-<!-- $Id: zoom.xml,v 1.6 2001-11-06 17:05:19 adam Exp $ -->
+<!-- $Id: zoom.xml,v 1.7 2001-11-11 22:25:25 adam Exp $ -->
<chapter id="zoom"><title>Building clients with ZOOM</title>
<para>
that because of typedefs. All destroy methods should gracefully ignore a
<literal>NULL</literal> pointer.
</para>
+ <para>
+ In each of the sections below you'll find a sub section called
+ protocol behavior, that descries how the API maps to the Z39.50
+ protocol.
+ </para>
<sect1 id="zoom.connections"><title>Connections</title>
<para>The Connection object is a session with a target.
<row><entry>
async</entry><entry>If true (1) the connection operates in
asynchronous operation which means that all calls are non-blocking
- except <function>Z3950_event</function>.
+ except
+ <link linkend="zoom.events"><function>Z3950_event</function></link>.
</entry><entry>0</entry></row>
<row><entry>
maximumRecordSize</entry><entry> Maximum size of single record.
holds messages for the error and additional-info if passed as
non-<literal>NULL</literal>.
</para>
+ <sect2><title>Protocol behavior</title>
+ <para>
+ The calls <function>Z3950_connection_new</function> and
+ <function>Z3950_connection_connect</function> establises a TCP/IP
+ connection and sends an Initialize Request to the target if
+ possible. In addition, the calls waits for an Initialize Response
+ from the target and the result is inspected (OK or rejected).
+ </para>
+ <para>
+ If <literal>proxy</literal> is set then the client will establish
+ a TCP/IP connection with the peer as specified by the
+ <literal>proxy</literal> host and the hostname as part of the
+ connect calls will be set as part of the Initialize Request.
+ The proxy server will then "forward" the PDU's transparently
+ to the target behind the proxy.
+ </para>
+ <para>
+ For the authentication parameters, if option <literal>user</literal>
+ is set and both options <literal>group</literal> and
+ <literal>pass</literal> are unset, then Open style
+ authentication is used (Version 2/3) in which case the username
+ is usually followed by a slash, then by a password.
+ If either <literal>group</literal>
+ or <literal>pass</literal> is set then idPass authentication
+ (Version 3 only) is used. If none of the options are set, no
+ authentication parameters are set as part of the Initialize Request
+ (obviously).
+ </para>
+ <para>
+ When option <literal>async</literal> is 1, it really means that
+ all network operations are postponed (and queued) until the
+ function <literal>Z3950_event</literal> is invoked. When doing so
+ it doesn't make sense to check for errors after
+ <literal>Z3950_connection_new</literal> is called since that
+ operation "connecting - and init" is still incomplete and the
+ API cannot tell the outcome (yet).
+ </para>
+ </sect2>
</sect1>
<sect1 id="zoom.query"><title>Queries</title>
<para>
sort criteria using the same string notation for sort as offered by
the <link linkend="sortspec">YAZ client</link>.
</para>
+ <sect2><title>Protocol behavior</title>
+ <para>
+ The query object is just an interface for the member Query
+ in the SearchRequest. The sortby-function is an interface to the
+ sortSequence member of the SortRequest.
+ </para>
+ </sect2>
</sect1>
<sect1 id="zoom.resultsets"><title>Result sets</title>
<para>
value, then target will return all records using small element set name
</entry><entry>0</entry></row>
<row><entry>
- largeSetLowerBound</entry><entry>If hits is greator than this value, the target
- will return no records.
+ largeSetLowerBound</entry><entry>If hits is greator than this
+ value, the target will return no records.
</entry><entry>1</entry></row>
<row><entry>
mediumSetPresentNumber</entry><entry>This value represents
</tbody>
</tgroup>
</table>
+ <sect2>
+ <title>Protocol behavior</title>
+ <para>
+ The creation of a result set involves at least a SearchRequest
+ - SearchResponse protocol handshake. Following that, if a sort
+ critieria was specified as part of the query, a sortRequest -
+ SortResponse handshake takes place. Note that it is necessary to
+ perform sorting before any retrieval takes place, so no records will
+ be returned from the target as part of the SearchResponse because these
+ would be unsorted. Hence, piggyback is disabled when sort critieria
+ is set. Following Search - and a Possible sort, Retrieval takes
+ place - as one or more Present Requests - Present Response being
+ transferred.
+ </para>
+ <para>
+ The API allows for two different modes for retrieval. A high level
+ mode which is somewhat more powerful and a low level one.
+ The low level is "enabled" when the settings
+ <literal>smallSetUpperBound</literal>,
+ <literal>mediumSetPresentNumber</literal> and
+ <literal>largeSetLowerBound</literal> are set. The low level mode
+ thus allows you to precisely set how records are returned as part
+ of a search response as offered by the Z39.50 protocol.
+ Since the client may be retrieving records as part of the
+ search response, this mode doesn't work well if sorting is used.
