Bovs
Version 2.0
Bryan's Overlay Supervisor
Copyright 1992 Bryan Ford
Disclaimer
~~~~~~~~~~
This program is provided "as is" without warranty of any kind, either
expressed or implied, including, but not limited to, the implied warranty
of fitness for a particular purpose. The entire risk as to the results,
reliability and performance of this program is assumed by you.
Distribution
~~~~~~~~~~~~
Bovs may be freely distributed as long as it is complete (none of the
files listed below are left out), all of the files are unmodified, and no
charge is made for such distribution other than a small fee to cover the
cost of copying. (In effect, no profit may be made through distribution
of
this program). You may modify this code for your own personal use, but
you
may not distribute modified versions. (If you improve Bovs, send your
modifications back to me and I'll incorporate them into the master,
giving
credit to you of course. This way we won't have hundreds of different
versions of Bovs floating around.)
You may use the object code derived from this source code, or your
own
modified version of this source code, in your own public domain,
freeware,
shareware, or otherwise freely distributable programs, as long as you
visibly state that you used Bovs and mention the name of its author
(Bryan
Ford) somewhere in your documentation. You do not need to obtain any
kind
of license or special permission to use it in freely distributable
software.
If you want to distribute Bovs or any modified version of it, in
either
source or object form, with a commercial package, you must send me a copy
of that commercial package as soon as it is released, as well as any
major
updates to the program released in the future. (This only applies for as
long as the program uses Bovs.) Other than that, there is no cost for
using Bovs in a commercial program.
The Bovs distribution must contain the following files, complete and
unmodified:
Bovs.doc
Bovs.o
OVERLAY=1)
Bst.o
OVERLAY=0)
Bovs.h
Bovs.asm
Bovs documentation (this file)
Linkable version of Bovs (assembled with
Same, without overlay support (assembled with
Header file for SAS/C programs
Source code for Bovs (A68k)
MemMan/MemMan.doc
MemMan/memman.library
MemMan/MemMan.o
MemMan/MemMan.fd
MemMan/MemMan.h
MemMan/MemMan.i
MemMan/MemMan.asm
MemMan/MemManLib.asm
MemMan/Macros.i
MemMan documentation
MemMan runtime library
Linkable version of MemMan
Entrypoints for memman.library
Header for C programs using MemMan
Same, for assembly language
Source to MemMan (A68k)
Source to memman.library interface (A68k)
Macros you'll need for the above two
MemMan/LMKfile
MemMan/test
MemMan/test.c
SAS/C makefile for MemMan
Program to test memman.library
Source for MemMan test program
files
Introduction
~~~~~~~~~~~~
Bovs (pronounced "boffs") is an overlay system designed to replace
the
standard overlay system supplied with SAS/C. It acts as both overlay
supervisor and startup code. It provides many powerful features that the
standard SAS/C startup code and overlay supervisor don't provide. It is
intended especially for large and medium-size applications which need
flexible and efficient use of memory. It is used in my Shareware music
player, MultiPlayer. For smaller programs, there is an alternative form
of
Bovs without the overlay system, but with all of Bovs's other features.
Bovs currently requires SAS/C's linker, BLink (or a very compatible
linker, if there is such), to create overlaid files. However, other than
that, SAS/C is not necessarily needed: you can write programs that use
Bovs
in assembly language or any other language that supports register
arguments
and the standard Amiga object file format.
Programs that use Bovs can be started from Workbench or the CLI.
Both
1.3 and 2.0 are supported, but when run under 2.0 Bovs supports automatic
command-line parsing with ReadArgs(). If started from the CLI, Bovs
detaches and runs your program in a separate process, freeing up the
original CLI.
Instead of the standard, rigidly-structured, hierarchially-organized
overlay system that standard Amiga overlay systems use, Bovs uses a very
flexible overlay method based on dynamic memory caching. When overlaid
code is called, the overlay node in which it is contained is loaded in if
necessary, and the function is called. However, overlay nodes are not
flushed from memory when other overlay nodes at the same level are
called.
