of external calls to the <tt>cycle</tt> and the total number
of calls in the cycle internally
</p>
</ul><p></p>
<p>
Member procedures:
</p>
<ul>
<li><p>the meaning of the fields are the same as above, except<tt>...</tt></p>
<li><p><tt>calls</tt>: the number of calls to the member procedure from within the cycle
</p>
</ul><p></p>
<p>
External procedures (here <tt>c</tt>):
</p>
<ul>
<li><p>the meaning of the fields are the same as with the member procedures, except<tt>...</tt></p>
<li><p><tt>calls</tt>: the number of calls to the external procedure from the cycle
and
the total number of non-recursive calls to the external procedure
</p>
</ul><p></p>
<p>
In the current implementation, there are some issues that need to be considered:
in the default configuration, samples are taken every 50 milliseconds. Procedures with a by-call
run time shorter than the interrupt time are likely to be profiled inaccurately or may not be seen
at all, if they have not been instrumented.</p>
<p>
The default interrupt time can be set with the <tt>profiler-interrupt-time</tt> command processor setting,
see <a href="#node_sec_3.7">3.7</a>. For example,
</p>
<pre class=verbatim>,set profiler-interrupt-time 200
</pre><p>
sets the
default profiler sampling interrupt time to 200ms.</p>
<p>
Because the interrupt is shared with the thread system, profiling may cause performance issues
with multi-threaded programs. Also, programs with deep recursion (resulting in large
continuations) can cause the profiler to use a significant percentage of the total run time.</p>
<p>
By default, call times are measured by the sampling process. When code had been instrumented,
call times of those procedures will be measured exactly. When only instrumented code should be
considered in the output of the profiler, the command processor setting
<tt>profiler-measure-noninstr</tt> can be set to <tt>off</tt>.</p>
<p>
Scheme&nbsp;48 optimizes tail calls. Thus, the profiler cannot &ldquo;see&rdquo;
tail calls and in some situations,
the output of the profiler may show that procedures directly call their &ldquo;grandchildren&rdquo;.</p>
<p>
</p>
<a name="node_sec_3.6.1"></a>
<h3 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.6.1">3.6.1&nbsp;&nbsp;How the profiler works</a></h3>
<p>The profiler is based on taking samples of the current stack of continuations, the <i>call-stack</i>.
Therefore it schedules the alarm interrupt at a regular interval specified by the
<tt>profiler-interrupt-time</tt> command processor setting. Because the thread system uses the
alarm interrupt as well, the profiler interrupt handler
calls the handler of the thread system after processing the sample.</p>
<p>
The profiler interrupt handler first captures the current continuation and follows the continuation-stack
down to the continuation of the <tt>profile-thunk</tt> procedure, which called the thunk to be profiled.
Each continuation with the corresponding code template is stored in a stack of <tt>stackentry</tt> records.</p>
<p>
This <tt>*cur-stack*</tt> is compared from bottom to top with the <tt>*last-stack*</tt>,
the stack captured while the prior sampling interrupt.
The statistics gathered are stored in <tt>profinfo</tt> records.</p>
<p>
For each stack-entry in the <tt>*cur-stack*</tt>, several possibilities exist in
conjunction to the entry at the same depth in the <tt>*last-stack*</tt>:
</p>
<ul>
<li><p>there is no such entry: the continuation must be the result of a new call
</p>
<li><p>both entries are the same continuation: nothing changed, the procedure is still running
</p>
<li><p>entries are not the same continuation:
the following entries up in the stack must have changed. If the templates of the current
entries are the same, the continuation counts as the still same call to the procedure (but at another
expression in the procedure). A such situation only counts as a new call if the continuation
object has changed, but the continuation program-counter, arguments and the template stayed the same.
</p>
</ul><p>
As we ascend in the call-stack, changes in the lower levels indicate that all entries above have to be new calls.
