dhewm3/neo/openal/docs/alc-context.sgml
Timothee 'TTimo' Besset fb1609f554 hello world
2011-11-22 15:28:15 -06:00

846 lines
33 KiB
Text

<chapter id="alc">
<title>AL Contexts and the ALC API</title>
<para>
This section of the AL specification describes ALC, the AL Context API.
ALC is a portable API for managing &AL; contexts, including
resource sharing, locking, and unlocking. Within the core AL API the
existence of a Context is implied, but the Context is not exposed. The
Context encapsulates the state of a given intance of the AL state machine.
</para>
<para>
To avoid confusion with the AL related prefixes implied throughout
this document, the "alc" and "ALC_" prefixes have been made explicit
in the ALC related sections.
</para>
<para>
ALC defines the following objects: Contexts.
</para>
<![ %Annote [
<note><title>Annotation (ALC entry points)</title><para>
While the actual ALC implemention might be supplied
as a separate library, or as part of a server or
daemon, the specification requires that the AL library
provides the actual ALC entry points.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (ALC OS independend)</title><para>
ALC is meant to be OS-independent. OS specifics are
expected to be adressed by defining proper device
specifiers strings, and configuration attributes.
In this, ALC differs from GLX/WGL, which (due to the
tighter coupling with the window manager and
operating system) attempt to abstract OS specifics
to a much lesser degree.
</para></note>
]]>
<![ %RFC [
<note id="rfc-bk001024-02"><title>RFC/ Issues </title><para>
Hardware init/deinit, mapping many processes with
multiple contexts to many devices, sharing resources
among processes.
</para></note>
]]>
<sect1>
<title>Managing Devices</title>
<para>
ALC introduces the notion of a Device. A Device can be,
depending on the implementation, a hardware device, or
a daemon/OS service/actual server. This mechanism also
permits different drivers (and hardware) to coexist
within the same system, as well as allowing several
applications to share system resources for audio, including
a single hardware output device. The details are left
to the implementation, which has to map the available
backends to uniq7ue device specifiers (represented as
strings).
</para>
<![ %Annote [
<note><title>Annotation (Network transparency)</title><para>
&AL; is meant for interoperability with &OGL;. Some
implementations of &OGL; bindings (e.g. GLX) are network
transparent. The Device API theoretically allows for
a network transparent AL implementation. No wire protocol
is specified, no specification or implementation is planned.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (Device Enumeration)</title><para>
At this time, ALC does not provide mechanism to query
for available devices, and request device enumerations.
This might be added at a later time, depending on demand
and the ability to abstract OS and configuration specifics.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (X11 Audio)</title><para>
The ALC API intentionally mimicks XOpenDisplay
and XCloseDisplay. There is no X Audio standard,
although proposals have been made in the past.
The ALC API design accounts for this possibility
in a minimal way.
</para></note>
]]>
<sect2>
<title>Connecting to a Device</title>
<para>
The alcOpenDevice function allows the application (i.e. the
client program) to connect to a device (i.e. the server).
<funcsynopsis><funcprototype>
<funcdef> &device; * <function>alcOpenDevice</function></funcdef>
<paramdef> const &ubyte;* <parameter>deviceSpecifier</parameter></paramdef>
</funcprototype></funcsynopsis>
If the function returns NULL, then no sound driver/device has
been found. The argument is a null terminated string that requests
a certain device or device configuration. If NULL is specified,
the implementation will provide an implementation specific default.
</para>
<![ %Annote [
<note><title>Annotation (Operating system dependencies)</title><para>
At this point, system specific configuration, and operating system
specific details, are handled by leaving the details of the
string specifier to the implementation. The application coder
has to determine how he wants to obtain this information from the
OS or the user. If, at a later point, device enumeration and
configuration requests are supported through ALC, the resulting
string might still be operating system and implementation specific.
</para></note>
]]>
</sect2>
<sect2>
<title>Disconnecting from a Device</title>
<para>
The alcCloseDevice function allows the application (i.e. the
client program) to disconnect from a device (i.e. the server).
<funcsynopsis><funcprototype>
<funcdef> &void; <function>alcCloseDevice</function></funcdef>
<paramdef> &device; * <parameter>deviceHandle</parameter></paramdef>
</funcprototype></funcsynopsis>
If deviceHandle is NULL or invalid, an ALC_INVALID_DEVICE error
will be generated. Once closed, a deviceHandle is invalid.