(In fact, Bovs doesn't even use overlay "levels".) Instead, they remain
in
memory and immediately available until the memory they occupy is needed
for
something else.
This is where MemMan comes in. The Bovs distribution includes
MemMan,
a low-memory manager also written by myself, which allows programs to be
called when any call to AllocMem() is about to fail, so they can free
memory they don't need. (Note that while Bovs currently only supports
the
BLink that comes with SAS/C, MemMan can be used by many compilers. See
MemMan's documentation for details.) Bovs uses MemMan to unload overlay
nodes that aren't currently in use when the system is about to run out of
memory. Your program can also use MemMan alongside Bovs.
If the user of a Bovs program has lots of memory, Bovs ends up acting
as a "delayed-loading" system, to speed up initial loading and load other
parts of the program only when necessary. After all parts of the program
are loaded, if the user never runs out of memory, no more loading is done
and calls to overlaid functions are instantaneous. However, if the user
does run out of memory, Bovs can immediately free inactive overlay nodes,
which gives the system some extra "padding" and helps prevent
memory-shortage problems.
To keep track of which overlay nodes are in use and which may be
freed
on demand, Bovs uses a system of nestable locks associated with each
overlay node. When an overlaid function is called, locking is handled
automatically: Bovs locks the overlay node before calling the function,
and unlocks it after the function returns. However, your program may
also
specifically lock and unlock overlay nodes to ensure that they remain in
memory at certain times.
Startup Code
~~~~~~~~~~~~
Before getting into Bovs's overlay management, I'll describe how Bovs
handles startup and exit. First of all, Bovs may not be used along with
"c.o" or any of the other standard startup code files. Bovs does not
support SAS's standard C-style memory allocation or file handling
functions, or any other non-trivial standard library functions. (You can
still use string functions and such.)
Bovs is, therefore, "Amiga-only."
Bovs supports auto-detaching. When your program is run from the CLI,
a
separate process is started for your program and the original CLI is
given
back to the user. Unlike typical auto-detaching startup code, however,
Bovs lets the program decide whether or not to detach before actually
splitting off into a separate process. For example, MultiPlayer (which
uses Bovs) provides a "NODETACH" command-line switch which will cause it
to
keep running in the normal CLI. If for some reason you don't even want
the
option to detach, you can reassemble Bovs or Bst with the DETACH flag set
to 0.
Somewhere in the non-overlaid part of your program (probably your
main
module), you'll have to define several variables containing various
pieces
of information Bovs needs during startup. I'll give the C definitions
for
these variables; for assembly language, just add an underscore ('_')
before
the variable names.
First, you need to define a longword variable called "stack" which
contains the stack size required for your program. Note that this value
is
only used if the program was loaded from the CLI and thus was detached
from
the original CLI process. If you want a larger-than-normal stack, make
sure you set this variable AND the stack size specified in the Workbench
icon for your program.
Second, you must define a pointer variable called "progname" which is
a
pointer to a string containing the name for the process created when
auto-detaching from the CLI. This string is also used as the program
name
when Bovs encounters a serious error and must display a message and
terminate before ever calling your program. Like the "stack" variable,
this is not used when the program is run from Workbench.
Third, you need to define a longword variable called "priority" which
contains the Exec priority level at which your program is to run. Unlike
the above variables, this is used when your program is started from
either
the Workbench or the CLI. This priority will be in effect through both
PreStart() and Main(). If run from the CLI without detaching, Bovs will
automatically switch back to the CLI's original priority before exiting.
Finally, you must define three variables which Bovs passes to
AmigaDOS
2.0's ReadArgs() function if your program is run under 2.0. The variable
"argtemplate" must be a pointer to a string containing the standard
argument template. For example, part of MultiPlayer's template is
"DIRECTORY,PLAY=MODULES/M,PROG=PROGRAM/K". The variable "argarray" must
be
an array of longwords (it's often convenient to define it as a structure
in
C) which AmigaDOS 2.0 will fill in with the parameters from the parsed
argument template. The variable "argexthelp" must contain a pointer to
an
extended help string which will be displayed to the user as a prompt if
'?'
is entered as an argument on the command line. (If this pointer is NULL,
the argtemplate will be used for this purpose.) I won't go into the
details of argument parsing; Commodore's documentation should suffice.
