mirrors/emacs
1
Fork 0
mirror of git://git.sv.gnu.org/emacs.git synced 2026-01-26 07:00:35 -08:00
Code Issues Projects Releases Wiki Activity
emacs/mps/manual/html/todo.html
Gareth Rees bd034b6eb0 Update html. 
Copied from Perforce
 Change: 180468
 ServerID: perforce.ravenbrook.com
2012-11-14 14:13:19 +00:00

1032 lines
No EOL
79 KiB
HTML
Raw Blame History

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<title>To do &mdash; Memory Pool System 1.111.0 documentation</title>
<link rel="stylesheet" href="_static/mps.css" type="text/css" />
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<script type="text/javascript">
var DOCUMENTATION_OPTIONS = {
URL_ROOT: '',
VERSION: '1.111.0',
COLLAPSE_INDEX: false,
FILE_SUFFIX: '.html',
HAS_SOURCE: true
};
</script>
<script type="text/javascript" src="_static/jquery.js"></script>
<script type="text/javascript" src="_static/underscore.js"></script>
<script type="text/javascript" src="_static/doctools.js"></script>
<link rel="copyright" title="Copyright" href="copyright.html" />
<link rel="top" title="Memory Pool System 1.111.0 documentation" href="index.html" />
<link rel="prev" title="Contact us" href="contact.html" />
</head>
<body>
<div class="related">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="genindex.html" title="General Index"
accesskey="I">index</a></li>
<li class="right" >
<a href="contact.html" title="Contact us"
accesskey="P">previous</a> |</li>
<li><a href="index.html">Memory Pool System 1.111.0 documentation</a> &raquo;</li>
</ul>
</div>
<div class="document">
<div class="documentwrapper">
<div class="bodywrapper">
<div class="body">
<div class="section" id="to-do">
<h1>To do<a class="headerlink" href="#to-do" title="Permalink to this headline"></a></h1>
<div class="section" id="outstanding">
<h2>Outstanding<a class="headerlink" href="#outstanding" title="Permalink to this headline"></a></h2>
<ol class="arabic simple" start="3">
<li>Glossary entries need permalink markers. See Sphinx <a class="reference external" href="https://bitbucket.org/birkenfeld/sphinx/issue/996/expose-glossary-entry-link-on-hover">issue 996</a>.</li>
</ol>
<ol class="arabic simple" start="5">
<li>Fix the <a class="reference internal" href="genindex.html"><em>Index</em></a> so that glossary entries like &#8220;byte (2)&#8221;
aren&#8217;t interpreted as subentries? (Or is it OK as is?)</li>
</ol>
<ol class="arabic simple" start="9">
<li>Hyphenate long function names across line endings. It would be
possible to do this by inserting soft hyphens but if you copied
the function name you&#8217;d copy the soft hyphens too.</li>
</ol>
<ol class="arabic simple" start="11">
<li>Support MMREF-style anchors to the glossary (<tt class="docutils literal"><span class="pre">#garbage.collection</span></tt>
as well as <tt class="docutils literal"><span class="pre">#garbage-collection</span></tt>).</li>
</ol>
<ol class="arabic" start="37">
<li><p class="first">Need a <tt class="docutils literal"><span class="pre">scheme-advanced.c</span></tt> to illustrate advanced topics:</p>
<ul class="simple">
<li>Segregate objects into multiple pools (e.g. strings and integers
into <a class="reference internal" href="pool/amcz.html#pool-amcz"><em>AMCZ (Automatic Mostly-Copying Zero-rank)</em></a>). [DONE]</li>
<li>Global symbol table under MPS management. [DONE]</li>
<li>Tagged references.</li>
<li>Unboxed values.</li>
<li>&#8220;Bumming cycles out of scan&#8221; using tags.</li>
<li>Weak hashtable. [DONE]</li>
</ul>
<p>Compare the performance with the plain <tt class="docutils literal"><span class="pre">scheme.c</span></tt>. The advanced
version better be faster!</p>
</li>
</ol>
<ol class="arabic simple" start="76">
<li>Develop debugging examples that illustrate each of the warnings
about consistency. Also to demonstrate the &#8220;garbage collect
frequently to expose errors as soon as possible&#8221; advice.</li>
</ol>
<ol class="arabic" start="124">
<li><p class="first">In <a class="reference internal" href="guide/debug.html#guide-debug"><em>Debugging with the Memory Pool System</em></a>, we need to talk about other debuggers.</p>
<p>In particular, you can&#8217;t really work with the MPS with LLDB on
the Mac yet. Apple knows about this.</p>
<p>There are a couple of magic commands to type when using WinDbg to
get it to pass exceptions. We should document those and things
that need doing with Visual Studio.</p>
</li>
<li><p class="first">Need overview for people who already know about garbage
collection. (RB needs to write this.)</p>
</li>
<li><p class="first">Things that are &#8220;within reach&#8221;, i.e. that we could do if people
needed them. (RB needs to write this.)</p>
</li>
</ol>
</div>
<div class="section" id="complete">
<h2>Complete<a class="headerlink" href="#complete" title="Permalink to this headline"></a></h2>
<ol class="arabic">
<li><p class="first">Create a new <a class="reference external" href="http://sphinx.pocoo.org/ext/appapi.html#sphinx.domains.Domain">domain</a> for MPS stuff.</p>
</li>
<li><p class="first">Create new types of Sphinx objects for pools and topics.</p>
<p><em>Action:</em> doesn&#8217;t seem necessary. Ordinary cross-references work fine.</p>
</li>
</ol>
<ol class="arabic simple" start="4">
<li>Make &#8220;(1)&#8221;, &#8220;(2)&#8221; into superscripts.</li>
</ol>
<ol class="arabic" start="6">
<li><p class="first">Identify <tt class="docutils literal"><span class="pre">:term:</span></tt> references that point to &#8220;see&#8221; cross-references.</p>
</li>
<li><p class="first">Re-do the diagrams in vector form and using the colour palette.</p>
</li>
<li><p class="first">Use a better bibliography extension and update the bibliography.
See for example <a class="reference external" href="http://wnielson.bitbucket.org/projects/sphinx-natbib/">sphinx-natbib</a>.</p>
<p>I tried downloading and using this but it&#8217;s clearly not ready yet.
It uses Pybtex to parse BibTeX, but Pybtex is buggy (its
<tt class="docutils literal"><span class="pre">CaseInsensitiveDict</span></tt> class can&#8217;t be serialized).</p>
<p>We could do this &#8220;by hand&#8221; as it were by using normal Sphinx
cross-references. For example:</p>
<div class="highlight-c"><div class="highlight"><pre><span class="o">*</span> <span class="p">..</span> <span class="n">_WIL95</span><span class="o">:</span>
<span class="n">Paul</span> <span class="n">R</span><span class="p">.</span> <span class="n">Wilson</span><span class="p">,</span> <span class="n">Mark</span> <span class="n">S</span><span class="p">.</span> <span class="n">Johnstone</span><span class="p">,</span> <span class="n">Michael</span> <span class="n">Neely</span><span class="p">,</span> <span class="n">David</span>
<span class="n">Boles</span><span class="p">.</span> <span class="mf">1995.</span> <span class="s">&quot;Dynamic Storage Allocation: A Survey and</span>
<span class="n">Critical</span> <span class="n">Review</span><span class="s">&quot;.</span>
</pre></div>
</div>
</li>
</ol>
<ol class="arabic simple" start="10">
<li>Pluralize &#8220;Topic&#8221; to &#8220;Topics&#8221; and so on.</li>
</ol>
<ol class="arabic" start="12">
<li><p class="first">Superscripts in aka sections.</p>
</li>
<li><p class="first">Does the object format description (<a class="reference internal" href="topic/format.html#mps_fmt_A_s" title="mps_fmt_A_s"><tt class="xref c c-type docutils literal"><span class="pre">mps_fmt_A_s</span></tt></a> and so
on) have to be static as in the Scheme example? Or is it OK to
throw it away after calling <tt class="xref c c-func docutils literal"><span class="pre">mps_fmt_create()</span></tt>?</p>
<p>Same question for the generation structure
(<a class="reference internal" href="topic/collection.html#mps_gen_param_s" title="mps_gen_param_s"><tt class="xref c c-type docutils literal"><span class="pre">mps_gen_param_s</span></tt></a>) for a pool and the size classes
(<a class="reference internal" href="topic/cache.html#mps_sac_class_s" title="mps_sac_class_s"><tt class="xref c c-type docutils literal"><span class="pre">mps_sac_class_s</span></tt></a>) for a segregated allocation cache?</p>
<p><em>Answer:</em> examination of the code shows that all of these structures
are copied into internally allocated MPS control structures, and
so it is OK for the client program to put these structures on the
stack otherwise dispose of them.</p>
</li>
<li><p class="first">What is the difference, if any, between <a class="reference internal" href="topic/interface.html#mps_word_t" title="mps_word_t"><tt class="xref c c-type docutils literal"><span class="pre">mps_word_t</span></tt></a> and
<a class="reference internal" href="topic/platform.html#MPS_T_WORD" title="MPS_T_WORD"><tt class="xref c c-macro docutils literal"><span class="pre">MPS_T_WORD</span></tt></a>?</p>
<p><em>Answer:</em> <a class="reference internal" href="topic/platform.html#MPS_T_WORD" title="MPS_T_WORD"><tt class="xref c c-macro docutils literal"><span class="pre">MPS_T_WORD</span></tt></a> comes from <tt class="docutils literal"><span class="pre">mpstd.h</span></tt> which
contains no C code (only macro definitions). It used to be the
case that <tt class="docutils literal"><span class="pre">mpstd.h</span></tt> was included in <tt class="docutils literal"><span class="pre">mps.h</span></tt> but it isn&#8217;t any
more? There&#8217;s something to do here: why don&#8217;t we include it?
