AROS Application Development Manual -- The Exec Library
Index
Avis
This document is not finished! It is highly likely that many parts are
out-of-date, contain incorrect information, or are simply missing
altogether. If you want to help rectify this, please contact us.
In exec/types.h the following short-cuts are typedef'd. They are used
often in AROS, so you should nearly always include exec/types.h.
- APTR
- A generic pointer for multiple purposes.
- STRPTR
- A pointer to a null-terminated string.
- UQUAD
- Unsigned 64-bit integer variable.
- QUAD
- Signed 64-bit integer variable.
- DOUBLE
- 64bit IEEE floating-point variable.
- ULONG
- Unsigned 32-bit integer variable (longword).
- LONG
- Signed 32-bit integer variable (longword).
- FLOAT
- 32 bit IEEE floating-point variable.
- UWORD
- Unsigned 16-bit integer variable (word).
- WORD
- Signed 16-bit integer variable (word).
- UBYTE
- Unsigned 8-bit integer variable (byte).
- BYTE
- Signed 8-bit integer variable (byte).
- BOOL
- Boolean variable, TRUE and FALSE are also defined in exec/types.h.
- VOID
- Void.
There is another important typedef, IPTR. It is really important in AROS,
as it the only way to declare a field that can contain both an integer and a
pointer.
Note
AmigaOS does not know this typedef. If you are porting a program
from AmigaOS to AROS, you have to search your source for occurrences
of ULONG that can also contain pointers, and change them into
IPTR. If you don't do this, your program will not work on systems
which have pointers with more than 32 bits (for example DEC Alphas
that have 64-bit pointers).
The so-called BPTR`s were always a problem in AmigaOS and this problem was
inherited by AROS. In binary-compatible AROS versions a `BPTR is in fact the
fourth of the real pointer. If, for example, a pointer points to address
$80000 the BPTR pointing to the same address would contain $20000.
On systems without binary-compatibility, a BPTR is equal to an APTR.
To convert between a normal pointer and a BPTR use the macros:
#include <dos/bptr.h>
APTR BADDR( BPTR bptr );
BPTR MKBADDR( APTR ptr );
There also exists something called BSTR which is a special kind of string.
We will not discuss this here, though, because it is used only very rarely.
When the development of the Amiga started, it was designed as a pure
module-based games-console. As such, it didn't need any means of file system
handling. The OS was created without a file system in mind. But Commodore,
who bought the Amiga, wanted a full-fletched home-computer instead of another
games-platform. So, a short time before the Amiga's initial presentation,
a file system was needed. Instead of wasting time in developing a custom one,
the file system of an operating system called TRIPOS was ported to the Amiga.
Unfortunately TRIPOS was written in BCPL, a programming language with a quite
eccentric pointer handling. This pointer handling was inherited by the
AmigaDOS and later by AROS (even though later versions of AmigaOS and also
AROS are written in C).
AROS implements a system of linked lists, so-called exec lists.
A linked-list consists of a number of nodes that link to each other.
Two types of nodes are defined in exec/nodes.h:
- struct MinNode
- is the basic node. You don't need to know about its structure, since
every possible action on them is handled by some library function.
- struct Node
extends the simple struct MinNode. It provides some additional fields:
- ln_Name
- Each Node contains a pointer to a string, describing that node.
- ln_Type
- A list of types is defined in exec/nodes.h.
- ln_Pri
- A priority, used for sorting the list.
Both structures can be embedded into other structures. For example,
struct Library (defined in exec/libraries.h) contains a struct Node at
the beginning. This way all libraries can be contained in a list. The field
ln_Name points to the name of the library, ln_Type is set to NT_LIBRARY
to show that this node is a library and ln_Pri reflects the importance of
a library.
Of course, we need node containers: lists. These are defined in
exec/lists.h.
Like nodes, we have two different kind of lists:
- struct MinList
- is the minimal list. You do not need to know about its members; look at
it as a black-box.
- struct List
- contains an additional field lh_Type, which corresponds to ln_Type
of struct Node.
MinList's take MinNode's as members, while List's use Node's; they
are not interchangeable. While it's technically possible to use`Node`'s in
MinList's, you loose all their advantages.