In addition to these six variables, you must define two functions in
your main program. First, the PreStart() function ("@PreStart" in
assembly
language) must be defined like this:
void __regargs PreStart(int arglen,char *argstr)
D0
A0
(You may omit the "__regargs" keyword if you are compiling with
register arguments as the default.) This function will always be called
before Main(). If your program was run from the CLI, PreStart() will be
called before detaching, while Main() will be called after detaching. At
this point, all other variables have been set up and the command line has
been parsed (if your program was run under 2.0). You may call BExit()
and
BRExit() from this routine; if you do this, and your program was called
from the CLI, you will never detach from the original CLI. Otherwise, if
you return from this function normally, Bovs will detach if necessary and
then call Main().
Take MultiPlayer as an example of the use of PreStart(). It contains
a
"NODETACH/K" entry in its argument template. MultiPlayer's PreStart()
function checks for this switch, and if the user specified it on the
command line, PreStart() simply calls Main() immediately, and when it
returns, it calls BExit() or BRExit(). Note that if Bovs calls
PreStart(),
unless you call BExit() or BRExit() from within that function, Bovs will
AlWAYS also call Main() at a later point, even if it runs into a fatal
error between the two calls. If it can't detach for some reason, it
simply
remains attached and calls Main() directly. That way, you can set things
up (allocate memory, etc.) in PreStart() without having to worry that
Main() will never get called for some reason.
The Main() function ("@Main" in assembly language) is the normal
entrypoint of your program. It must be defined like this:
void __regargs Main(int arglen,char *argstr)
D0
A0
When either PreStart() or Main() is called, "arglen" contains the
length of the unparsed command line received directly from AmigaDOS, and
"argstr" is a pointer to that command line. (Bovs doesn't separate the
command line into separate arguments like the standard startup code does;
the name change from "main" to "Main" is intended to emphasize this
difference.) If your program is called from the Workbench, "arglen" will
contain zero and "argstr" will contain the Workbench startup message.
Note that the raw command line is available on entry even if the
command line was also parsed with 2.0's ReadArgs() function. Under 1.3,
your "argarray" variable will remain unmodified and unused; under 2.0,
you'll get both the parsed and the unparsed arguments. Bovs defines the
long integer symbol "argsparsed" ("_argsparsed" in assembly language),
which it sets to nonzero if the arguments were successfully parsed with
2.0's ReadArgs(). If you want compatibility with both 1.3 and 2.0, you
should use the parameters in "argargarray" if "argsparsed" is nonzero;
otherwise parse the raw command line yourself.
When Main() is called, you don't have any standard input or output
streams, so don't be printing error messages to nonexistant file handles.
(You may use standard I/O from PreStart() if your program was called from
the CLI.) However, your current directory will be whatever the current
directory was in the CLI your program started from, or if you started
from
Workbench, the directory containing your program icon. You don't have to
return to this directory before exiting your program, although you must
not
UnLock() it (Bovs will do that if necessary).
When the Main() function returns, Bovs frees all resources, closes
the
executable file if it was using overlays, and ends the program with a
return code of zero. You can also do this from anywhere in your program
(as long as you're running from the original process Bovs started you in)
by calling BExit() or BRExit(), described later. Remember, Bovs provides
NO resource tracking like standard C startup code does: you must free
anything you allocate. (Of course, Bovs will free anything it
allocates.)
Overlays
~~~~~~~~
First of all, be warned that the use of Bovs overlays is not
completely
transparent. Even if you're already using overlays in your program,
converting it to use Bovs will require a little work. How much it
requires
depends on your program. In this document, I'll assume that you are
trying
to convert a non-overlaid program into an overlaid program that uses
Bovs.