Because <tt class="docutils literal"><span class="pre">mpstd.h</span></tt> is a bit picky and could easily be
broken. <a class="reference internal" href="topic/platform.html#MPS_T_WORD" title="MPS_T_WORD"><tt class="xref c c-macro docutils literal"><span class="pre">MPS_T_WORD</span></tt></a> could be eliminated from the
external interface.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003315/">job003315</a>.</p>
</li>
<li><p class="first">How can I explain why the Scheme example uses
<tt class="docutils literal"><span class="pre">sizeof(mps_word_t)</span></tt> as its alignment? Why not
<a class="reference internal" href="topic/platform.html#MPS_PF_ALIGN" title="MPS_PF_ALIGN"><tt class="xref c c-macro docutils literal"><span class="pre">MPS_PF_ALIGN</span></tt></a> (or are client programs not supposed to
look at <tt class="docutils literal"><span class="pre">mpstd.h</span></tt>)? Why not something of its own manufacture, like
<tt class="docutils literal"><span class="pre">sizeof(union</span> <span class="pre">{long,</span> <span class="pre">size_t,</span> <span class="pre">void*})</span></tt>?</p>
<p><em>Answer:</em> This a puzzle! More thought needed.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003316/">job003316</a>.</p>
</li>
<li><p class="first">Why does the Scheme example have a copy method in its object
format when the reference manual says it&#8217;s obsolete?</p>
<p><em>Answer:</em> It wasn&#8217;t obsolete when it was written. I removed it.</p>
</li>
<li><p class="first">What is the difference between the &#8220;event stream&#8221; and the
&#8220;telemetry stream&#8221;? Are these names for the same thing? Or is
there a distinction (for example, &#8220;event stream&#8221; refers to the
internal, unfiltered, stream of events and &#8220;telemetry stream&#8221;
refers to the filtered stream)?</p>
<p><em>Answer:</em> the event stream is the implementation of the telemetry
stream, so the user doc can refer to &#8220;telemetry stream&#8221;.</p>
</li>
<li><p class="first">The location dependency functions all take an arena as an
argument. What is the role of this argument?</p>
<p><em>Answer:</em> if you have multiple arenas, you are going to have to
have an <a class="reference internal" href="topic/location.html#mps_ld_t" title="mps_ld_t"><tt class="xref c c-type docutils literal"><span class="pre">mps_ld_t</span></tt></a> for each arena, and add each address
you are interested in for each arena. (If you happen to know which
arena it&#8217;s in, you can just specify it for that arena.) So for
isstale, it better be the same arena as add.</p>
</li>
<li><p class="first">What is the role of the third (<tt class="docutils literal"><span class="pre">addr</span></tt>) argument to
<a class="reference internal" href="topic/location.html#mps_ld_isstale" title="mps_ld_isstale"><tt class="xref c c-func docutils literal"><span class="pre">mps_ld_isstale()</span></tt></a>? <tt class="docutils literal"><span class="pre">LDIsStale</span></tt> says <tt class="docutils literal"><span class="pre">UNUSED(addr);</span></tt> so
maybe it is unused.</p>
<p><em>Answer:</em> in theory it&#8217;s the address you want to check, but in the
implementation it tells you for all addresses. The LD functions
have an intention that is not quite the same as the design
documentation. (But what then, asks RB, is the point of
<a class="reference internal" href="topic/location.html#mps_ld_add" title="mps_ld_add"><tt class="xref c c-func docutils literal"><span class="pre">mps_ld_add()</span></tt></a>?) Perhaps in the case of
<a class="reference internal" href="topic/location.html#mps_ld_isstale" title="mps_ld_isstale"><tt class="xref c c-func docutils literal"><span class="pre">mps_ld_isstale()</span></tt></a> it&#8217;s &#8220;a piece of information that might
be useful for debugging&#8221;.</p>
</li>
<li><p class="first">Is the material in the pool class comparison table at all accurate?</p>
<p><em>Answer:</em> It will be better to have a flowchart approach rather than
a table of properties.</p>
</li>
<li><p class="first">This code seems a bit confused about what to do:</p>
<div class="highlight-c"><div class="highlight"><pre><span class="n">assert</span><span class="p">(</span><span class="mi">0</span><span class="p">);</span>
<span class="n">fprintf</span><span class="p">(</span><span class="n">stderr</span><span class="p">,</span> <span class="s">&quot;Unexpected object on the heap</span><span class="se">\n</span><span class="s">&quot;</span><span class="p">);</span>
<span class="n">abort</span><span class="p">();</span>
<span class="k">return</span> <span class="n">MPS_RES_FAIL</span><span class="p">;</span>
</pre></div>
</div>
<p>What should it say?</p>
<p><em>Answer:</em> the assertion and the return seem to be bogus, so I
removed them.</p>
</li>
<li><p class="first">How does fixing interact with tagged references? Do I need to
remove the tag before fixing a reference? Do I need to restore the
tag afterwards? I thought that both would be necessary but the
<a class="reference external" href="https://info.ravenbrook.com/project/mps/master/design/critical-path.txt">critical path</a> documentation has an example from OpenDylan with
tagged references that does neither:</p>
<p><em>Answer:</em> we&#8217;ll document that all references need to be
decrypted/de-tagged. There ought to be some slack in practice but
it needs thought.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003317/">job003317</a>.</p>
</li>
<li><p class="first">This code from <tt class="docutils literal"><span class="pre">mps_chat</span></tt> in the Scheme example is wrong:</p>
<div class="highlight-c"><div class="highlight"><pre><span class="k">if</span> <span class="p">(</span><span class="n">type</span> <span class="o">==</span> <span class="n">mps_message_type_gc_start</span><span class="p">())</span> <span class="p">{</span>
<span class="n">printf</span><span class="p">(</span><span class="s">&quot;Collection %lu started.</span><span class="se">\n</span><span class="s">&quot;</span><span class="p">,</span> <span class="p">(</span><span class="kt">unsigned</span> <span class="kt">long</span><span class="p">)</span><span class="n">mps_collections</span><span class="p">(</span><span class="n">arena</span><span class="p">));</span>
</pre></div>
</div>
<p><a class="reference internal" href="topic/arena.html#mps_collections" title="mps_collections"><tt class="xref c c-func docutils literal"><span class="pre">mps_collections()</span></tt></a> returns the total number of collections
to date, not the number of the collection that posted the
message. This means that if there have been multiple collections
since the last time the message queue was emptied, the output will
look like this:</p>
<div class="highlight-c"><div class="highlight"><pre><span class="n">Collection</span> <span class="mi">47</span> <span class="n">started</span><span class="p">.</span>
<span class="p">...</span>
<span class="n">Collection</span> <span class="mi">47</span> <span class="n">started</span><span class="p">.</span>
<span class="p">...</span>
<span class="n">Collection</span> <span class="mi">47</span> <span class="n">started</span><span class="p">.</span>
<span class="p">...</span>
</pre></div>
</div>
<p><em>Answer:</em> mis-feature. Update the scheme example.</p>
<p><em>Action:</em> updated the Scheme example. Made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003318/">job003318</a>.</p>
</li>
<li><p class="first">It seems &#8220;tricky&#8221; to re-use fowarding objects as padding objects
by setting their forwarding pointer to <tt class="docutils literal"><span class="pre">NULL</span></tt>. Wouldn&#8217;t it be
simpler to explain if we had <tt class="docutils literal"><span class="pre">TYPE_PAD</span></tt> for multiple-word
padding objects? Things are difficult enough to explain as it is!</p>
<p><em>Answer:</em> It would be simpler, so I made this change.</p>
</li>
<li><p class="first">The Scheme example says, &#8220;Adapting it to use the MPS took
approximately two hours&#8221;. I doubt this would be the common case,
and it would be better to under-promise here and over-deliver.</p>
<p><em>Answer:</em> take it out.</p>
</li>
<li><p class="first">I need to document the values in <a class="reference internal" href="topic/collection.html#mps_gen_param_s" title="mps_gen_param_s"><tt class="xref c c-type docutils literal"><span class="pre">mps_gen_param_s</span></tt></a>. I
believe they are the capacity (size of the generation in
kilobytes) and the mortality (the proportion of objects in this
generation that are expected to die in a collection). But what do
they mean to the MPS? And how should one go about choosing values?</p>
<p><em>Answer:</em> requires a whole document. Talk to RB. The intro
document should say, &#8220;take these values on trust&#8221;. There&#8217;s a
design document about this based on the <a class="reference external" href="https://info.ravenbrook.com/project/mps/doc/2002-06-18/obsolete-mminfo/mminfo/strategy/lisp-machine/">Lisp Machine</a>.</p>
</li>
<li><p class="first">Wouldn&#8217;t it make <a class="reference internal" href="pool/amc.html#mps_amc_apply" title="mps_amc_apply"><tt class="xref c c-func docutils literal"><span class="pre">mps_amc_apply()</span></tt></a> easier to document if
there were a typedef for the stepper function type, something like
this:</p>