FIXME: Macros
exec.library and the link-library amiga.lib contain some functions for
manipulating exec lists. Before a list can be used, it must be
initialized. This can be done using this amiga.lib function:
#include <proto/alib.h>
void NewList( struct List *list );
Nodes can be added to lists with these exec.library functions:
#include <proto/exec.h>
void AddHead( struct List *list, struct Node *node );
void AddTail( struct List *list, struct Node *node );
void Enqueue( struct List *list, struct Node *node );
void Insert( struct List *list, struct Node *node, struct Node *pred );
With AddHead() and AddTail() node is inserted at the beginning or the
end of list respectively. Enqueue() inserts node according to its
ln_Pri field. A node can be inserted after another by using Insert().
A pointer to the node to insert node after, must be provided as pred.
Nodes can be removed using these exec.library functions:
#include <proto/exec.h>
void Remove( struct Node *node );
struct Node *RemHead( struct List *list );
struct Node *RemTail( struct List *list );
While RemHead() and RemTail() remove the first or last node of a list
respectively and return a pointer to it, Remove() removes node from
whatever list it is in.
Of course, apart from Enqueue(), all list functions can process struct
MinList and struct MinNode's, as well.
A list can be searched for a named node, using:
#include <proto/exec.h>
struct Node *FindName( struct List *list, STRPTR name );
name is a pointer to a string that is to be compared with the ln_Name
of the nodes in list. The comparison is case-sensitive! If name
matches any ln_Name field, a pointer to the corresponding node is
returned. If no field matches, NULL is returned.
Note
A list used with FindName() must not contain any struct MinList
entries. If it does, memory could get corrupted!
In the following example, we create a list, add three nodes to it, search
a named node and then remove it:
#include <proto/alib.h>
#include <proto/exec.h>
#include <exec/types.h>
#include <exec/lists.h>
#include <exec/nodes.h>
#include <dos/dos.h> /* For RETURN_OK */
struct List list;
/* Our nodes */
struct Node node1 =
{
NULL, NULL, /* No predecessor and successor, yet */
NT_UNKNOWN, 0, /* Unknown type, priority ignored */
"First node" /* Name of the node */
};
struct Node node2 =
{
NULL, NULL,
NT_UNKNOWN, 0,
"Second node"
};
struct Node node3 =
{
NULL, NULL,
NT_UNKNOWN, 0,
"Third node"
};
int main(int argc, char *argv[])
{
struct Node *node;
/* Prepare the list for use. */
NewList(&list);
/* Add the first two nodes at the end of the list. */
AddTail(&list, &node1);
AddTail(&list, &node2);
/* Insert the third node after the first node. */
Insert(&list, &node3, &node1);
/* Find the second node */
node = FindName(&list, "Second node");
/*
If the node was found (which is always the case in this example),
remove it.
*/
if (node)
Remove(&node);
return RETURN_OK;
}
AROS defines a couple of macros in various header files. All macros
cast their parameters to the correct type, so you must provide a
valid input but can safe the casts (macros are meant to make life
simpler).
- NEWLIST(list)
Compatible: | Yes |
Location: | exec/lists.h |
Initializes a list. You should not use any list before you have
initialized it.
- GetHead(list)
Compatible: | Yes |
Location: | exec/lists.h |
Returns a pointer to the first node of a list, or NULL if the list
is empty.
- GetTail(list)
Compatible: | Yes |
Location: | exec/lists.h |
Returns a pointer to the last node of a list, or NULL if the list
is empty.
- GetSucc(node)
Compatible: | Yes |
Location: | exec/lists.h |
Returns a pointer to the next node of a list, or NULL if there is
none.
- GetPred(list)
Compatible: | Yes |
Location: | exec/lists.h |
Returns a pointer to the previous node of a list, or NULL if there is
none.
- ForeachNode(list,node)
Compatible: | Yes |
Location: | exec/lists.h |
Iterates through a list. A block of code must follow this macro. The
block doesn't get executed if the list is empty. When the list terminates
node doesn't contain NULL but node->ln_Succ will be NULL.
This macro can't be used if you want to delete the nodes in the list
(i.e.. you must not call Remove() inside the block of code following the
macro). Use ForeachNodeSafe() if you have to delete nodes.