If you have already written an overlaid program, you shouldn't have too
much trouble figuring out what you need to change.
Your program will have to have at least one non-overlaid object file,
as well as one or more overlaid files. You don't have to work out any
complicated overlay call tree or anything: all the overlay nodes are on
the
same overlay "level", and any combination of them may be loaded at one
time. Overlay nodes may even call other overlay nodes, although you will
have to fudge a little to get around some of BLink's error checking.
You'll have to set up a WITH file that you pass to BLink to link your
program. In it, along with whatever other BLink options you want, supply
an OVERLAY construct that looks something like this:
OVERLAY
window.o windowspec.o
progwin.o progwinspec.o
prefswin.o prefswinspec.o
flashy.o flash.o flashyspec.o
rexx.o
#
(This is part of MultiPlayer's WITH file.) All the object files on
one
line are linked into the same overlay node: they are loaded and unloaded
together, and they may share data, make no-overhead calls to each other,
and so on. Each line specifies a separate overlay node. Note that you
should never use asterisks ('*') before the filenames, which normally
denote overlay sublevels, as Bovs neither needs nor supports hierarchial
overlays.
There are certain limitations to what you can do when programming
with
overlays. First, you can't directly reference data in one overlay node
from another overlay node, or from the root. The only exception to this
is if you name the overlay node's data hunks "__MERGED", in which case
BLink will merge the data into the root node's global data segment. In
effect, the data is no longer associated with the overlay node it was
defined in, but rather in the permanently resident root node. Compiling
with SAS/C using the small data model will cause this to happen. You
should avoid putting too much overlay-node-specific data into the
__MERGED
hunk, because that data will be loaded all the time the program runs,
even
when the actual overlay node isn't loaded. (To keep certain variables
out
of the __MERGED hunk under SAS/C while still using the small data model,
just define your overlay-node-private variables as __far.)
Bovs imposes very few restrictions on calling functions in overlay
nodes. Any time an overlaid function is called, BLink reroutes the call
through an entrypoint in Bovs, which loads the required overlay node if
it
isn't already loaded, locks it in memory, calls the function, and unlocks
it when the function returns. Theoretically, functions within overlay
nodes could call functions in other overlay nodes with no problem.
However, BLink has some very inconvenient error checking code that won't
let you link a program in which one overlay node calls another. (This is
based on the assumption that loading one overlay node causes all other
overlay nodes on the same level to be unloaded - which, with Bovs, is not
the case.) For now, unless you find a way to hack BLink to disable this
code (if you do, tell me how!), you'll have to create "stub" functions in
some non-overlaid module to bounce calls back into other overlay nodes.
Function calling conventions in Bovs are quite different from those
used with the standard overlay manager. In particular, no stack
parameters
are allowed when calling an overlaid function from outside the function's
overlay node (when BLink has to route the function through Bovs). This
is
because Bovs has to keep some information on the stack during the call so
it can unlock the overlay node on returning. (It can't just "load and
jump" like the standard overlay supervisor can.) The only parameters
that
are guaranteed to get to the called function are register parameters
passed
in A0 and A1. (In SAS/C, with register arguments active, this amounts to
two pointer-type parameters; if you want to pass value parameters, just
convert them them to void pointers.) In addition, the register A4 is
passed through to allow easy use of the small data model.
If you pass register parameters to overlay functions when using Bovs,
you'll get a bunch of annoying but meaningless warning messages from
BLink
telling you that the register parameters will get destroyed, because it
thinks that overlay functions are supposed to be called with stack
parameters. For now, just let the warnings roll up your screen; I'm
trying
to get SAS to put an option in BLink to disable this checking, but so far
I've had no luck.
Now we get into the really unconventional bit. When an overlaid Bovs
function is called, it actually receives one extra parameter that the
caller never passed to it. This parameter is passed in D0 (just define
it
as a 'long' in a SAS/C function prototype), and contains a "handle" on
the
overlay node in which the function resides. This handle is only valid
throughout that function call, unless the function locks the overlay node
into memory with LockOverlay(), described later, before returning. A
valid
handle passed to you will never be zero (unless you are using "Bst"
instead
of "Bovs" and thus have overlays disabled - this is described later), but
other than that you can't assume anything about what it actually is.