</para>
</sect2>
</sect1>
<sect1>
<title>Managing Rendering Contexts</title>
<para>
All operations of the AL core API affect a current AL context.
Within the scope of AL, the ALC is implied - it is not visible as
a handle or function parameter. Only one AL Context per INprocess
can be current at a time. Applications maintaining multiple
AL Contexts, whether threaded or not, have to set the current
context accordingly. Applications can have multiple threads
that share one more or contexts. In other words, AL and ALC are
threadsafe.
</para>
<para>
The default AL Context interoperates with a hardware device driver.
The application manages hardware and driver resources by
communicating through the ALC API, and configures and uses
such Contexts by issuing AL API calls. A default AL Context
processes AL calls and sound data to generate sound output.
Such a Context is called a Rendering Context. There might
be non-rendering contexts in the future.
</para>
<para>
The word "rendering" was chosen intentionally to emphasize the
primary objective of the AL API - spatialized sound - and the
underlying concept of AL as a sound synthesis pipeline that
simulates sound propagation by specifying spatial arrangements
of listeners, filters, and sources. If used in describing an
application that uses both &OGL; and &AL;, "sound rendering
context" and "graphics rendering context" should be used for
clarity. Throughout this document, "rendering" is used
to describe spatialized audio synthesis (avoiding ambiguous
words like "processing", as well as proprietary and restrictive
terms like "wavetracing").
</para>
<sect2>
<title>Context Attributes</title>
<para>
The application can choose to specify certain attributes for
a context. Attributes not specified explicitely are set to
implementation dependend defaults.
</para>
<para>
<table>
<title>Context Attributes</title>
<tgroup cols="2" align="left" colsep=1 rowsep=1>
<colspec colname=c1>
<colspec colname=c2>
<thead>
<row>
<entry>Name</>
<entry>Description</>
</row>
</thead>
<tbody>
<row>
<entry>ALC_FREQUENCY</>
<entry>Frequency for mixing output buffer, in units of Hz.</>
</row>
<row>
<entry>ALC_REFRESH</>
<entry>Refresh intervalls, in units of Hz.</>
</row>
<row>
<entry>ALC_SYNC</>
<entry>Flag, indicating a synchronous context.</>
</row>
</tbody>
</tgroup>
</table>
</para>
<![ %Annote [
<note><title>Annotation (Refresh Control)</title><para>
Applications might have a fixed, or bounded, schedule for
state changes (e.g. synchronously with the GL framerate).
In this case it is desirable to specify the mixahead
interval (milliseconds), or refresh rate (Hz), for the
mixing thread. This is especially important for a synchronous
context, where the application has to specify the refresh
interval it intends to keep.
</para></note>
]]>
<![ %RFC [
<note id="rfc-bk001026-05"><title>RFC / Mixing Buffer Configuration </title><para>
ALC_RESOLUTION was originally used to specify the accuracy of the
mixing output buffer. For the time being this is not supported, but
could be added if mixing path and result accuracy control is desirable
to scale resource requirements.
A full ALC_FORMAT (including channel and other attributes) does not
make sense for rendering contexts, but will be necessary for PBuffers
(mix to memory).
</para></note>
]]>
<![ %RFC [
<note id="rfc-bk001026-01"><title>RFC / LOKI extensions </title><para>
<programlisting>
ALC_SOURCES_HINT_LOKI, /* # of sources to pre-allocate */
ALC_BUFFERS_HINT_LOKI, /* # of buffers to pre-allocate */
ALC_CD_LOKI, /* demand CD/DVD control */
</programlisting>
</para></note>
]]>
</sect2>
<sect2>
<title>Creating a Context</title>
<para>
A context is created using alcCreateContext. The device parameter
has to be a valid device. The attribute list can be NULL,
or a zero terminated list of integer pairs composed of valid
ALC attribute tokens and requested values.
<funcsynopsis><funcprototype>
<funcdef> &context; * <function> alcCreateContext </function></funcdef>
<paramdef> const &device; * <parameter> deviceHandle </parameter></paramdef>
<paramdef> &int; * <parameter> attrList </parameter></paramdef>
</funcprototype></funcsynopsis>
Context creation will fail
if the application requests attributes that, by themselves,
can not be provided. Context creation will fail if the
combination of specified attributes can not be provided.
Context creation will fail if a specified attribute, or
the combination of attributes, does not match the default
values for unspecified attributes.
</para>
</sect2>
<sect2>
<title>Selecting a Context for Operation</title>
<para>
To make a Context current with respect to AL Operation (state
changes by issueing commands), alcMakeContextCurrent is used.