This way, the profiler gathers information about how often a template is called by another template.</p>
<p>
Additionally, every time a procedure is seen, we record this &ldquo;occurrence&rdquo;. If the procedure is a the top of
the stack, it is currently running. Based on these numbers, the average interrupt time and the number of calls,
we can estimate the self and child times.</p>
<p>
While running, the profiler collects information about heap memory usage as well.
After comparing the current with the last stack, the profiler knows which procedures finished, which procedures are new
and which procedure called the new procedures. This information is used to distribute the difference in
heap memory usage between the prior and the current sample.
The profiler uses the <tt>memory-status</tt> primitive for retrieving available heap space and the number of
garbage collector runs (<tt>gc-count</tt>).
Difference in used memory between two samples is distributed by the following principles:
</p>
<ul>
<li><p>if <tt>gc-count</tt> increased, distribution cannot be done
</p>
<li><p>if the same continuation is still at top, credit its template with all used memory between the two samplings
</p>
<li><p>if there are stackentry-templates gone or new, distribute the used memory <i>equally</i> between them:
</p>
<ul>
<li><p>gone templates could have used some memory before &ldquo;returning&rdquo;
</p>
<li><p>new  templates could have used some memory while running
</p>
<li><p>the caller of new templates could have used some memory between the calling of gone and new procedures
</p>
</ul><p>
</p>
</ul><p></p>
<p>
The byte-code optimizer <tt>profiler-instrumentation</tt> adds a call to the &ldquo;profile-count&rdquo; procedure of the
profiler structure. This procedure is responsible for the exact call-times measurement.</p>
<p>
After collection of the data, procedures are numbered and cycles are detected by ascending in the
call tree. Time is being propagated from top to bottom, remembering self and child times. Most
of the time, cycles are being considered as one procedure.</p>
<p>
</p>
<a name="node_sec_3.7"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.7">3.7&nbsp;&nbsp;Settings</a></h2>
<p></p>
<p>
There are a number of settings that control the behavior of the
command processor; most of them are booleans.
They can be set using the <tt>,set</tt> and <tt>,unset</tt> commands.</p>
<p>
</p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>,set <i>setting</i> [on | off | literal | ?]</tt><br>
This sets the value of setting <i>setting</i>.
For a boolean setting, the second argument must be <tt>on</tt> or
<tt>off</tt>; it then defaults to <tt>on</tt>.
Otherwise, the value must be a literal, typically a positive number.
If the second argument is <tt>?</tt> the value of <i>setting</i> is
is displayed and not changed.
Doing <tt>,set ?</tt> will display a list of the setting and
their current values.<p>
</p>
</dd><dt></dt><dd><tt>,unset <i>setting</i></tt><br>
<tt>,unset <i>setting</i></tt> is the same as
<tt>,set <i>setting</i> off</tt>.
</dd></dl><p></p>
<p>
The settings are as follows:
</p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>batch</tt> (boolean)<br>
In `batch mode' any error or interrupt that comes up will cause
Scheme&nbsp;48 to exit immediately with a non-zero exit status.  Also,
the command processor doesn't print prompts.  Batch mode is
off by default.<p>
</p>
<p>
</p>
</dd><dt></dt><dd><tt>levels</tt> (boolean)<br>
Enables or disables the automatic pushing of a new command level when
an error, interrupt, or other breakpoint occurs.
When enabled (the default), breakpoints push a new command level,
and &lang;eof&rang; (see above)
or <tt>,reset</tt> is required to return to top level.  The effects of
pushed command levels include:
<ul>
<li><p>a longer prompt
</p>
<li><p>retention of the continuation in effect at the point of error
</p>
<li><p>confusion among some newcomers
</p>
</ul><p>
With <tt>levels</tt> disabled one must issue a
<tt>,push</tt> command immediately
following an error in order to retain the error continuation for
debugging purposes; otherwise the continuation is lost as soon as
the focus object changes.  If you don't know anything about the
available debugging tools, then <tt>levels</tt> might as well be disabled.</p>
<p>
</p>
</dd><dt></dt><dd><tt>break-on-warnings</tt> (boolean)<br>
Enter a new command level when a warning is produced, just as
when an error occurs.  Normally warnings only result in a displayed
message and the program does not stop executing.<p>
</p>
</dd><dt></dt><dd><tt>ask-before-loading</tt> (boolean)<br>
If on, the system will ask before loading modules that are arguments
to the <tt>,open</tt> command.  <tt>Ask-before-loading</tt> is off by
default.