At this point, you should be able to implement overlaid programs
using
Bovs. Bovs also has some extra features to make your programs sleek and
efficient, which are described in the following sections.
Functions
~~~~~~~~~
Bovs defines a few extra functions which you may call during your
program for various purposes. Besides these, since MemMan must also be
linked into any program that uses Bovs, you may also call MemMan's memory
management functions. Do not call MMInit() or MMFinish() from your
program
- Bovs does that for you. However, if you use MemMan directly, you must
be
sure to remove all MMNodes before terminating your program. (See
MemMan.doc for details on using MemMan.)
void BExit(void)
void BRExit(long retcode,long retcode2)
D0
D1
Call one of these functions to terminate your program. BExit()
returns
with a returncode of zero; BRExit() allows you to set the return codes
before exiting. "retcode" should be RETURN_OK, RETURN_WARN,
RETURN_ERROR,
or RETURN_FAIL (defined in dos/dos.h). if you are returning an error,
"retcode2" should contain the secondary return code - one of the
ERROR_xxx
definitions in dos/dos.h. Note that these return codes will normally be
"seen" only if you call BRExit() from within your PreStart() routine. If
you call it from Main(), your program will have already been detached and
no one will be paying attention to whatever return code you terminate
with.
(Of course, if you have reassembled Bovs with the DETACH flag set to
zero,
then the return code will always be seen by the CLI.) Calling BExit() is
effectively the same as returning from your Main() function.
Of course, since Bovs does no resource tracking, you must be sure to
free all resources before calling BExit() or BRExit(). Also, they may
only
be called from the original process that Bovs gave you.
void LockOverlay(long ovhan)
D0
void UnlockOverlay(long ovhan)
D0
Given an overlay handle passed by Bovs to one of your overlaid
functions, LockOverlay() will lock the overlay node containing that
function in memory. It will not be unloaded until you unlock the overlay
node. LockOverlay() calls may be nested, and every call to LockOverlay()
must be paired with exactly one call to UnlockOverlay(). The only
exception to this rule is that you don't have to unlock every locked
overlay node before your program exits: Bovs will unload all loaded
overlay
nodes before exiting, locked or unlocked. In any case, you must never
try
to unlock an overlay node that you haven't already locked.
You may only call LockOverlay() on a given overlay handle from within
the function that was passed the handle, or from within a sub-function
that
it calls. You MAY NOT pass an overlay handle back from an overlay
function
to the function that calls it, unless you lock it first. In other words,
the overlay handle is only valid during the call to the overlaid
function,
unless you lock it within that function.
When using LockOverlay() and UnlockOverlay() to ensure that a certain
overlay node stays in memory at certain times, it is permissible for the
main module or other overlay nodes to access the overlay node's data
(indirectly through pointers, since direct references are not allowed),
or
to call functions in the overlay node through function pointers. This
can
be desirable, for example, if you have some function that is very
frequently called and you don't want the slight extra overhead of running
through the overlay supervisor every time the function is called.
However,
any time "outsiders" access a node's data or functions this way, you must
be sure that the overlay node is always locked at the time, so it can't
disappear on you.
You may call UnlockOverlay() from anywhere in your program, as long
as
you are sure the overlay handle you're passing to it is valid and locked,
and you're sure that the call will not pull out the carpet from under you
in some way. To put it in simpler terms, an overlay node may be unlocked
by (a) an "outsider" who has been given the overlay handle to unlock, or
(b) a function within the overlay node which was called normally through
the overlay supervisor (not directly through a function pointer). Since
functions called through function pointers are not automatically locked
by
Bovs throughout the call, if such a function were to unlock the last lock
on its own overlay node, it could be unloaded while the function is still
running. The only time when you may unlock the last lock on an overlay
node from within such a function is in assembly language: if the call to
UnlockOverlay() is done with a JMP instruction, and the return address
points to some other overlay node (or the root), the call to
UnlockOverlay() will return directly to whatever "outsider" called the
function in the first place, so there is no problem.