<div class="highlight-c"><div class="highlight"><pre><span class="k">typedef</span> <span class="kt">void</span> <span class="p">(</span><span class="o">*</span><span class="n">mps_amc_apply_stepper_t</span><span class="p">)(</span><span class="n">mps_addr_t</span> <span class="n">object</span><span class="p">,</span> <span class="kt">void</span> <span class="o">*</span><span class="n">p</span><span class="p">,</span> <span class="kt">size_t</span> <span class="n">s</span><span class="p">)</span>
</pre></div>
</div>
<p><em>Answer:</em> RB says I can make this change since it&#8217;s
backwards-compatible.</p>
</li>
<li><p class="first">Wouldn&#8217;t the Scheme example be better without TAB characters?</p>
<p><em>Answer:</em> maybe, but it would lead to merge conflicts. So no change.</p>
</li>
<li><p class="first">The example code looks better (easier to see the structure) if I
use an indentation of four spaces. There are also cases where the
original code is inconsistent (compare the indentation of the case
labels in <tt class="docutils literal"><span class="pre">print</span></tt> versus <tt class="docutils literal"><span class="pre">obj_scan</span></tt>). I&#8217;ve made these consistent
in the user guide.</p>
<p>Should these changes be propagated back to the example code?</p>
<p><em>Answer:</em> no.</p>
</li>
<li><p class="first">The Scheme example is inconsistent in its use of whitespace: for
example sometimes there&#8217;s a space after <tt class="docutils literal"><span class="pre">if</span></tt> and sometimes not.</p>
<p><em>Answer:</em> maybe so, but it would lead to merge conflicts. So no
change.</p>
</li>
<li><p class="first">What is a root mode and how do I explain it?</p>
<p><em>Answer:</em> currently a root mode has no effect. It tells the MPS
whether it&#8217;s OK for the MPS to put a barrier on the
root. Recommend pass zero. For future expansion.</p>
</li>
<li><p class="first">You create a marker on the stack and pass it to
<a class="reference internal" href="topic/root.html#mps_root_create_reg" title="mps_root_create_reg"><tt class="xref c c-func docutils literal"><span class="pre">mps_root_create_reg()</span></tt></a> to tell it where the bottom of the
stack is. Fine. But then you are supposed to call your program
via <a class="reference internal" href="topic/thread.html#mps_tramp" title="mps_tramp"><tt class="xref c c-func docutils literal"><span class="pre">mps_tramp()</span></tt></a>. If the MPS is trampolining your whole
program, why does the MPS need your help to work out where the
stack is? It could work it out for itself surely?</p>
<p>I guess this is some kind of separation of concerns &#8211;
<a class="reference internal" href="topic/thread.html#mps_tramp" title="mps_tramp"><tt class="xref c c-func docutils literal"><span class="pre">mps_tramp()</span></tt></a> is something to do with protection &#8211; but it&#8217;s
hard for me to explain.</p>
<p><em>Answer:</em> <a class="reference internal" href="topic/thread.html#mps_tramp" title="mps_tramp"><tt class="xref c c-func docutils literal"><span class="pre">mps_tramp()</span></tt></a> is needed for your program to work
on Windows because a barrier hit generates a Structured Exception
which has to be caught by a handler higher up the stack.</p>
</li>
<li><p class="first">Status. At what point will the work be &#8220;good enough&#8221; to merge back
to the master sources?</p>
<p><em>Answer:</em> After RB reviews it.</p>
</li>
<li><p class="first">The generic example of using <a class="reference internal" href="topic/thread.html#mps_tramp" title="mps_tramp"><tt class="xref c c-func docutils literal"><span class="pre">mps_tramp()</span></tt></a> needs to pass
<tt class="docutils literal"><span class="pre">argv</span></tt> and <tt class="docutils literal"><span class="pre">argc</span></tt>, and return an exit code, so maybe it would
make sense to do that in the Scheme example, even though Scheme
doesn&#8217;t use these parameters.</p>
<p><em>Answer:</em> no.</p>
</li>
<li><p class="first">There&#8217;s a lot of stuff to explain here, and I think some of it
could be simplified:</p>
<ol class="loweralpha simple">
<li>The common trampoline case (passing <tt class="docutils literal"><span class="pre">argv</span></tt> and <tt class="docutils literal"><span class="pre">argc</span></tt>, and
returning an exit code).</li>
<li>The common single-thread root registration case (e.g. putting a
marker on the stack in <a class="reference internal" href="topic/thread.html#mps_tramp" title="mps_tramp"><tt class="xref c c-func docutils literal"><span class="pre">mps_tramp()</span></tt></a>).</li>
</ol>
<p><em>Answer:</em> leave it as it is.</p>
</li>
<li><p class="first">The discussion in the Scheme example about <a class="reference internal" href="topic/allocation.html#mps_reserve" title="mps_reserve"><tt class="xref c c-func docutils literal"><span class="pre">mps_reserve()</span></tt></a>
suggests that <a class="reference internal" href="topic/allocation.html#mps_alloc" title="mps_alloc"><tt class="xref c c-func docutils literal"><span class="pre">mps_alloc()</span></tt></a> doesn&#8217;t require aligned
sizes. Is that right? Needs to be added to <a class="reference internal" href="topic/allocation.html#mps_alloc" title="mps_alloc"><tt class="xref c c-func docutils literal"><span class="pre">mps_alloc()</span></tt></a>
reference if so.</p>
<p><em>Answer:</em> in fact there&#8217;s no rule about this. Depends on the pool
class. &#8220;It doesn&#8217;t unless the pool class says it does&#8221;.</p>
</li>
</ol>
<ol class="arabic" start="38">
<li><p class="first">Document about interface conventions and interface policies. What
do we guarantee about support for the external symbols?</p>
<p><em>Action:</em> added new topic <a class="reference internal" href="topic/interface.html#topic-interface"><em>Interface conventions</em></a>.</p>
</li>
<li><p class="first">We don&#8217;t support scanning the stack/registers except via
mps_stack_scan_ambig? Document this?</p>
<p><em>Answer:</em> this is the only one we support at the moment.</p>
</li>
<li><p class="first">The <tt class="docutils literal"><span class="pre">fragmentation_limit</span></tt> argument to <a class="reference internal" href="pool/mvt.html#mps_class_mvt" title="mps_class_mvt"><tt class="xref c c-func docutils literal"><span class="pre">mps_class_mvt()</span></tt></a> is
an integer representing a percentage between 1 and 100. For
consistency with mps_gen_param_s this should be a double between 0
(exclusive) and 1. Can we change this?</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003319/">job003319</a>.</p>
</li>
<li><p class="first">Move symbol reference from <tt class="docutils literal"><span class="pre">mpsio.h</span></tt>, <tt class="docutils literal"><span class="pre">mpstd.h</span></tt> and
<tt class="docutils literal"><span class="pre">mpslib.h</span></tt> to <a class="reference internal" href="topic/plinth.html#topic-plinth"><em>Plinth</em></a>.</p>
</li>
<li><p class="first">Move symbol references for the pool classes to the corresponding
pool document.</p>
</li>
<li><p class="first">In the &#8220;choosing a pool&#8221; procedure there&#8217;s no mention of ambiguous
references. I omitted them because the NB/RIT chart of pool
properties seemed to suggest that no pools can contain ambiguous
references. But is this right? Couldn&#8217;t you allocate your block
containing ambiguous references in a non-moving, non-scanning pool
like MVT and register it as an ambiguous root?</p>
<p><em>Answer:</em> Ought to be a note to explain that there may more thing to
take into account for experts. &#8220;For beginners&#8221;. Ambiguous
references not currently supported via scanning/automatic pools
etc: use workaround.</p>
</li>
<li><p class="first">In the &#8220;choosing a pool&#8221; procedure there&#8217;s no mention of
protection. Can we subsume this under &#8220;moving&#8221;?</p>
<p><em>Answer:</em> Use &#8220;movable and protectable&#8221;.</p>
</li>
<li><p class="first">Does <a class="reference internal" href="topic/arena.html#mps_arena_step" title="mps_arena_step"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_step()</span></tt></a> offer any guarantees about how long
it will pause for? (I presume not: I&#8217;ve written &#8220;makes every
effort to return within interval seconds, but does not guarantee
to do so&#8221;.)</p>
<p><em>Answer:</em> No guarantee. &#8220;Since it calls your scanning code...&#8221;</p>
</li>
<li><p class="first">RHSK&#8217;s documentation for <a class="reference internal" href="topic/arena.html#mps_arena_step" title="mps_arena_step"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_step()</span></tt></a> says &#8220;Note that
<a class="reference internal" href="topic/arena.html#mps_arena_step" title="mps_arena_step"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_step()</span></tt></a> will still step, even if the arena has
been clamped. This is to allow a client to advance a collection
only at these mps_arena_step points (but note that barrier hits
will also cause collection work).&#8221; However, looking at the
implementation, it seems to me that if <a class="reference internal" href="topic/arena.html#mps_arena_step" title="mps_arena_step"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_step()</span></tt></a>
does anything, then it releases the arena. Is the client program
expected to call <a class="reference internal" href="topic/arena.html#mps_arena_clamp" title="mps_arena_clamp"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_clamp()</span></tt></a> after
<a class="reference internal" href="topic/arena.html#mps_arena_step" title="mps_arena_step"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_step()</span></tt></a> returns, if they want to support this use
case?</p>
<p>Similar question if the arena is in the parked state.</p>