<pre class=verbatim>&gt; ,set ask-before-loading
will ask before loading modules
&gt; ,open random
Load structure random (y/n)? y
&gt;
</pre><p></p>
<p>
</p>
</dd><dt></dt><dd><tt>load-noisily</tt> (boolean)<br>
When on, the system will print out the names of modules and files
as they are loaded.  <tt>Load-noisily</tt> is off by default.
<pre class=verbatim>&gt; ,set load-noisily
will notify when loading modules and files
&gt; ,open random
[random /usr/local/lib/scheme48/big/random.scm]
&gt; 
</pre><p></p>
<p>
</p>
</dd><dt></dt><dd><tt>inline-values</tt> (boolean)<br>
This controls whether or not the compiler is allowed to substitute
variables' values in-line.
When <tt>inline-values</tt> mode is on,
some Scheme procedures will be substituted in-line; when it is off,
none will.
Section&nbsp;<a href="manual-Z-H-3.html#node_sec_2.4">2.4</a>
has more information.<p>
</p>
</dd><dt></dt><dd><tt>inspector-menu-limit</tt> (positive integer)<br>
This controls how many items the displayed portion of the inspector
menu contains at most.  (See Section&nbsp;<a href="#node_sec_3.8">3.8</a>.)<p>
</p>
</dd><dt></dt><dd><tt>inspector-writing-depth</tt> (positive integer)<br>
This controls the maximum depth to which the inspector
prints values.  (See Section&nbsp;<a href="#node_sec_3.8">3.8</a>.)<p>
</p>
</dd><dt></dt><dd><tt>inspector-writing-length</tt> (positive integer)<br>
This controls the maximum length to which the inspector
prints values.  (See Section&nbsp;<a href="#node_sec_3.8">3.8</a>.)<p>
</p>
</dd><dt></dt><dd><tt>condition-writing-depth</tt> (positive integer)<br>
This controls the maximum depth to which conditions
are printed.<p>
</p>
</dd><dt></dt><dd><tt>condition-writing-length</tt> (positive integer)<br>
This controls the maximum length to which conditions
are printed.<p>
</p>
</dd><dt></dt><dd><tt>profiler-interrupt-time</tt> (positive integer)<br>
This controls the time between two profiler sampling interrupts
(in milliseconds, see section <a href="#node_sec_3.6">3.6</a>).<p>
</p>
</dd><dt></dt><dd><tt>profiler-measure-noninstr</tt> (boolean)<br>
When this flag is enabled, call-times will be measured by the sampling process.
When it is disabled, call-times will only be measured when the procedure has been
instrumented (see section <a href="#node_sec_3.6">3.6</a>).<p>
</p>
</dd><dt></dt><dd><tt>trace-writing-length</tt> (positive integer)<br>
This controls the maximum length to which tracing
prints procedure calls.
</dd></dl><p></p>
<p>
</p>
<a name="node_sec_3.8"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.8">3.8&nbsp;&nbsp;Inspection mode</a></h2>
<p></p>
<p>
There is a data inspector available via the <tt>,inspect</tt> and
<tt>,debug</tt> commands.
The inspector is particularly useful with procedures, continuations,
and records.
The command processor can be taken out of inspection mode by
using the <tt>q</tt> command.
When in inspection mode, input that begins with
a letter or digit is read as a command, not as an expression.
To see the value of a variable or number, do <tt>(begin <i>exp</i>)</tt>
or use the <tt>,run <i>exp</i></tt> command.</p>
<p>
In inspection mode the command processor prints out a
menu of selectable components for the current focus object.