+ </para>
+ <para>
+ The high-level mode allows you to fetch a range of records from
+ the result set with a given start offset. When you use this mode
+ the client will automatically use piggyback if that is possible
+ with the target and perform one or more present requests as needed.
+ Even if the target returns fewer records as part of a present response
+ because of a record size limit, etc. the client will repeat sending
+ present requests. As an example, if option <literal>start</literal>
+ is 0 (default) and <literal>count</literal> is 4, and
+ <literal>piggyback</literal> is 1 (default) and no sorting critieria
+ is specified, then the client will attempt to retrieve the 4
+ records as part the search response (using piggyback). On the other
+ hand, if either <literal>start</literal> is positive or if
+ a sorting criteria is set, or if <literal>piggyback</literal>
+ is 0, then the client will not perform piggyback but send Present
+ Requests instead.
+ </para>
+ <para>
+ If either of the options <literal>mediumSetElementSetName</literal> and
+ <literal>smallSetElementSetName</literal> are unset, the value
+ of option <literal>elementSetName</literal> is used for piggyback
+ searches. This means that for the high-level mode you only have
+ to specify one elementSetName option rather than three.
+ </para>
+ </sect2>
</sect1>
<sect1 id="zoom.records"><title>Records</title>
<para>
Function <function>Z3950_resultset_records</function> retrieves
a number of records from a result set. Parameter <literal>start</literal>
and <literal>count</literal> specifies the range of records to
- be returned. Upon completion array <literal>recs[0], ..recs[count-1]</literal>
+ be returned. Upon completion array
+ <literal>recs[0], ..recs[count-1]</literal>
holds record objects for the records. The array of records
<literal>recs</literal> should be allocate prior to calling
<function>Z3950_resultset_records</function>. Note that for those
nature (type) of the pointer depends on the <function>type</function>
given. In addition for certain types, the length
<literal>len</literal> passed will be set to the size in bytes of
- the returned information. The types <literal>database</literal>,
- <literal>syntax</literal> and <literal>render</literal> are
- supported. More will be added later.
+ the returned information.
+ <variablelist>
+ <varlistentry><term><literal>database</literal></term>
+ <listitem><para>The database that holds the record is returned
+ as a C string. Return type <literal>char *</literal>.
+ </para></listitem>
+ </varlistentry>
+ <varlistentry><term><literal>syntax</literal></term>
+ <listitem><para>The transfer syntax (OID) of the record is returned
+ as a C string. Return type <literal>char *</literal>.
+ </para></listitem>
+ </varlistentry>
+ <varlistentry><term><literal>render</literal></term>
+ <listitem><para>The record is returned in a display friendly
+ format. Upon completion buffer is returned
+ (type <literal>char *</literal>) and length is stored in
+ <literal>*len</literal>.
+ </para></listitem>
+ </varlistentry>
+ <varlistentry><term><literal>raw</literal></term>
+ <listitem><para>The record is returned in the internal
+ YAZ specific format. The raw data is returned as type
+ <literal>Z_External *</literal> is just the type for
+ the member <literal>retrievalRecord</literal> in
+ type <literal>NamePlusRecord</literal>.
+ </para></listitem>
+ </varlistentry>
+ </variablelist>
</para>
+ <sect2><title>Protocol behavior</title>
+ <para>
+ The functions <function>Z3950_resultset_record</function> and
+ <function>Z3950_resultset_records</function> inspects the client-side
+ record cache. If the records(s) were not found, i.e. not yet retrieved
+ from, they are fetched using Present Requests.
+ </para>
+ </sect2>
</sect1>
<sect1 id="zoom.options"><title>Options</title>
<para>
- Most &zoom; objects provide a way to specify options to default behaviour.
+ Most &zoom; objects provide a way to specify options to default behavior.
From an implementation point of view a set of options is just like
an associate array / hash array, etc.
</para>
occurred for connection <literal>cs[n-1]</literal>.
When no events are pending for the connections, a value of zero is
returned.
- To make sure all outstanding requests are performed call this function
+ To ensure that all outstanding requests are performed call this function
repeatedly until zero is returned.
</para>
</sect1>
/*
- * $Id: zoom-c.c,v 1.3 2001-11-06 17:05:19 adam Exp $
+ * $Id: zoom-c.c,v 1.4 2001-11-11 22:25:25 adam Exp $
*
* ZOOM layer for C, connections, result sets, queries.