<p><em>Answer:</em> this is a bug.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003320/">job003320</a>.</p>
</li>
<li><p class="first">How does <a class="reference internal" href="topic/arena.html#mps_arena_start_collect" title="mps_arena_start_collect"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_start_collect()</span></tt></a> interact with the arena
state? (It seems from looking at the implementation that it puts
the arena into the unclamped state, like
<a class="reference internal" href="topic/arena.html#mps_arena_release" title="mps_arena_release"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_release()</span></tt></a>.)</p>
<p><em>Answer:</em> this is fine.</p>
</li>
<li><p class="first">How do I explain the condition on <a class="reference internal" href="topic/pool.html#mps_addr_pool" title="mps_addr_pool"><tt class="xref c c-func docutils literal"><span class="pre">mps_addr_pool()</span></tt></a> and
<a class="reference internal" href="topic/format.html#mps_addr_fmt" title="mps_addr_fmt"><tt class="xref c c-func docutils literal"><span class="pre">mps_addr_fmt()</span></tt></a>? It&#8217;s clear what this is if the address is
the address of an allocated block? But do we guarantee anything if
it isn&#8217;t? The implementation succeeds if the address points
anywhere within a page allocated to a pool.</p>
<p><em>Answer:</em> don&#8217;t specify result except in supported cases. Not
false positive but never false negative.</p>
</li>
<li><p class="first">Constraints on order of destructors. Is it an error to destroy an
object format while there are still pools that refer to it?</p>
<p><em>Answer:</em> yes.</p>
</li>
<li><p class="first"><tt class="docutils literal"><span class="pre">mps.h</span></tt> declares the type <tt class="docutils literal"><span class="pre">mps_shift_t</span></tt> for a &#8220;shift amount&#8221;
but this is not used anywhere in the MPS. Remove?</p>
<p><em>Action:</em> removed in <a class="reference external" href="http://info.ravenbrook.com/infosys/cgi/perfbrowse.cgi?&#64;describe+179944">change 179944</a>.</p>
</li>
<li><p class="first">The structures <tt class="docutils literal"><span class="pre">mps_sac_s</span></tt> and <tt class="docutils literal"><span class="pre">mps_sac_freelist_block_s</span></tt> are
declared in mps.h. I presume that this is so for the benefit of
the macros <a class="reference internal" href="topic/cache.html#MPS_SAC_ALLOC_FAST" title="MPS_SAC_ALLOC_FAST"><tt class="xref c c-func docutils literal"><span class="pre">MPS_SAC_ALLOC_FAST()</span></tt></a> and
<a class="reference internal" href="topic/cache.html#MPS_SAC_FREE_FAST" title="MPS_SAC_FREE_FAST"><tt class="xref c c-func docutils literal"><span class="pre">MPS_SAC_FREE_FAST()</span></tt></a>, and the details of these structues
are not actually part of the public interface.</p>
<p><em>Answer:</em> RB thinks the client doesn&#8217;t need to know about these
structures, so they ought to have names starting with
underscores.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003321/">job003321</a>.</p>
</li>
<li><p class="first"><tt class="docutils literal"><span class="pre">mps_fmt_fixed_s</span></tt> is just like <a class="reference internal" href="topic/format.html#mps_fmt_A_s" title="mps_fmt_A_s"><tt class="xref c c-type docutils literal"><span class="pre">mps_fmt_A_s</span></tt></a> but with no
&#8220;skip&#8221;. I presume it&#8217;s for fixed-size pools. But the only
fixed-size pool is MFS, which has no public header, so I presume
it&#8217;s for MPS internal use only. What should I say about this?</p>
<p><em>Answer:</em> leave it undocumented. Maybe it (and
<tt class="docutils literal"><span class="pre">mps_fmt_create_fixed</span></tt>) should be removed?</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003322/">job003322</a>.</p>
</li>
<li><p class="first">It&#8217;s rather unfortunate that <a class="reference internal" href="topic/arena.html#mps_arena_create" title="mps_arena_create"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_create()</span></tt></a> take
arguments in the order <tt class="docutils literal"><span class="pre">size</span></tt>, <tt class="docutils literal"><span class="pre">base</span></tt> but
<a class="reference internal" href="topic/arena.html#mps_arena_extend" title="mps_arena_extend"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_extend()</span></tt></a> takes them in the order <tt class="docutils literal"><span class="pre">base</span></tt>,
<tt class="docutils literal"><span class="pre">size</span></tt>. I guess there&#8217;s nothing to be done about this now.</p>
<p><em>Answer:</em> RB plans to remove <a class="reference internal" href="topic/arena.html#mps_arena_create" title="mps_arena_create"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_create()</span></tt></a> and
other varargs functions.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003323/">job003323</a>.</p>
</li>
<li><p class="first">Need notes about performance. It&#8217;s important to know that
<a class="reference internal" href="topic/pool.html#mps_addr_pool" title="mps_addr_pool"><tt class="xref c c-func docutils literal"><span class="pre">mps_addr_pool()</span></tt></a> is really quick and it&#8217;s fine to call
it. But other things are slow, e.g. walking over all objects.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003324/">job003324</a>.</p>
</li>
<li><p class="first">Mark all the allocation frame stuff as deprecated in the manual.</p>
<p><em>Action:</em> I marked them all as deprecated in version 1.111.</p>
</li>
<li><p class="first">&#8220;since version 1.111&#8221; is wrong: &#8220;starting with version 1.111&#8221;
would be right.</p>
</li>
<li><p class="first"><a class="reference internal" href="topic/arena.html#mps_collections" title="mps_collections"><tt class="xref c c-func docutils literal"><span class="pre">mps_collections()</span></tt></a> should of course have been called
<tt class="xref c c-func docutils literal"><span class="pre">mps_arena_collections()</span></tt> but it&#8217;s pretty meaningless
because the collector is asynchronous (as we saw in the Scheme
example). It might make more sense to deprecate it and add a new
function <tt class="xref c c-func docutils literal"><span class="pre">mps_message_gc_start_collection()</span></tt> that returns the
collection number for a <a class="reference internal" href="topic/collection.html#mps_message_type_gc" title="mps_message_type_gc"><tt class="xref c c-func docutils literal"><span class="pre">mps_message_type_gc()</span></tt></a>.</p>
<p><em>Action:</em> added to <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003318/">job003318</a>.</p>
</li>
<li><p class="first">After you destroy an allocation point, can you continue to use the
objects you allocated on it?</p>
<p><em>Answer:</em> the objects you have committed are fine.</p>
</li>
<li><p class="first">What&#8217;s the condition for <a class="reference internal" href="topic/thread.html#mps_thread_reg" title="mps_thread_reg"><tt class="xref c c-func docutils literal"><span class="pre">mps_thread_reg()</span></tt></a> when you have
multiple arenas? Do you need to register each thread with each
arena? Or is it acceptable to register a thread with only some of
your arenas so long as you satisfy some condition? (e.g. that the
thread promises only to access blocks allocated in the arenas it
is registered with?)</p>
<p>Similarly for <a class="reference internal" href="topic/thread.html#mps_thread_dereg" title="mps_thread_dereg"><tt class="xref c c-func docutils literal"><span class="pre">mps_thread_dereg()</span></tt></a>: &#8220;after deregistration,
the thread must not access any blocks allocated in that arena.&#8221;</p>
<p><em>Answer:</em> a thread that never uses a pointer to an address in an
automatically managed pool need not be registered. Note: recommend
that the user register all threads and scan all their stacks.</p>
</li>
<li><p class="first">What&#8217;s the condition for needing to register a thread at all? If
you have a single-threaded program does that thread still need to
be registered? What if you have no moving pools? etc.</p>
<p><em>Answer:</em> see above.</p>
</li>
<li><p class="first">Start adding <a class="reference internal" href="genindex.html"><em>Index</em></a> entries.</p>
</li>
<li><p class="first">I documented the requirement for tags to be removed when calling
<tt class="xref c c-func docutils literal"><span class="pre">MPS_FIX()</span></tt>, but does this apply to format auto-header?</p>
<p><em>Answer:</em> auto-header pool does the subtraction, so this is an
exception to the rule.</p>
</li>
<li><p class="first">How do you create an auto-header format with a class method?</p>
<p><em>Answer:</em> this is just missing functionality. Also, not clear what
the class method is for.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003325/">job003325</a>.</p>
</li>
<li><p class="first">Doc for <tt class="xref c c-func docutils literal"><span class="pre">mps_arena_class_vmnz()</span></tt> says, &#8220;This class is
similar to <a class="reference internal" href="topic/arena.html#mps_arena_class_vm" title="mps_arena_class_vm"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_class_vm()</span></tt></a>, except that it has a
simple placement policy (“no zones”) that makes it slightly
faster.&#8221; Presumably there&#8217;s a corresponding disadvantage,
otherwise why would you use <a class="reference internal" href="topic/arena.html#mps_arena_class_vm" title="mps_arena_class_vm"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_class_vm()</span></tt></a> rather