To inspect a particular component, just type the corresponding number in
the menu.
That component becomes the new focus object.
For example:
</p>
<pre class=verbatim>&gt; ,inspect '(a (b c) d)
(a (b c) d)

[0] a
[1] (b c)
[2] d
: 1
(b c)

[0] b
[1] c
: 
</pre><p></p>
<p>
When a new focus object is selected the previous one is pushed onto a
stack.
You can pop the stack, reverting to the previous object, with
the <tt>u</tt> command, or use the <tt>stack</tt> command to move to
an earlier object.</p>
<p>
</p>
<p>
Commands useful when in inspection mode:
</p>
<ul>
<li><p><tt>u</tt> (up) pop object stack
</p>
<li><p><tt>m</tt> (more) print more of a long menu
</p>
<li><p><tt>(<tt>...</tt>)</tt> evaluate a form and select result
</p>
<li><p><tt>q</tt> quit
</p>
<li><p><tt>template</tt> select a closure or continuation's template
(Templates are the static components of procedures; these are found
inside of procedures and continuations, and contain the quoted
constants and top-level variables referred to by byte-compiled code.)
</p>
<li><p><tt>d</tt> (down) move to the next continuation
(current object must be a continuation)
</p>
<li><p><tt>menu</tt> print the selection menu for the focus object
</p>
</ul><p></p>
<p>
Multiple selection commands (<tt>u</tt>, <tt>d</tt>, and menu indexes)
may be put on a single line.</p>
<p>
</p>
<p>
All ordinary commands are available when in inspection mode.
Similarly, the inspection commands can be used when not in inspection
mode.
For example:
</p>
<pre class=verbatim>&gt; (list 'a '(b c) 'd)
'(a (b c) d)
&gt; ,1
'(b c)
&gt; ,menu
[0] b
[1] c
&gt; 
</pre><p></p>
<p>
If the current command level was initiated because of
a breakpoint in the next level down, then 
<tt>,debug</tt> will invoke the inspector on the
continuation at the point of the error.
The <tt>u</tt> and <tt>d</tt> (up and down)
commands then make the inspected-value stack look like a conventional stack
debugger, with continuations playing the role of stack frames.  <tt>D</tt> goes
to older or deeper continuations (frames), and <tt>u</tt> goes back up to more
recent ones.</p>
<p>
</p>
<a name="node_sec_3.9"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.9">3.9&nbsp;&nbsp;Command programs</a></h2>
<p></p>
<p>
The <tt>exec</tt> package contains procedures that are used
to execute the command processor's commands.
A command <tt>,<i>foo</i></tt> is executed by applying the value of
the identifier <i>foo</i> in the <tt>exec</tt> package to
the (suitably parsed) command arguments.</p>
<p>
</p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>,exec [<i>command</i>]</tt><br>
Evaluate <i>command</i> in the <tt>exec</tt> package.
For example, use
<pre class=verbatim>,exec ,load <i>filename</i>
</pre><p>
to load a file containing commands.
If no <i>command</i> is given, the <tt>exec</tt> package becomes the
execution package for future commands.
</p>
</dd></dl><p></p>
<p>
The required argument types are as follows:
</p>
<ul>
<li><p>filenames should be strings
</p>
<li><p>other names and identifiers should be symbols
</p>
<li><p>expressions should be s-expressions
</p>
<li><p>commands (as for <tt>,config</tt> and <tt>,exec</tt> itself)
should be lists of the form
<tt>(<i>command-name</i> <i>argument</i> <tt>...</tt>)</tt>
where <i>command-name</i> is a symbol.
</p>
</ul><p></p>
<p>
For example, the following two commands are equivalent:
</p>
<pre class=verbatim>,config ,load my-file.scm

,exec (config '(load &quot;my-file.scm&quot;))
</pre><p></p>
<p>
The file <tt>scheme/vm/load-vm.scm</tt> in the source directory contains an
example of an <tt>exec</tt> program.</p>
<p>
</p>
<a name="node_sec_3.10"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.10">3.10&nbsp;&nbsp;Building images</a></h2>
<p></p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>,dump <i>filename</i> [<i>identification</i>]</tt><br>
Writes the current heap out to a file, which can then be run using the
virtual machine.  The new image file includes the command processor.