If this all sounds a little complicated at this point, don't panic:
you
only need to worry about this if you're using function pointers called by
external modules. If you stick to normal functions, there is not much
you
need to worry about.
long ResCall(void *routine,void *arg1,void *arg2)
D0
A2
A0
A1
This function is a little gimmick that you might find useful for
creating sleek, fast programs. It is used to call overlaid functions in
a
way slightly different than normal. If some part of the program wants to
call a function in another overlay node, normally it makes a call
normally,
and BLink reroutes the call to go through Bovs. However, instead of
calling the function directly, you can pass a pointer to it (which BLink
turns into a pointer to one of its dynamically created rerouter routines)
to ResCall, along with the parameters "arg1" and "arg2" which you would
normally pass to the real function in A0 and A1. ResCall then calls the
function normally, with the parameters that you specify.
If the overlay node containing the needed function is already loaded,
then the function is called normally, and its return value (whatever it
may
be) is passed back in D0, as usual. However, if the overlaid function is
NOT resident, the function is not called at all, and Bovs returns with
zero
in D0 immediately, without loading the overlay node.
Thus, ResCall provides a simple and convenient way to call an
overlaid
function only if that overlay node is already resident. I have found
this
useful in many cases, but I'll let you figure them out yourself. In any
case, if the overlay node is resident and the function is called, all
standard overlay function calling and locking conventions still apply.
Bst.o
~~~~~
The alternate object file "Bst.o" (pronounced "beast") can be used in
place of "Bovs.o" if you want Bovs's startup features without its overlay
system, or if you want to create a non-overlaid version of a program that
is normally overlaid. Your normally overlaid functions may end up be
called with anything (including zero) as the overlay handle parameter in
D0, since they are not called through the overlay supervisor at all.
While
the LockOverlay() and UnlockOverlay() functions still exist in Bst, they
do
absolutely nothing. Therefore, as long as you're somewhat careful, you
should be able to link your program in both overlaid and non-overlaid
versions, and have it perform exactly the same either way. MultiPlayer's
registered distribution, for example, contains both overlaid and
non-overlaid versions of the MultiPlayer program, to best suit the user's
preferences.
History
~~~~~~~
2.0 (10-Apr-92)
Removed the commercial license fee. (All I require now for
commercial use is a copy
of the commercial program when it is completed.)
Added PreStart call to allow program to avoid auto-detachment.
Added conditional assembly to allow Bovs to be assembled without
auto-detach capability.
Changed 'argtemplate' to be a _pointer_ to a string, not an
actual string.
Argument parsing on 2.0 now allows a client-defined extended help
string ('argexthelp').
Renamed 'procname' to 'progname'.
Fixed a bug in setting priority when starting from Workbench.
Added 'argsparsed' flag which is set if the arguments were parsed
with 2.0's ReadArgs().
Fixed a memory deallocation bug that could cause problems with
programs linked in a
certain (unusual) way.
1.0 (18-Feb-92)
First public release.
Contact Address
~~~~~~~~~~~~~~~
I tend to move around a great deal, so mail sent directly to me
sometimes has a hard time catching up. If you want mail to reach me (it
may take a while, but it WILL reach me), send it to this address:
Bryan Ford
8749 Alta Hills Circle
Sandy, UT 84093
I can be reached more quickly (for the time being anyway) on the
phone
or through one of the electronic mail addresses below:
(801) 585-4619
bryan.ford@m.cc.utah.edu
baf0863@cc.utah.edu
baf0863@utahcca.bitnet
If you want to get something to me through the mail more quickly,
FIRST
call or E-mail me to make sure I'm still here, then send it to this
address:
Bryan Ford
27104 Ballif Hall
University of Utah
Salt Lake City, UT 84112
Enjoy!