than this?</p>
<p><em>Answer:</em> There&#8217;s a massive disadvantage: the lack of zones makes
automatic memory management much slower. Remove the sentence. Not
clear what <tt class="xref c c-func docutils literal"><span class="pre">mps_arena_class_vmnz()</span></tt> is for (maybe when you
have no automatic pools?)</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003326/">job003326</a>.</p>
</li>
<li><p class="first">Some of the <a class="reference internal" href="topic/allocation.html#mps_ap_s" title="mps_ap_s"><tt class="xref c c-type docutils literal"><span class="pre">mps_ap_s</span></tt></a> structure is public. What are the
use cases for the client program accessing these values other than
via <a class="reference internal" href="topic/allocation.html#mps_reserve" title="mps_reserve"><tt class="xref c c-func docutils literal"><span class="pre">mps_reserve()</span></tt></a> and <a class="reference internal" href="topic/allocation.html#mps_commit" title="mps_commit"><tt class="xref c c-func docutils literal"><span class="pre">mps_commit()</span></tt></a>? Wouldn&#8217;t they
need to know about <a class="reference internal" href="topic/allocation.html#mps_ap_fill" title="mps_ap_fill"><tt class="xref c c-func docutils literal"><span class="pre">mps_ap_fill()</span></tt></a> and <a class="reference internal" href="topic/allocation.html#mps_ap_trip" title="mps_ap_trip"><tt class="xref c c-func docutils literal"><span class="pre">mps_ap_trip()</span></tt></a>
if they were doing their own thing? But these function have
comments saying &#8220;should never be &#8220;called&#8221; directly by the client
code.&#8221;</p>
<p><em>Answer:</em> if you&#8217;re not writing client code in C (e.g. writing a
compiler, and you want to inline your allocation: you can&#8217;t use
the macros, you can generate the equivalent code). The comments in
<tt class="docutils literal"><span class="pre">mpsi.c</span></tt> are simply wrong. See RB&#8217;s <a class="reference external" href="https://github.com/dylan-lang/opendylan/issues/235">issue 235</a> on the OpenDylan bug
tracker for advice about what&#8217;s expected.</p>
<p><em>Action:</em> See what refers to these bogus comments and see if
there&#8217;s a reason. (There isn&#8217;t.) If not: reverse the sense of the
comments and refer to manual. (Done in <a class="reference external" href="http://info.ravenbrook.com/infosys/cgi/perfbrowse.cgi?&#64;describe+179971">change 179971</a>.)</p>
</li>
<li><p class="first">What about reservoirs? Is the idea that the client&#8217;s handler for
low-memory situations (whether it&#8217;s an error message or whatever)
might need to allocate? And so you can ask the MPS to reserve a
reservoir for this situation?</p>
<p><em>Answer:</em> leave it undocumented for now.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003327/">job003327</a>.</p>
</li>
<li><p class="first">Potential optimization of reserve/commit protocol: maybe we don&#8217;t
need to call <a class="reference internal" href="topic/allocation.html#mps_ap_trip" title="mps_ap_trip"><tt class="xref c c-func docutils literal"><span class="pre">mps_ap_trip()</span></tt></a>?</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003328/">job003328</a>.</p>
</li>
<li><p class="first">What are we supposed to say about <a class="reference internal" href="topic/allocation.html#mps_ap_fill" title="mps_ap_fill"><tt class="xref c c-func docutils literal"><span class="pre">mps_ap_fill()</span></tt></a> and
<a class="reference internal" href="topic/allocation.html#mps_ap_trip" title="mps_ap_trip"><tt class="xref c c-func docutils literal"><span class="pre">mps_ap_trip()</span></tt></a>?</p>
<p><em>Answer:</em> Say that they should only be called as part of the
allocation point protocol.</p>
</li>
<li><p class="first">What is the purpose of <a class="reference internal" href="topic/allocation.html#MPS_RESERVE_BLOCK" title="MPS_RESERVE_BLOCK"><tt class="xref c c-func docutils literal"><span class="pre">MPS_RESERVE_BLOCK()</span></tt></a>? It does the
same thing as <a class="reference internal" href="topic/allocation.html#mps_reserve" title="mps_reserve"><tt class="xref c c-func docutils literal"><span class="pre">mps_reserve()</span></tt></a>, but can only be used as a
statement, whereas the latter can also be used as an
expression. So I can&#8217;t say, &#8220;in such-and-such a circumstance use
<a class="reference internal" href="topic/allocation.html#MPS_RESERVE_BLOCK" title="MPS_RESERVE_BLOCK"><tt class="xref c c-func docutils literal"><span class="pre">MPS_RESERVE_BLOCK()</span></tt></a>&#8221;.</p>
<p><em>Answer:</em> it takes an lvalue instead of a pointer so it may
generate better code.</p>
</li>
<li><p class="first">&#8220;Reserve/commit protocol&#8221; or &#8220;allocation point protocol&#8221;?</p>
<p><em>Answer:</em> the latter.</p>
</li>
<li><p class="first">What about <tt class="xref c c-func docutils literal"><span class="pre">mps_alert_collection_set()</span></tt>? Seems dodgy to me.</p>
<p><em>Answer:</em> leave it undocumented.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003329/">job003329</a>.</p>
</li>
<li><p class="first">When a <tt class="docutils literal"><span class="pre">..</span> <span class="pre">note::</span></tt> block contains a numbered list with multiple
items (as <a class="reference internal" href="guide/lang.html#guide-lang-scan"><em>here</em></a>) or multiple footnotes (as
<a class="reference internal" href="pool/intro.html#pool-properties"><em>here</em></a>) the heading should say &#8220;Notes&#8221;.</p>
</li>
<li><p class="first">The names of the sectors on the &#8220;treadmill&#8221; diagram don&#8217;t
correspond exactly to the names in <a class="reference external" href="http://www.pipeline.com/~hbaker1/NoMotionGC.html">Baker 1991</a>.</p>
<p><em>Answer:</em> these names come from the diagram in <a class="reference internal" href="mmref/bib.html#jones96"><em>Jones (1996)</em></a> and refer to concepts that are missing from the
glossary: <a class="reference internal" href="glossary/f.html#term-fromspace"><em class="xref std std-term">fromspace</em></a>, <a class="reference internal" href="glossary/t.html#term-tospace"><em class="xref std std-term">tospace</em></a> (with synonyms
<a class="reference internal" href="glossary/o.html#term-old-space"><em class="xref std std-term">old space</em></a> and <a class="reference internal" href="glossary/n.html#term-new-space"><em class="xref std std-term">new space</em></a>).</p>
<p><em>Action:</em> add these concepts to the glossary. Label the treadmill
diagram with &#8220;fromspace&#8221;, &#8220;tospace&#8221;, and so on, rather than just
&#8220;from&#8221; and &#8220;to&#8221;. Here and elsewhere, refer to <a class="reference internal" href="mmref/bib.html#jones12"><em>Jones (2012)</em></a> for preferred terminology. See in particular page 138.</p>
</li>
<li><p class="first"><tt class="docutils literal"><span class="pre">eventcnv</span> <span class="pre">-b</span> <span class="pre">&lt;bucket</span> <span class="pre">size&gt;</span></tt> seems to be useless. See
<a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003331/">job003331</a>, <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003332/">job003332</a>, <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003333/">job003333</a>, <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003334/">job003334</a>, and
<a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003335/">job003335</a>. NB&#8217;s work on eventsql suggests that maybe there&#8217;s no
point in trying to make it work again (because eventsql will be
able to do this kind of aggregation and much more). So maybe best
to deprecate or remove this feature? For the moment I&#8217;ve left it
undocumented.</p>
<p><em>Answer:</em> <tt class="docutils literal"><span class="pre">eventcnv</span></tt> was always intended to be a simple tool that
just translates events, not a report generator.</p>
<p><em>Action:</em> remove the -e, -b and -v options. (-v should be always
turned on.) Also remove the label processing. Look at NB&#8217;s branch
in case he&#8217;s done some of this work already. (He hasn&#8217;t.)</p>
</li>
<li><p class="first">Write something about telemetry labels in the telemetry chapter of
the reference manual.</p>
</li>
</ol>
<ol class="arabic" start="77">
<li><p class="first">The debugging section on <a class="reference internal" href="guide/debug.html#guide-debug-underscanning"><em>Example: underscanning</em></a> is not
entirely convincing as to the utility of the telemetry. I&#8217;d like
to have a more compelling example here.</p>
<p><em>Answer:</em> a better example is going to have to wait for better
tools.</p>
</li>
<li><p class="first">Should memory addresses increase going up the page, or down the
page? The output of gdb&#8217;s <tt class="docutils literal"><span class="pre">x</span></tt> command always has them going down
the page so maybe we should follow that? It would involve turning
a bunch of diagrams upside down.</p>
<p><em>Answer:</em> <a class="reference internal" href="mmref/bib.html#jones12"><em>Jones (2012)</em></a> has addresses increasing
as they go down the page, so let&#8217;s follow that.</p>
</li>
<li><p class="first">Suppose that in the Scheme interpreter you wanted to return some
statistics from the <tt class="docutils literal"><span class="pre">(gc)</span></tt> function, how would you do this? You
could poll the message queue after calling
<a class="reference internal" href="topic/arena.html#mps_arena_collect" title="mps_arena_collect"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_collect()</span></tt></a> and before calling