If present, <i>identification</i>
should be a string (written with double quotes); this string will
be part of the greeting message as the image starts up.<p>
</p>
</dd><dt></dt><dd><tt>,build <i>exp</i> <i>filename</i> [<i>option <tt>...</tt></i>]</tt><br>
Like <tt>,dump</tt>, except that the image file contains the value of
<i>exp</i>, which should be a procedure of one argument, instead of
the command processor.  When
<i>filename</i> is resumed, that procedure will be invoked on the VM's
<tt>-a</tt> arguments, which are passed as a list of OS strings (see
section&nbsp;<a href="manual-Z-H-6.html#node_sec_5.15">5.15</a>.  The
procedure should return an integer, which is
returned to the program that invoked the VM.  The command
processor and debugging system are not included in the image
(unless you go to some effort to preserve them, such as retaining
a continuation).<p>
If <tt>no-warnings</tt> appears as an <i>option</i> after the file
name, no warnings about undefined external bindings
(see Section&nbsp;<a href="manual-Z-H-9.html#node_sec_8.2">8.2</a>) will be printed upon
resuming the image.  This is useful when the definitions of
external bindings appear in shared objects that are only loaded
after the resumption of the image.</p>
<p>
Doing <tt>,flush</tt> before building an image will reduce the amount
of debugging information in the image, making for a smaller
image file, but if an error occurs, the error message may be less
helpful.  Doing <tt>,flush source maps</tt> before loading any programs
used in the image will make it still smaller.
See section&nbsp;<a href="#node_sec_3.11">3.11</a>
for more information.</p>
<p>
</p>
</dd></dl><p></p>
<p>
</p>
<a name="node_sec_3.11"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.11">3.11&nbsp;&nbsp;Resource query and control</a></h2>
<p></p>
<p>
</p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>,time <i>exp</i></tt><br>
Measure execution time.<p>
</p>
</dd><dt></dt><dd><tt>,collect</tt><br>
Invoke the garbage collector.  Ordinarily this happens
automatically, but the command tells how much space is available
before and after the collection.<p>
</p>
</dd><dt></dt><dd><tt>,keep <i>kind</i></tt>
</dd><dt></dt><dd><tt>,flush <i>kind</i></tt><br>
These control the amount of debugging information retained after
compiling procedures.  This information can consume a fair amount
of space.  <i>kind</i> is one of the following:
<ul>
<li><p><tt>maps</tt> - environment maps (local variable names, for inspector)
</p>
<li><p><tt>source</tt> - source code for continuations (displayed by inspector)
</p>
<li><p><tt>names</tt> - procedure names (as displayed by <tt>write</tt> and in error
messages)
</p>
<li><p><tt>files</tt>  - source file names
</p>
</ul><p>
These commands refer to future compilations only, not to procedures
that already exist.  To have any effect, they must be done before
programs are loaded.  The default is to keep all four types.
</p>
<p>
</p>
</dd><dt></dt><dd><tt>,flush</tt><br>
The flush command with no argument deletes the database of names
of initial procedures.  Doing <tt>,flush</tt> before a <tt>,build</tt> or
<tt>,dump</tt>
will make the resulting image significantly smaller, but will
compromise the information content of many error
messages.
</dd></dl><p></p>
<p>
</p>
<a name="node_sec_3.12"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.12">3.12&nbsp;&nbsp;Threads</a></h2>
<p></p>
<p>
Each command level has its own set of threads.  These threads are suspended
when a new level is entered and resumed when the owning level again becomes
the current level.
A thread that raises an error is not resumed unless
explicitly restarted using the <tt>,proceed</tt> command.