The context parameter can be NULL or a valid context pointer.
The operation will apply to the device that the context was
created for.
<funcsynopsis><funcprototype>
<funcdef> &bool; <function> alcMakeContextCurrent </function></funcdef>
<paramdef> &context; * <parameter> context </parameter></paramdef>
</funcprototype></funcsynopsis>
For each OS process (usually this means for each application), only
one context can be current at any given time. All AL commands apply
to the current context. Commands that affect objects shared among
contexts (e.g. buffers) have side effects on other contexts.
</para>
<![ %Annote [
<note><title>Annotation (No Explicit Device)</title><para>
An ALC context is bound to the device it was created for.
The context carries this information, thus removing the
need to specify the device explicitely. Contexts can not
be made current for any other device aside from the one
they were created for.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (No Multiple Current)</title><para>
There is only one current context per process, even
in multithreaded applications, even if multiple
devices are used.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (Current NULL)</title><para>
The implementation is encouraged to exploit optimizations
possible if the application sets the current context
to NULL, indicating that no state changes are intended
for the time being. The application should not set the
current context to NULL if more state changes are
pending on the most recent, or another context created
for the same device.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (Shared Objects)</title><para>
Buffers are shared among contexts. As mutiple contexts
can exist at the same time, the state of shared objects
is also shared among contexts.
</para></note>
]]>
<![ %RFC [
<note id="rfc-bk001026-06"><title>RFC / Buffer Deletion </title><para>
Buffers that have not yet been processed in another context
can not be deleted from any other context.
</para></note>
]]>
</sect2>
<sect2>
<title>Initiate Context Processing</title>
<para>
The current context is the only context accessible to state
changes by AL commands (aside from state changes affecting
shared objects). However, multiple contexts can be processed
at the same time. To indicate that a context should be
processed (i.e. that internal execution state like offset
increments are supposed to be performed), the application
has to use alcProcessContext.
<funcsynopsis><funcprototype>
<funcdef> &void; <function>alcProcessContext</function></funcdef>
<paramdef> &context; * <parameter> context </parameter></paramdef>
</funcprototype></funcsynopsis>
Repeated calls to alcProcessContext are legal, and do not
affect a context that is already marked as processing. The
default state of a context created by alcCreateContext is
that it is not marked as processing.
</para>
<![ %Annote [
<note><title>Annotation (Sync and async implementations)</title><para>
Unfortunately, the exact semantics of alcProcessContext is
not independend of the implementation. Ideally it should be completely
transparent to the application whether the sound driver is threaded or
synced. Unfortunately a synced context has to have its execution
initiated by the application, which requires calls of
alcProcessContext timed in accordance to the drivers mixahead,
or the rendering buffer will underflow. For a threaded driver,
the implementation is free to consider alcProcessContext
a NOP once the context has been marked as processing.
</para><para>
One consequence is that an application that was developed
using a threaded implementation of AL might not work
properly with a synchronous implementation of AL (on
the other hand, an AL application that works using a
synchronous implementation is guaranteed to work with
a threaded implementation.
</para><para>
Enforcing alcProcessContext calls would defeat the purpose
of a threaded implementation. Permitting the AL implementation
to e.g. schedule optimizations based on alcProcessContext
calls would similarly obfuscate the exact semantincs.
Consequently, the application coder has to accept this
implementation dependency, and has to rely on the ALC_SYNC
attribute to explicitely request a synchronous implementation.
The implementation can expect the application to be aware
of the additonal constraints imposed on alcProcessContext in
this case.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (Multiple Contexts and SYNC refresh)</title><para>
The application can request SYNC contexts or threaded contexts,
however, the implementation is not obliged to provide both, or provide
a mixture of both on the same device.
</para></note>
]]>
<![ %RFC [
<note id="rfc-bk001024-03"><title>RFC/ Implications for Networking</title><para>
What does alcProcessContext imply for networking?
For AL, will we add alFlush/alFinish?
</para></note>
]]>
</sect2>
<sect2>
<title>Suspend Context Processing</title>
<para>
The application can suspend any context from processing
(including the current one). To indicate that a context should be
suspended from processing (i.e. that internal execution state like
offset increments is not supposed to be changed), the application
has to use alcSuspendContext.
<funcsynopsis><funcprototype>
<funcdef> &void; <function>alcSuspendContext</function></funcdef>
<paramdef> &context; * <parameter> context </parameter></paramdef>
</funcprototype></funcsynopsis>
Repeated calls to alcSuspendContext are legal, and do not
affect a context that is already marked as suspended. The
default state of a context created by alcCreateContext is
that it is marked as suspended.