<a class="reference internal" href="topic/arena.html#mps_arena_release" title="mps_arena_release"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_release()</span></tt></a>, I suppose.</p>
<p><em>Answer:</em> this might happen to work, but it&#8217;s not supported.</p>
</li>
<li><p class="first">Need to migrate the new Scheme functionality (new make_string, fix
to append, new entry points) back to the &#8220;before the MPS&#8221; version.</p>
<p><em>Action:</em> done in <a class="reference external" href="http://info.ravenbrook.com/infosys/cgi/perfbrowse.cgi?&#64;describe+180069">change 180069</a>.</p>
</li>
<li><p class="first">The Scheme interpreter could avoid printing &#8220;<tt class="docutils literal"><span class="pre">#[undefined]</span></tt>&#8221; at
the REPL (like Python with <tt class="docutils literal"><span class="pre">None</span></tt>).</p>
<p><em>Answer:</em> not useful.</p>
</li>
<li><p class="first">Is the message example correct?</p>
<p><em>Answer:</em> no, the MPS doesn&#8217;t guarantee the timely delivery of
messages, so this example is invalid. Remove it.</p>
<p><em>Action:</em> what we could do is return a collection ID from
<a class="reference internal" href="topic/arena.html#mps_arena_collect" title="mps_arena_collect"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_collect()</span></tt></a>, <a class="reference internal" href="topic/arena.html#mps_arena_step" title="mps_arena_step"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_step()</span></tt></a> and
<a class="reference internal" href="topic/arena.html#mps_arena_park" title="mps_arena_park"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_park()</span></tt></a>, and provide a mechanism to get the
collection ID from the collection message. I added this suggestion
to <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003318/">job003318</a>.</p>
</li>
<li><p class="first">If a block has been finalized, can you &#8220;resurrect&#8221; it and
re-register it for finalization?</p>
<p><em>Answer:</em> make a note about resurrection in the MPS not being the
same as the usual notion. It&#8217;s fine to re-register after
retrieving the message.</p>
</li>
<li><p class="first">If a block has been registered for finalization more than once,
what happens if you deregister it? Do you have to deregister it as
many times as you registered it? Or do you only need to deregister
it once?</p>
<p><em>Answer:</em> say nothing about this esoteric subject.</p>
</li>
<li><p class="first">What exactly is our policy about support for the documented
behaviour of public identifiers. &#8220;The documented behaviour of
public identifiers will not be changed in a backward-incompatible
fashion without a period of notice lasting at least one version.&#8221;</p>
<p><em>Answer:</em> &#8220;We intend to support the existence and behaviour of
documented symbols. We will only remove or change these on a
version change (not between patch releases). Normally one
version&#8217;s notice.&#8221;</p>
</li>
<li><p class="first">In <tt class="docutils literal"><span class="pre">buffer.c</span></tt> it says:</p>
<div class="highlight-c"><div class="highlight"><pre><span class="cm">/* Assumes pun compatibility between Rank and mps_rank_t */</span>
<span class="cm">/* Which is checked by mpsi_check in &lt;code/mpsi.c&gt; */</span>
</pre></div>
</div>
<p>but I see no such check in <tt class="docutils literal"><span class="pre">mpsi_check</span></tt>.</p>
<p><em>Answer:</em> the comment was wrong: ranks are no longer part of the
public interface. Removed in <a class="reference external" href="http://info.ravenbrook.com/infosys/cgi/perfbrowse.cgi?&#64;describe+180031">change 180031</a>.</p>
</li>
<li><p class="first">Could simplify a lot of glossary references if plurals were
handled automatically. That is, if a glossary entry for &#8220;bytes
(1)&#8221; is found to be missing, then &#8220;byte (1)&#8221; should be tried
instead. (Similarly for hyphens versus spaces.)</p>
</li>
<li><p class="first">Where glossary entries differ on in hyphens versus spaces there&#8217;s
no need to give both.</p>
</li>
<li><p class="first"><tt class="docutils literal"><span class="pre">MPS_TELEMETRY_CONTROL=65535</span></tt> is ugly. Why not
<tt class="docutils literal"><span class="pre">MPS_TELEMETRY_CONTROL=all</span></tt>?</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003340/">job003340</a>.</p>
</li>
<li><p class="first">Chapters in the User Guide need a conclusion of some kind. e.g. in
debugging, what should you do if this hasn&#8217;t helped?</p>
<p><em>Action:</em> added &#8220;What now?&#8221; section.</p>
</li>
<li><p class="first">Chapter about the community resources. &#8220;Get in touch&#8221;. See LLVM
and Boost for examples of this kind of thing. Put it in the
sidebar so it&#8217;s always there.</p>
<p><em>Action:</em> added &#8220;Contact us&#8221; page and link in the sidebar of every
page.</p>
</li>
<li><p class="first">Extensibility: &#8220;Writing your own pool or need a pool that&#8217;s not
listed here: have a go, but get in touch.&#8221; We can write new pool
classes for you.</p>
<p><em>Action:</em> added a section to the Pool reference.</p>
</li>
<li><p class="first">Multi-core is not supported (except in the case of multiple
arenas). See the <a class="reference external" href="http://news.ycombinator.com/item?id=4521988">Hacker News thread</a>. Could be done by
running the MPS in a separate process but that might be horrible
and heavyweight.</p>
<p>There may be more documentation needed here (in
<a class="reference internal" href="topic/thread.html#topic-thread"><em>Threads</em></a>): if the mutator is multi-threaded, it needs
to reserve <tt class="docutils literal"><span class="pre">SIGUSR1</span></tt> (or whatever the actual signal is) for use
by the MPS. [It turned out that <tt class="docutils literal"><span class="pre">SIGBUS</span></tt> or <tt class="docutils literal"><span class="pre">SIGSEGV</span></tt> is used
for this purpose, not <tt class="docutils literal"><span class="pre">SIGUSR1</span></tt>.]</p>
<p>Also <tt class="docutils literal"><span class="pre">SIGBUS</span></tt> (xc) or <tt class="docutils literal"><span class="pre">SIGSEGV</span></tt> (fr, li) or &#8220;first chance
exception handler&#8221; (w3). Affects debugging. Also, if you handle your
own SIGBUS etc then you need to give us a call.</p>
<p><em>Action:</em> added a section on &#8220;Signal handling issues&#8221; to
<a class="reference internal" href="topic/thread.html#topic-thread"><em>Threads</em></a>. (Not clear that this is the right place,
since it isn&#8217;t really thread-specific, but I couldn&#8217;t find a
better place and I can always move it later.)</p>
</li>
<li><p class="first">Maybe target R4RS instead of R6RS? (R4RS is the &#8220;one true&#8221;
Scheme.)</p>
<p><em>Action:</em> fixed in <a class="reference external" href="http://info.ravenbrook.com/infosys/cgi/perfbrowse.cgi?&#64;describe+180033">change 180033</a>.</p>
</li>
<li><p class="first">Bring <a class="reference internal" href="mmref/lang.html#lang"><em>Memory management in various languages</em></a> up to date. Add C#, Lua, Python.</p>
</li>
<li><p class="first">Change &#8220;event&#8221; to &#8220;event category&#8221; in discussion of the telemetry
filter.</p>
</li>
<li><p class="first">In <a class="reference internal" href="guide/debug.html#guide-debug"><em>Debugging with the Memory Pool System</em></a>, need to provide more context around each
assertion, using GDB command <tt class="docutils literal"><span class="pre">list</span></tt>.</p>
</li>
<li><p class="first">In the &#8220;Cautions&#8221; section of <a class="reference internal" href="topic/finalization.html#topic-finalization"><em>Finalization</em></a>, add
something to effect of, &#8220;A block might be dead and yet not
finalizable, because it is being kept alive by an ambiguous
reference.&#8221;</p>
</li>
<li><p class="first">What does this mean in <a class="reference internal" href="topic/finalization.html#topic-finalization"><em>Finalization</em></a>:
&#8220;mps_pool_destroy() should therefore not be invoked on pools
containing objects registered for finalization.&#8221;</p>
<p><em>Answer:</em> This is clearly a misfeature if true, because how could
you destroy a pool containing finalizable objects?</p>
<p><em>Action:</em> make <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003341/">job003341</a></p>
</li>
<li><p class="first">Try to find &#8220;GARBAGE COLLECTING... HERE ARE SOME INTERESTING
STATISTICS&#8221; early Lisp anecdote and put it in
<a class="reference internal" href="topic/telemetry.html#topic-telemetry"><em>Telemetry</em></a>.</p>
<p>The anecdote is at the end of <a class="reference internal" href="mmref/bib.html#mccarthy79"><em>McCarthy (1979)</em></a>.</p>
</li>
<li><p class="first">Are there any other use cases for the clamped and parked states?
Are there any use cases that apply specifically to the parked
state?</p>
<p><em>Answer:</em> clamp prevents a flip and so buffers can&#8217;t be trapped,
and so <a class="reference internal" href="topic/allocation.html#mps_commit" title="mps_commit"><tt class="xref c c-func docutils literal"><span class="pre">mps_commit()</span></tt></a> will always succeed, and so
allocation will always run at max speed.</p>
<p>It might be helpful for debugging, for example you might want to
type it at the GDB prompt to ensure that nothing moves around
while you are debugging.</p>
</li>
<li><p class="first">It&#8217;s kind of a shame that the MPS has two means for &#8220;committed&#8221;.