In addition to any threads spawned by the user, each level has a thread
that runs the command processor on that level.
A new command-processor thread is started if the current one
dies or is terminated.
When a command level is abandoned for a lower level, or when
a level is restarted using <tt>,reset</tt>, all of the threads on that
level are terminated and any <tt>dynamic-wind</tt> &ldquo;after&rdquo; thunks are run.</p>
<p>
The following commands are useful when debugging multithreaded programs:
</p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>,resume [<i>number</i>]</tt><br>
Pops out to a given level and resumes running all threads at that level.
<i>Number</i> defaults to zero.<p>
</p>
</dd><dt></dt><dd><tt>,threads</tt><br>
Invokes the inspector on a list of the threads running at the
next lower command level.<p>
</p>
</dd><dt></dt><dd><tt>,exit-when-done [<i>exp</i>]</tt><br>
Waits until all user threads have completed and then
exits back out to shell (or executive or whatever invoked Scheme&nbsp;48
in the first place).
<i>Exp</i> should evaluate to an integer which is then
returned to the calling program.
<p>
</p>
<p>
</p>
</dd></dl><p></p>
<p>
</p>
<a name="node_sec_3.13"></a>
<h2 class=section><a href="manual-Z-H-1.html#node_toc_node_sec_3.13">3.13&nbsp;&nbsp;Quite obscure</a></h2>
<p></p>
<dl><dt></dt><dd>
</dd><dt></dt><dd><tt>,go <i>exp</i></tt><br>
This is like <tt>,exit <i>exp</i></tt> except that the evaluation of <i>exp</i>
is tail-recursive with respect to the command processor.  This
means that the command processor itself can probably be GC'ed,
should a garbage collection occur in the execution of <i>exp</i>.
If an error occurs Scheme&nbsp;48 will exit with a non-zero value.<p>
</p>
</dd><dt></dt><dd><tt>,translate <i>from</i> <i>to</i></tt><br>
For <tt>load</tt> and the <tt>,load</tt> command
(but not for <tt>open-{in|out}put-file</tt>), file
names beginning with the string <i>from</i> will be changed so that the
initial <i>from</i> is replaced by the string <i>to</i>.  E.g.
<pre class=verbatim><tt>,translate /usr/gjc/ /zu/gjc/</tt>
</pre><p>
will cause <tt>(load &quot;/usr/gjc/foo.scm&quot;)</tt> to have the same effect as
<tt>(load &quot;/zu/gjc/foo.scm&quot;)</tt>.
</p>
<p>
</p>
</dd><dt></dt><dd><tt>,from-file <i>filename</i> <i>form</i> <tt>...</tt> ,end</tt><br>
This is used by the <tt>cmuscheme48</tt> Emacs library to indicate the file
from which the <i>form</i>s came.  <i>Filename</i> is then used by the
command processor to determine the package in which the <i>form</i>s
are to be evaluated.
</dd></dl><p></p>
<p>

</p>
<p>
</p>
<p>
</p>
<div class=footnoterule><hr></div><p></p>
<div class=footnote><p><a name="node_footnote_Temp_3"></a><sup><small><a href="#node_call_footnote_Temp_3">1</a></small></sup> part of the GNU Binary Utilities: http://www.gnu.org/software/binutils/</p>
</div>
<div class=smallskip></div>
<p style="margin-top: 0pt; margin-bottom: 0pt">
<div align=right class=navigation>[Go to <span><a href="manual.html">first</a>, <a href="manual-Z-H-3.html">previous</a></span><span>, <a href="manual-Z-H-5.html">next</a></span> page<span>; &nbsp;&nbsp;</span><span><a href="manual-Z-H-1.html#node_toc_start">contents</a></span><span><span>; &nbsp;&nbsp;</span><a href="manual-Z-H-11.html#node_index_start">index</a></span>]</div>
</p>
<p></p>
</div>
</body>
</html>
scheme48-doc 1.9-3 / usr / share / doc / scheme48 / html / manual-Z-H-4.html