Example:
/* Iterate through a list with complete nodes and print their names */
t = 1;
ForeachNode(list,node)
{
if (node->ln_Name)
{
printf ("Node %d: %s\n", t++, node->ln_Name);
if (!strcmp (node->ln_Name, "end"))
break;
}
}
if (node->ln_Succ)
printf ("Not all nodes have been processed\n");
else
printf ("The list doesn't contain a node with the name \"end\"\n");
- ForeachNodeSafe(list,node,tmpNode)
Compatible: | Yes |
Location: | exec/lists.h |
Iterates through a list. A block of code must follow this macro. The
block doesn't get executed if the list is empty. When the list terminates
node doesn't contain NULL but node->ln_Succ will be NULL.
This macro can be used with code that deletes nodes in the list.
- SetNodeName(node,name)
Compatible: | Yes |
Location: | exec/lists.h |
Sets a new name for a node. The name is not copied, the macro will just
make ln_Name point to name.
The macro casts node to struct Node *
so you better make sure that node is a full featured node.
- GetNodeName(node)
Compatible: | Yes |
Location: | exec/lists.h |
Return the current name of a node.
The macro casts node to struct Node *
so you better make sure that node is a full featured node.
- ListLength(list,count)
Compatible: | Yes |
Location: | exec/lists.h |
This puts the number of nodes in list into count.
You need memory for nearly everything in a program. Many things can be done
using the stack, but often you need larger chunks of memory or don't want
to use the stack for some reason. In these cases you have to allocate memory
by yourself. The exec.library provides several methods for allocating
memory. The two most important functions are:
#include <proto/exec.h>
APTR AllocMem( ULONG size, ULONG flags );
APTR AllocVec( ULONG size, ULONG flags );
Both functions return a pointer to a memory area of the requested size
provided as argument. If not enough memory was available, NULL is
returned, instead. You must check for this condition, before using the
memory. If the memory was successfully allocated, you can do with it
whatever you want to.
You can provide additional flags to get a special kind of memory. The
following flags are defined in exec/memory.h:
- MEMF_CLEAR
- The allocated memory area is initialized with zeros.
- MEMF_LOCAL
- Get memory that will not be flushed, if the computer is reset.
- MEMF_CHIP
- Get memory that is accessible by graphics and sound chips. Some
functions require this type of memory.
- MEMF_FAST
- Get memory that is not accessible by graphics and sound chips. You
should normally not set this flag! It is needed only for some very
esoteric functions. Many systems don't have this kind of memory.
- MEMF_PUBLIC
- This flag must be set, if the memory you allocate is to be accessible by
other tasks. If you do not set it, the allocated memory is private to
your task. This issue will be discussed in detail in the chapter about
.. FIXME:: inter-task communication.
- MEMF_REVERSE
- If this flag is set, the order of the search for empty memory blocks is
reversed. Blocks that are at the end of the list of empty memory will be
found first.
- MEMF_NO_EXPUNGE
- Normally, if not enough free memory of the requested size is found, AROS
tries to free unused memory, for example by flushing unused libraries
out of the memory. If this flag is set, this behaviour is turned off.
Memory allocated with these functions must be freed after use with one of
the following functions. Note well that you should never access memory after
it has been freed.:
#include <proto/exec.h>
void FreeMem( APTR memory, ULONG size );
void FreeVec( APTR memory );
Of course, FreeMem() must be used for memory allocated with AllocMem()
and FreeVec() for memory allocated with AllocVec(). The synopsis for
these two functions shows the difference between AllocMem() and
AllocVec(): AllocVec() remembers the size of the chunk of memory it
allocated. So, if you use AllocVec(), you don't have to store the
requested size, while you have to if you use AllocMem().
Sometimes you may want to make multiple memory allocations at once. The usual
way to do this is calling AllocVec() with the size of all memory-blocks
added together and then making pointers relative to the returned pointer.
But what do you do, if you need memory of different kinds, with different
MEMF_ flags set?
You could make multiple allocations or simply use this function:
#include <proto/exec.h>
struct MemList *AllocEntry( struct MemList *oldlist );
As you will have noticed, AllocEntry() uses a pointer to a
struct MemList as only argument and as result. We find the definition of
this structure in exec/memory.h:
struct MemEntry
{
union
{
ULONG meu_Reqs;
APTR meu_Addr;
} me_Un;
ULONG me_Length;
};
struct MemList
{
struct Node ml_Node;
UWORD ml_NumEntries;
struct MemEntry ml_ME[1];
};
The array ml_ME of struct MemList has a variable number of elements.