<a class="reference internal" href="glossary/c.html#term-committed-1"><em class="xref std std-term">committed<sup>(1)</sup></em></a> meaning &#8220;mapped to RAM&#8221;, as in
<a class="reference internal" href="topic/arena.html#mps_arena_commit_limit" title="mps_arena_commit_limit"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_commit_limit()</span></tt></a>; and <a class="reference internal" href="glossary/c.html#term-committed-2"><em class="xref std std-term">committed<sup>(2)</sup></em></a>
meaning &#8220;initialized and placed under management by the MPS&#8221;, as
in <a class="reference internal" href="topic/allocation.html#mps_commit" title="mps_commit"><tt class="xref c c-func docutils literal"><span class="pre">mps_commit()</span></tt></a>. Probably too late to do anything about
this.</p>
<p><em>Answer:</em> too late to change.</p>
</li>
<li><p class="first">The documentation for <a class="reference internal" href="topic/arena.html#mps_arena_has_addr" title="mps_arena_has_addr"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_has_addr()</span></tt></a> says, &#8220;call
this function and interpret the result while the arena is in the
<a class="reference internal" href="glossary/p.html#term-parked-state"><em class="xref std std-term">parked state</em></a>&#8221;. Similarly, <a class="reference internal" href="topic/root.html#mps_arena_roots_walk" title="mps_arena_roots_walk"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_roots_walk()</span></tt></a>
says &#8220;This function may only be called when the arena is in the
<a class="reference internal" href="glossary/p.html#term-parked-state"><em class="xref std std-term">parked state</em></a>.&#8221; What&#8217;s wrong with the clamped state in
these cases? (I can see that <a class="reference internal" href="topic/root.html#mps_arena_roots_walk" title="mps_arena_roots_walk"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_roots_walk()</span></tt></a>
asserts if not in the parked state, but I guess I&#8217;d like an
explanation.)</p>
<p><em>Answer:</em> there&#8217;s still stuff going on in the clamped state.</p>
</li>
<li><p class="first">Are there use cases for <a class="reference internal" href="topic/arena.html#mps_arena_collect" title="mps_arena_collect"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_collect()</span></tt></a> other than
development and testing?</p>
<p><em>Answer:</em> probably not. &#8220;You might think that...&#8221;</p>
</li>
<li><p class="first">It&#8217;s a shame that the names <a class="reference internal" href="topic/arena.html#mps_arena_release" title="mps_arena_release"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_release()</span></tt></a> and
&#8220;unclamped state&#8221; don&#8217;t match. Could I call the &#8220;unclamped state&#8221;
the &#8220;released state&#8221; in the documentation?</p>
<p><em>Answer:</em> too late to change.</p>
</li>
<li><p class="first">After calling <a class="reference internal" href="topic/arena.html#mps_arena_expose" title="mps_arena_expose"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_expose()</span></tt></a>, how do you restore the
protection? Do you have to call <a class="reference internal" href="topic/arena.html#mps_arena_release" title="mps_arena_release"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_release()</span></tt></a> or
are there other functions that will do the job, e.g.
<a class="reference internal" href="topic/arena.html#mps_arena_collect" title="mps_arena_collect"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_collect()</span></tt></a>, <a class="reference internal" href="topic/arena.html#mps_arena_start_collect" title="mps_arena_start_collect"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_start_collect()</span></tt></a>,
or <a class="reference internal" href="topic/arena.html#mps_arena_step" title="mps_arena_step"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_step()</span></tt></a>?</p>
<p><em>Answer:</em> the protection will arise naturally: no need to call
anything specific.</p>
</li>
<li><p class="first"><a class="reference internal" href="topic/arena.html#mps_arena_unsafe_expose_remember_protection" title="mps_arena_unsafe_expose_remember_protection"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_unsafe_expose_remember_protection()</span></tt></a> and
<a class="reference internal" href="topic/arena.html#mps_arena_unsafe_restore_protection" title="mps_arena_unsafe_restore_protection"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_unsafe_restore_protection()</span></tt></a> are kind of hairy
(as well as having absurd names). What&#8217;s the use case? Exposing
the MPS &#8220;is expected only to be useful for debugging&#8221; so why have
special unsafe functions for optimizing the expose/restore
procedure? Maybe these should be left undocumented?</p>
<p><em>Answer:</em> Deprecate them for the moment. Make a job to
investigate the use case.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003342/">job003342</a></p>
</li>
<li><p class="first">Have I correctly explained why the documentation is not very
forthcoming about the set of result codes a function might
return?</p>
<p><em>Answer:</em> the paragraph is fine, but there&#8217;s a job here. For some
functions we could say something along the lines of: &#8220;if you
receive MPS_RES_X, that means Y&#8221;.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003343/">job003343</a></p>
</li>
<li><p class="first">NB&#8217;s <a class="reference external" href="http://info.ravenbrook.com/project/mps/doc/2002-06-18/obsolete-mminfo/mmdoc/protocol/mps/format/index.html">MPS Format Protocol</a>
says that formats A and B are deprecated. Surely this isn&#8217;t the
case? That would leave only auto-header supported.</p>
<p><em>Answer:</em> ask NB, he wrote it.</p>
</li>
<li><p class="first">Say something about assertions and varieties in the error chapter.</p>
</li>
<li><p class="first">Need discussion in <a class="reference internal" href="topic/format.html#topic-format"><em>Object formats</em></a> under
<a class="reference internal" href="topic/format.html#mps_fmt_auto_header_s" title="mps_fmt_auto_header_s"><tt class="xref c c-type docutils literal"><span class="pre">mps_fmt_auto_header_s</span></tt></a> about client pointers versus base
pointers.</p>
</li>
<li><p class="first">What&#8217;s the purpose of allocation frames and the SNC pool class?
Perhaps ask Pekka about how it&#8217;s worked out for Global Graphics.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003344/">job003344</a></p>
</li>
<li><p class="first">Note about sizes of generations in the Scheme example are
deliberately chosen to be small so you can see it working.</p>
</li>
<li><p class="first">What&#8217;s the purpose of segregated allocation caches?</p>
<p><em>Answer:</em> There might be a paper about SACs. Run program once and
write down the sizes of the objects you allocate. Change malloc
into a macro that tests the size (because the size is almost
always constant). Dramatic improvement in performance and
reduction in fragmentation. Interface to a pool that allows it to
exploit this pattern. Perhaps in use in ScriptWorks? Write to
Pekka and ask him how this has worked out. (After 1.111.0.)</p>
</li>
<li><p class="first">Need to associate paragraph of text with deprecation notice. (See
for example <a class="reference internal" href="topic/scanning.html#mps_fix" title="mps_fix"><tt class="xref c c-func docutils literal"><span class="pre">mps_fix()</span></tt></a>.) Also, deprecation notices need
nicer formatting.</p>
</li>
<li><p class="first">&#8220;The maximum number of arenas that the system can support is
around 10.&#8221; Is this true? What goes wrong? (I tried creating lots
of arenas and nothing seemed to go wrong.)</p>
<p><em>Answer:</em> this was probably a finger in the air. Remove it.</p>
</li>
<li><p class="first">Consistent pluralization in headings. &#8220;Arena states&#8221;. &#8220;Other
notes&#8221; → &#8220;Ambiguous references&#8221;. &#8220;You could use this to optimize
the scan by avoiding the need to reassemble and store the updated
reference after calling <a class="reference internal" href="topic/scanning.html#MPS_FIX2" title="MPS_FIX2"><tt class="xref c c-func docutils literal"><span class="pre">MPS_FIX2()</span></tt></a>.</p>
</li>
<li><p class="first">Add comments giving links to sources.</p>
</li>
<li><p class="first">This (from <a class="reference internal" href="topic/critical.html#topic-critical"><em>The critical path</em></a>) needs mentioning in various
places: &#8220;it&#8217;s important to give a good estimate of the amount of
address space you will ever occupy with objects when you
initialize the arena&#8221;.</p>
</li>
<li><p class="first">&#8220;Magic&#8221; needs explaining in <a class="reference internal" href="topic/location.html#topic-location"><em>Location dependency</em></a>.</p>
<p><em>Answer:</em> that&#8217;s because we hit it: it didn&#8217;t move.</p>
</li>
<li><p class="first">Documentation needs copying into <tt class="docutils literal"><span class="pre">scheme.c</span></tt>.</p>
</li>
<li><p class="first">What&#8217;s a &#8220;class structure&#8221;? (See <a class="reference internal" href="topic/cache.html#mps_sac_alloc" title="mps_sac_alloc"><tt class="xref c c-func docutils literal"><span class="pre">mps_sac_alloc()</span></tt></a> and
<a class="reference internal" href="topic/cache.html#mps_sac_free" title="mps_sac_free"><tt class="xref c c-func docutils literal"><span class="pre">mps_sac_free()</span></tt></a>.) Does it mean &#8220;exactly the same set of
size classes in the same order&#8221;? Or even &#8220;exactly the same
pointer to array of size classes&#8221;?</p>
<p><em>Answer:</em> probably the former. Check the code.</p>
</li>
<li><p class="first">Needs to be a topic on low memory handling. How can you do this
while still keeping the allocation fast and inline? How does the
MPS behave when it&#8217;s low on memory? Performance degrades (due to
running out of zones) and then there are emergency collections.</p>
<p><em>Action:</em> created <em class="xref std std-ref">topic-low</em>.</p>
</li>
</ol>
<ol class="arabic" start="127">
<li><p class="first"><a class="reference internal" href="topic/root.html#mps_arena_roots_walk" title="mps_arena_roots_walk"><tt class="xref c c-func docutils literal"><span class="pre">mps_arena_roots_walk()</span></tt></a> says, &#8220;A client-supplied function
is called for every root reference which points to an object in
any automatically managed pools&#8221;. Is this right?</p>
<p><em>Answer:</em> the description seems to be right. What&#8217;s the use case?