</para>
<![ %Annote [
<note><title>Annotation (Sync and async implementations)</title><para>
Unfortunately, the exact semantics of alcSuspendContext is
also not independend of the implementation. For a threaded
implementation, alcSuspendContext is necessary to ensure a context
is not processed. For a synchronous implementation, omitting
alcProcessContext calls will ultimately have the same effect,
but will also generate rendering buffer underflow errors.
Again, the application coder that requests a synchronous
context using ALC_SYNC has to make sure that alcSuspendContext
is used accordingly.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (Suspending vs. Muting a context)</title><para>
By setting Listener GAIN to zero, an application can mute a
context, and expect the implementation to bypass all rendering.
However, the context is still processing, and the internal
execution state is still updated accordingly. Suspending a
context, whether muted or not, will incidentally suspend rendering
as well. However, it is the application's responsibility to
prevent artifacts (e.g. by proper GAIN control to fade in and
out). It is recommended to mute a context before suspending.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (Current Context Suspended)</title><para>
It is possible to make a suspended context current, or
suspend the current context. In this case, the implementation
is still obliged to immediately verify AL commands as they
are issued, and generate errors accordingly. The implementation
is permitted to postpone propagating the actual state changes
until the context is marked for processing again, with the
exception of dereferencing data (e.g. buffer contents).
For efficiency reasons (memory usage), most if not all
AL commands applied to a suspended context will usually
be applied immediately. State changes will have to be applied
in the sequence they were requested. It is possible to
use suspension of a current context as an explicit locking
(to enforce apparent synchronicity), but execution is still
guaranteed to be in sequence, and the implementation is not
expected to optimize this operation. A typical use would
be setting up the initial configuration while loading a
scene.
</para></note>
]]>
<![ %Annote [
<note><title>Annotation (Release of Hardware Resources)</title><para>
The specification does not guarantee that the implementation
will release hardware resources used by a suspended context.
This might well depend on the details of the hardware and driver.
Neither a muted context nor a suspended context can be expected
to free device resources. If all contexts for a given device
are suspended, and no context of this device is current, the
implementation is expected to release all hardware resources if possible.
</para></note>
]]>
</sect2>
<sect2>
<title>Destroying a Context</title>
<para>
<funcsynopsis><funcprototype>
<funcdef> &void; <function> alcDestroyContext </function></funcdef>
<paramdef> &context; * <parameter> context </parameter></paramdef>
</funcprototype></funcsynopsis>
The correct way to destroy a context is to first release it using
alcMakeCurrent and NULL. Applications should not attempt to destroy
a current context.
</para>
</sect2>
</sect1>
<sect1>
<title>ALC Queries</title>
<sect2>
<title>Query for Current Context</title>
<para>
The application can query for, and obtain an handle to, the current
context for the application. If there is no current context, NULL is
returned.
<funcsynopsis><funcprototype>
<funcdef> &context; * <function>alcGetCurrentContext</function></funcdef>
<void>
</funcprototype></funcsynopsis>
</para>
</sect2>
<sect2>
<title>Query for a Context's Device</title>
<para>
The application can query for, and obtain an handle to, the device
of a given context.
<funcsynopsis><funcprototype>
<funcdef> &device; * <function>alcGetContextsDevice</function></funcdef>
<paramdef> &context; * <parameter> context </parameter></paramdef>
</funcprototype></funcsynopsis>
</para>
</sect2>
<sect2>
<title>Query For Extensions</title>
<para>
To verify that a given extension is available for the current
context and the device it is associated with, use
<funcsynopsis>
<funcprototype>
<funcdef>&bool;<function>IsExtensionPresent</function></funcdef>
<paramdef> const &device; * <parameter>deviceHandle</parameter></paramdef>
<paramdef>const &ubyte; * <parameter> extName </parameter></paramdef>
</funcprototype>
</funcsynopsis>
A NULL name argument returns FALSE, as do invalid and unsupported string
tokens. A NULL deviceHandle will result in an INVALID_DEVICE error.
<![ %Annote [
<note><title>Annotation (Exlicit Device Parameter)</title><para>
Certain ALC Extensions might be relevant to context creation
(like additional attributes, or support for unusual multi-context
combinations), thus the application might have to query these
before a context is created. On the other hand, ALC Extensions
are specific to the device.