*/
#include "zoom-p.h"
+#define USE_POLL 0
+
+#if USE_POLL
+#include <sys/poll.h>
+#endif
+
static Z3950_record record_cache_lookup (Z3950_resultset r,
int pos,
const char *elementSetName);
if (ret >= 0)
{
c->state = STATE_CONNECTING;
- c->mask = Z3950_SELECT_READ | Z3950_SELECT_WRITE | Z3950_SELECT_EXCEPT;
+ c->mask = Z3950_SELECT_READ | Z3950_SELECT_WRITE |
+ Z3950_SELECT_EXCEPT;
return;
}
}
}
return 0;
}
+ else if (!strcmp (type, "raw"))
+ {
+ if (npr->which == Z_NamePlusRecord_databaseRecord)
+ {
+ *len = -1;
+ return (Z_External *) npr->u.databaseRecord;
+ }
+ return 0;
+ }
return 0;
}
else if (r == 1)
{
c->state = STATE_ESTABLISHED;
- c->mask = Z3950_SELECT_READ|Z3950_SELECT_WRITE;
+ c->mask = Z3950_SELECT_READ|Z3950_SELECT_WRITE|Z3950_SELECT_EXCEPT;
}
else
{
c->state = STATE_ESTABLISHED;
- c->mask = Z3950_SELECT_READ;
+ c->mask = Z3950_SELECT_READ|Z3950_SELECT_EXCEPT;
}
return 0;
}
{
#if 0
int r = cs_look(c->cs);
- yaz_log (LOG_DEBUG, "Z3950_connection_do_io c=%p mask=%d cs_look=%d",
+ yaz_log (LOG_LOG, "Z3950_connection_do_io c=%p mask=%d cs_look=%d",
c, mask, r);
if (r == CS_NONE)
}
else if (r == CS_CONNECT)
{
- yaz_log (LOG_DEBUG, "calling rcvconnect");
+ yaz_log (LOG_LOG, "calling rcvconnect");
if (cs_rcvconnect (c->cs) < 0)
{
c->error = Z3950_ERROR_CONNECT;
int Z3950_event (int no, Z3950_connection *cs)
{
+#if USE_POLL
+ struct pollfd pollfds[1024];
+ Z3950_connection poll_cs[1024];
+#else
struct timeval tv;
fd_set input, output, except;
- int i, r;
+#endif
+ int i, r, nfds;
int max_fd = 0;
for (i = 0; i<no; i++)
}
}
+#if USE_POLL
+
+#else
tv.tv_sec = 15;
tv.tv_usec = 0;
FD_ZERO (&input);
FD_ZERO (&output);
FD_ZERO (&except);
- r = 0;
+#endif
+ nfds = 0;
for (i = 0; i<no; i++)
{
Z3950_connection c = cs[i];
continue;
if (max_fd < fd)
max_fd = fd;
+
+#if USE_POLL
+ if (mask)
+ {
+ short poll_events = 0;
+
+ if (mask & Z3950_SELECT_READ)
+ poll_events += POLLIN;
+ if (mask & Z3950_SELECT_WRITE)
+ poll_events += POLLOUT;
+ if (mask & Z3950_SELECT_EXCEPT)
+ poll_events += POLLERR;
+ pollfds[nfds].fd = fd;
+ pollfds[nfds].events = poll_events;
+ pollfds[nfds].revents = 0;
+ poll_cs[nfds] = c;
+ nfds++;
+ }
+#else
if (mask & Z3950_SELECT_READ)
{
FD_SET (fd, &input);
- r++;
+ nfds++;
}
if (mask & Z3950_SELECT_WRITE)
{
FD_SET (fd, &output);
- r++;
+ nfds++;
}
if (mask & Z3950_SELECT_EXCEPT)
{
FD_SET (fd, &except);
- r++;
+ nfds++;
}
+#endif
}
- if (!r)
+ if (!nfds)
{
for (i = 0; i<no; i++)
{
yaz_log (LOG_DEBUG, "no more events");
return 0;
}
+#if USE_POLL
+ yaz_log (LOG_LOG, "poll start");
+ r = poll (pollfds, nfds, 15000);
+ yaz_log (LOG_LOG, "poll stop, returned r=%d", r);
+ for (i = 0; i<nfds; i++)
+ {
+ Z3950_connection c = poll_cs[i];
+ if (r && c->mask)
+ {
+ int mask = 0;
+ if (pollfds[i].revents & POLLIN)
+ mask += Z3950_SELECT_READ;
+ if (pollfds[i].revents & POLLOUT)
+ mask += Z3950_SELECT_WRITE;
+ if (pollfds[i].revents & POLLERR)
+ mask += Z3950_SELECT_EXCEPT;
+ if (mask)
+ Z3950_connection_do_io(c, mask);
+ }
+ else if (r == 0 && c->mask)
+ {
+ /* timeout and this connection was waiting */
+ c->error = Z3950_ERROR_TIMEOUT;
+ c->event_pending = 1;
+ }
+ }
+#else
yaz_log (LOG_DEBUG, "select start");
r = select (max_fd+1, &input, &output, &except, &tv);
yaz_log (LOG_DEBUG, "select stop, returned r=%d", r);
-
for (i = 0; i<no; i++)
{
Z3950_connection c = cs[i];
c->event_pending = 1;
}
}
+#endif
+
for (i = 0; i<no; i++)
{
Z3950_connection c = cs[i];