If none, deprecate it.</p>
</li>
<li><p class="first">From <a class="reference internal" href="topic/thread.html#topic-thread"><em>Threads</em></a>: &#8220;A thread must be registered with an
<a class="reference internal" href="glossary/a.html#term-arena"><em class="xref std std-term">arena</em></a> if it ever uses a pointer to a location in an
<a class="reference internal" href="glossary/a.html#term-automatic-memory-management"><em class="xref std std-term">automatically managed</em></a>
<a class="reference internal" href="glossary/p.html#term-pool"><em class="xref std std-term">pool</em></a> belonging to that arena.&#8221; This isn&#8217;t quite right,
and the real requirement is quite hard to document, so perhaps we
need to make this a documented property of the pool (&#8220;whether or
not threads that access memory allocated in the pool need to be
registered&#8221;). Similarly in <a class="reference internal" href="topic/thread.html#mps_thread_dereg" title="mps_thread_dereg"><tt class="xref c c-func docutils literal"><span class="pre">mps_thread_dereg()</span></tt></a>.</p>
</li>
<li><p class="first">What happens if a thread is killed or otherwise terminates
without being deregistered? Does the MPS go wrong?</p>
<p><em>Answer:</em> Probably all we do is try to send signals to it. What
does pthreads do with this? It may be hard for the client program
to deregister all their threads: better for the MPS to cope
gracefully.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003348/">job003348</a>.</p>
</li>
<li><p class="first"><tt class="docutils literal"><span class="pre">readme.txt</span></tt> should contain a brief overview and pointers to
more documentation, and should (only) duplicate other
documentation. There&#8217;s nothing wrong with it being somewhat
redundant. Its supported platforms section doesn&#8217;t exist
elsewhere. Note that the platforms we build with (in
<tt class="docutils literal"><span class="pre">build.txt</span></tt>) is not the same list.</p>
</li>
<li><p class="first">Is <a class="reference internal" href="topic/interface.html#mps_clock_t" title="mps_clock_t"><tt class="xref c c-type docutils literal"><span class="pre">mps_clock_t</span></tt></a> a transparent alias for <tt class="docutils literal"><span class="pre">unsigned</span>
<span class="pre">long</span></tt>? I presume it must be: if it were platform-specific, or
supplied by the plinth, then it wouldn&#8217;t be defined in <tt class="docutils literal"><span class="pre">mps.h</span></tt>,
surely?</p>
<p><em>Answer:</em> this is a bug, fixed in <a class="reference external" href="http://info.ravenbrook.com/infosys/cgi/perfbrowse.cgi?&#64;describe+180151">change 180151</a>.</p>
</li>
</ol>
<blockquote>
<div></div></blockquote>
<ol class="arabic" start="132">
<li><p class="first"><a class="reference internal" href="topic/plinth.html#mps_lib_get_stderr" title="mps_lib_get_stderr"><tt class="xref c c-func docutils literal"><span class="pre">mps_lib_get_stderr()</span></tt></a> appears in <tt class="docutils literal"><span class="pre">mpslib.h</span></tt> and
<tt class="docutils literal"><span class="pre">mpsliban.c</span></tt> but is not called by the MPS (it uses
<a class="reference internal" href="topic/plinth.html#mps_lib_assert_fail" title="mps_lib_assert_fail"><tt class="xref c c-func docutils literal"><span class="pre">mps_lib_assert_fail()</span></tt></a> for assertions). Should this be
documented to reserve the option of using it, or should it be
left out?</p>
<p>Similar question for <a class="reference internal" href="topic/plinth.html#mps_lib_get_stdout" title="mps_lib_get_stdout"><tt class="xref c c-func docutils literal"><span class="pre">mps_lib_get_stdout()</span></tt></a>: it&#8217;s used, but
only by the &#8220;diagnostic&#8221; code which has no public interface.</p>
<p><em>Action:</em> Keep <tt class="docutils literal"><span class="pre">stderr</span></tt> documented. Add a note that it&#8217;s not used
but for possible future use. Same thing about <tt class="docutils literal"><span class="pre">stdout</span></tt>.</p>
</li>
<li><p class="first">What&#8217;s the purpose of <tt class="docutils literal"><span class="pre">mps_SEH_filter</span></tt> and <tt class="docutils literal"><span class="pre">mps_SEH_handler</span></tt>?
Do they need to be documented?</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003349/">job003349</a>.</p>
</li>
<li><p class="first">What&#8217;s the use case for AMS? It&#8217;s suitable when you have blocks
that need to be automatically managed but can&#8217;t be moved. But
when does this happen? If foreign code maintains remembers their
location then it seems unlikely that they can be automatically
managed (how can these foreign references keep the blocks
alive?).</p>
<p><em>Answer:</em> It&#8217;s useful for a step in an integration because its
automatic but non-moving. But not ready for production: best to
plan to switch to AMC later on. Could be developed into a more
solid mark-and-sweep pool. Contact us.</p>
</li>
<li><p class="first">Does AMS use protection? I see that it calls <tt class="docutils literal"><span class="pre">ShieldExpose</span></tt> and
<tt class="docutils literal"><span class="pre">ShieldCover</span></tt> when calling the format&#8217;s skip method, but not
otherwise (e.g. when calling the format&#8217;s scan method). If it
does I need to update the pool choice algorithm.</p>
<p>Similar question for SNC? It calls <tt class="docutils literal"><span class="pre">ShieldExpose</span></tt> and
<tt class="docutils literal"><span class="pre">ShieldCover</span></tt> when calling the format&#8217;s pad method, but not
otherwise.</p>
<p><em>Answer:</em> The rule is &#8220;if it needs to be scanned, either it gets
scanned atomically on flip, or else it gets read-protected&#8221;. So
objects in both of these pools get protected.</p>
</li>
<li><p class="first">It seems possible that MV should not be used. What should I say
about this? Should we remove MV from the documentation
altogether.</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003350/">job003350</a>.</p>
</li>
<li><p class="first">What does the &#8220;No Checking&#8221; in &#8220;Stack No Checking&#8221; refer to?</p>
<p><em>Action:</em> Added to <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003344/">job003344</a>. Deprecated SNC.</p>
</li>
<li><p class="first">The hash table implementation is a bit rubbish: it waits until
the table is completely full before rehashing. Should ensure that
it never gets more than 50% full. Also the constructor should be
called <tt class="docutils literal"><span class="pre">make-eq-hashtable</span></tt> following R6RS.</p>
</li>
<li><p class="first">It seems a shame that MVT doesn&#8217;t support <a class="reference internal" href="topic/allocation.html#mps_alloc" title="mps_alloc"><tt class="xref c c-func docutils literal"><span class="pre">mps_alloc()</span></tt></a> as
this would be fairly trivial via an internal <a class="reference internal" href="glossary/a.html#term-allocation-point"><em class="xref std std-term">allocation
point</em></a>.</p>
<p><em>Action:</em> added to <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003350/">job003350</a>.</p>
</li>
<li><p class="first">Document MFS. Explain how it works (putting link in the free
block) and how this is unlike other pools.</p>
</li>
<li><p class="first">Do we support instruction emulation on x86-64? DL says that it&#8217;s
not in yet. What about OS X?</p>
<p><em>Action:</em> made <a class="reference external" href="https://info.ravenbrook.com/project/mps/issue/job003352/">job003352</a>.</p>
</li>
<li><p class="first">Weakness: how to detect splatting. (In scan.) Cope with NULL. Do
all references get splatted at the same time?</p>
<p><em>Action:</em> created <a class="reference internal" href="topic/weak.html#topic-weak"><em>Weak references</em></a>.</p>
</li>
<li><p class="first">Document the purpose of AWL and explain that it&#8217;s for OpenDylan.
&#8220;If you require more general implementation of weakness, contact
us.&#8221;</p>
<p><em>Action:</em> see <a class="reference internal" href="pool/awl.html#pool-awl"><em>AWL (Automatic Weak Linked)</em></a>.</p>
</li>
<li><p class="first">In AWL, what are the restrictions on the dependent object? Must
it be another object in AWL? What are we allowed to do with it?
What state will the object be in when we look at it? Does it have
to be fixed?</p>
<p><em>Answer:</em> better not be in a moving pool (otherwise might point
to forwarding object). You can fix it, but the fixed-up reference
won&#8217;t necessarily be exposed. Let&#8217;s document, &#8220;dependent object
must be in a non-moving pool, or in memory not managed by the
MPS.&#8221; (Should fix the dependent object; that may result in the
reference being splatted if the dependent object died.)</p>
</li>
<li><p class="first">Can you use the same generation chain with more than one pool?</p>
<p><em>Answer:</em> not only can you do this, but you ought to.</p>
</li>
</ol>
</div>
</div>
</div>
</div>
</div>
<div class="sphinxsidebar">
<div class="sphinxsidebarwrapper">
<p class="logo"><a href="index.html">
<img class="logo" src="_static/logo.png" alt="Logo"/>
</a></p>
<h3><a href="index.html">Table Of Contents</a></h3>
<ul>
<li><a class="reference internal" href="#">To do</a><ul>
<li><a class="reference internal" href="#outstanding">Outstanding</a></li>
<li><a class="reference internal" href="#complete">Complete</a></li>
</ul>
</li>
</ul>
<h4>Previous topic</h4>
<p class="topless"><a href="contact.html"
title="previous chapter">Contact us</a></p><h4>Contact us</h4>
<p class="topless"><a href="mailto:mps-questions@ravenbrook.com">mps-questions@ravenbrook.com</a></p>
</div>
</div>
<div class="clearer"></div>
</div>
<div class="related">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="genindex.html" title="General Index"
>index</a></li>
<li class="right" >
<a href="contact.html" title="Contact us"
>previous</a> |</li>
<li><a href="index.html">Memory Pool System 1.111.0 documentation</a> &raquo;</li>
</ul>
</div>
<div class="footer">
&copy; <a href="copyright.html">Copyright</a> 2012, Ravenbrook Limited.
Created using <a href="http://sphinx.pocoo.org/">Sphinx</a> 1.1.3.
</div>
</body>
</html>
Reference in a new issue View git blame Copy permalink