</para></note>
]]>
</para>
</sect2>
<sect2>
<title>Query for Function Entry Addresses</title>
<para>
The application is expected to verify the applicability of
an extension or core function entry point before requesting
it by name, by use of alcIsExtensionPresent.
<funcsynopsis><funcprototype>
<funcdef> &void; * <function> alcGetProcAddress</function></funcdef>
<paramdef> const &device; * <parameter>deviceHandle</parameter></paramdef>
<paramdef> const &ubyte; * <parameter> funcName </parameter></paramdef>
</funcprototype></funcsynopsis>
Entry points can be device specific, but are not context specific.
Using a NULL device handle does not guarantee that the entry
point is returned, even if available for one of the available
devices. Specifying a NULL name parameter will cause an
ALC_INVALID_VALUE error.
</para>
</sect2>
<sect2>
<title>Retrieving Enumeration Values</title>
<para>
Enumeration/token values are device independend, but tokens
defined for extensions might not be present for a given device.
Using a NULL handle is legal, but only the tokens defined by
the AL core are guaranteed. Availability of extension tokens
dependends on the ALC extension.
<funcsynopsis><funcprototype>
<funcdef> &uint; <function> alcGetEnumValue </function></funcdef>
<paramdef> const &device; * <parameter>deviceHandle</parameter></paramdef>
<paramdef> const &ubyte; <parameter> enumName </parameter></paramdef>
</funcprototype></funcsynopsis>
Specifying a NULL name parameter will cause an
ALC_INVALID_VALUE error.
</para>
</sect2>
<sect2>
<title>Query for Error Conditions</title>
<para>
ALC uses the same conventions and mechanisms as AL for error
handling. In particular, ALC does not use conventions
derived from X11 (GLX) or Windows (WGL).
The alcGetError function can be used to query ALC errors.
<funcsynopsis><funcprototype>
<funcdef>&enum;<function>alcGetError</function></funcdef>
<paramdef> &device; * <parameter>deviceHandle</parameter></paramdef>
</funcprototype></funcsynopsis>
Error conditions are specific to the device.
</para>
<para>
<table>
<title>Error Conditions</title>
<tgroup cols="2" align="left" colsep=1 rowsep=1>
<colspec colname=c1>
<colspec colname=c2>
<thead>
<row>
<entry>Name</>
<entry>Description</>
</row>
</thead>
<tbody>
<row>
<entry>ALC_NO_ERROR</>
<entry>The device handle or specifier does name an accessible driver/server.</>
</row>
<row>
<entry>ALC_INVALID_DEVICE</>
<entry>The Context argument does not name a valid context.</>
</row>
<row>
<entry>ALC_INVALID_CONTEXT</>
<entry>The Context argument does not name a valid context.</>
</row>
<row>
<entry>ALC_INVALID_ENUM</>
<entry>A token used is not valid, or not applicable.</>
</row>
<row>
<entry>ALC_INVALID_VALUE</>
<entry>An value (e.g. attribute) is not valid, or not applicable.</>
</row>
</tbody>
</tgroup>
</table>
</para>
<![ %Annote [
<note><title>Annotation (No UNDERFLOW error)</title><para>
Applications using synchronous (and, depending on CPU
load, even an asynchronous implementation itself) might
fail to prevent underflow of the rendering output
buffer. No ALC error is generated in these cases,
as it this error condition can not be applied to a
specific command.
</para></note>
]]>
</sect2>
<sect2>
<title>String Query</title>
<para>
The application can obtain certain strings from ALC.
<funcsynopsis><funcprototype>
<funcdef>const &ubyte; * <function>alcGetString</function></funcdef>
<paramdef> &device; * <parameter>deviceHandle</parameter></paramdef>
<paramdef> &enum; <parameter> token </parameter></paramdef>
</funcprototype></funcsynopsis>
For some tokens, NULL is is a legal value for the deviceHandle.
In other cases, specifying a NULL device will generate an
ALC_INVALID_DEVICE error.
</para>
<para>
<table>
<title>String Query Tokens</title>
<tgroup cols="2" align="left" colsep=1 rowsep=1>
<colspec colname=c1>
<colspec colname=c2>
<thead>
<row>
<entry>Name</>
<entry>Description</>
</row>
</thead>
<tbody>
<row>
<entry>ALC_DEFAULT_DEVICE_SPECIFIER</>
<entry>The specifier string for the default device (NULL handle is legal).</>
</row>
<row>
<entry>ALC_DEVICE_SPECIFIER</>
<entry>The specifier string for the device (NULL handle is not legal).</>
</row>
<row>
<entry>ALC_EXTENSIONS</>
<entry>The extensions string for diagnostics and printing.</>
</row>
</tbody>
</tgroup>
</table>
</para>
<para>
In addition, printable error message strings are provided for
all valid error tokens, including ALC_NO_ERROR, ALC_INVALID_DEVICE,
ALC_INVALID_CONTEXT, ALC_INVALID_ENUM, ALC_INVALID_VALUE.
</para>
</sect2>
<sect2>
<title>Integer Query</title>
<para>
The application can query ALC for information using an integer
query function.
<funcsynopsis><funcprototype>
<funcdef> &void; <function>alcGetIntegerv</function></funcdef>
<paramdef> &device; * <parameter>deviceHandle</parameter></paramdef>
<paramdef> &enum; <parameter> token </parameter></paramdef>
<paramdef> &sizei; <parameter> size </parameter></paramdef>
<paramdef> &int; <parameter> dest </parameter></paramdef>
</funcprototype></funcsynopsis>
For some tokens, NULL is a legal deviceHandle. In other
cases, specifying a NULL device will generate an ALC_INVALID_DEVICE
error. The application
has to specify the size of the destination buffer provided.
A NULL destination or a zero size parameter will cause ALC to ignore
the query.
</para>
<para>
<table>
<title>Integer Query Tokens</title>
<tgroup cols="2" align="left" colsep=1 rowsep=1>
<colspec colname=c1>
<colspec colname=c2>
<thead>
<row>
<entry>Name</>
<entry>Description</>
</row>
</thead>
<tbody>
<row>
<entry>ALC_MAJOR_VERSION</>
<entry>Major version query.</>
</row>
<row>
<entry>ALC_MINOR_VERSION</>
<entry>Minor version query.</>
</row>
<row>
<entry>ALC_ATTRIBUTES_SIZE</>
<entry>The size required for the zero-terminated attributes
list, for the current context. NULL is an invalid
device. NULL (no current context for the
specified device) is legal.</>
</row>
<row>
<entry>ALC_ALL_ATTRIBUTES</>
<entry>Expects a destination of ALC_CURRENT_ATTRIBUTES_SIZE,
and provides the attribute list for the current
context of the specified device. NULL is an invalid
device. NULL (no current context for the
specified device) will return the default attributes
defined by the specified device.</>
</row>
</tbody>
</tgroup>
</table>
</para>
<![ %Annote [
<note><title>Annotation (Backward Compatibility)</title><para>
Backward compatibility is guaranteed only for minor revisions.
Breaking ABI backwards compatibility will require a issuing
major revision.
</para></note>
]]>
<![ %RFC [
<note id="rfc-bk001026-02"><title>RFC / Version Matching </title><para>
The ALC version can be different from the AL version.
The ALC major version has to be identical between
application and driver (client and server).
The ALC minor version can differ between client and server,
and the minimum of the two minor version numbers is returned.
</para></note>
]]>
<![ %RFC [
<note id="rfc-bk001027-03"><title>RFC / Device Enumeration </title><para>
ALC_NUM_DEVICE_SPECIFIERS could be provided as an extension,
but it requires the number of device configurations to be
finite and small, as they are to be represented by strings.
</para></note>
]]>
</sect2>
</sect1>
<sect1>
<title>Shared Objects</title>
<para>
For efficiency reasons, certain AL objects are shared across
ALC contexts. At this time, AL buffers are the only shared
objects.
</para>
<![ %RFC [
<note id="rfc-bk001024-01"><title>RFC/ Sharing Sources?</title><para>
Mike Kelly from Dynamix describes an application scenario with
multiple camera views of the same scene (implemented using
several viewports on a single, "split screen", context)
in which he needs multiple listeners (one per view) to
preceive the same sources, with the rendering results of
each listener to mixed for a single output device. As an
alternative to permitting multiple listeners within a context,
it might be preferable to introduce a mechanism to explicitely
share certain AL objects among contexts, at a later time.
It is not desirable to share Sources by default, especially
as the more common multi-listener application is expected to
maintain listeners with disjunct sets of sources.
</para></note>
]]>
<sect2>
<title>Shared Buffers</title>
<para>
Buffers are shared among contexts. The processing state of a buffer
is determined by the dependencies imposed by all contexts, not just
the current context. This includes suspended contexts as well as
contexts that are processing.
</para>
</sect2>
</sect1>
</chapter>