The number of its elements is set in ml_NumEntries. The struct
MemEntry describes one memory-entry. Stored are its size
(me_Length), its requirements (i.e. the MEMF_, set in
me_Un.meu_Reqs) and possibly a pointer to the memory-block
(me_Un.meu_Addr). The struct MemList, you pass in as oldlist,
must have set the field ml_NumEntries to the actual number of struct
MemEntry's contained in ml_ME. The struct MemEntry's
must have set the fields me_Length and me_Un.meu_Reqs. The other
fields are ignored. The function returns a pointer to a copy of the struct
MemEntry, passed in as oldlist, with all the relevant fields set
(especially me_Un.meu_Addr). An error is indicated by setting the most
significant bit of the pointer returned. So you always have to check it,
before using the pointer returned. Memory allocated with AllocEntry() too
must be freed using FreeMem().
AROS manages several so-called memory-pools. Each memory-pool contains a list
of memory-areas. Of these, the most important memory-pool is the pool that
contains all free memory in the system. But you also can create memory-pools
yourself. This has some advantages:
- Every time you allocate some memory, the memory in the system becomes more
fragmented. This fragmentation causes the available memory chunks to become
smaller. This way larger allocations may fail. To prevent this problem,
memory-pools were introduced. Instead of allocating many small chunks of
memory, the pool-management routines allocate large chunks and then return
small chunks out of it, when memory-requests are made.
- Private memory-pools have the ability to keep track of all the allocations
you made so that all memory in a pool can be freed with one simple
function-call (but you can also free memory individually).
Before a memory-pool can be used, it must be created. This is done with the
function:
#include <proto/exec.h>
APTR CreatePool( ULONG flags, ULONG puddleSize, ULONG threshSize );
The flags specifies the type of memory you want to get from the
AllocPooled() function . All MEMF_ definitions as described above are
allowed here.
The puddleSize is the size of the chunks of memory that are allocated by
the pool functions. Usually a size about ten times bigger than the average
memory-size, you need to allocate, is a good guess. But on the other hand
the puddleSize should not be too large. Normally you should limit it to
about 50kb. Note well, though, that these are only suggestions and no
real limitations.
Finally, the threshSize specifies how large the requested chunk of memory
must be to be allocated automatically, rather than after checking the pool.
If, for example, the threshSize is set to 25kb and you want to allocate a
piece of memory with the size of 30kb, the internal lists of chunks of that
memory-pool is not searched at all. Instead, the memory is allocated
directly. If the memory to be allocated was only 20kb, first the chunk-list
would have been searched for a piece of free memory of that size. Of course,
the threshSize shouldn't be larger than the puddleSize and it should not
be too small, either. Half the puddleSize is a good guess here.
CreatePool() returns a private pointer to a pool-structure that must be
saved for further use. NULL is returned if no memory for the
pool-structure was available. You have to check for this condition.
After use, all memory-pools must be destroyed by calling:
#include <proto/exec.h>
void DeletePool( APTR pool );
This function deletes the pool passed in. Additionally all memory that was
allocated in this pool is freed. This way, you don't need to remember every
single piece of memory, you allocated in a pool. Just call DeletePool() at
the end. Note that you should be careful not to access pooled memory after
its pool was deleted!
If you want to allocate memory from a pool, you need to call:
#include <proto/exec.h>
void *AllocPooled( APTR pool, ULONG size );
Besides the pool to allocate memory from, the size of the memory to be
allocated must be passed in. Returned is a pointer to a block of memory of
the requested size or NULL to indicate that not enough memory was
available.
Memory allocated with AllocPooled() can be freed by either destroying the
whole pool with DeletePool() or individually by calling:
#include <proto/exec.h>
void FreePooled( APTR pool, void *memory, ULONG size );
This function frees exactly one piece of memory that was previously allocated
with AllocPooled(). The pointer to the memory pointer, returned by
AllocPooled(), its size and the pool it is in, have to be supplied as
arguments.
Note
You may ask yourself: "If DeletePool() deletes all the memory of a
pool, why should I ever use FreePooled()?" The answer is easy: to save
memory. Normally it's good style to free memory as soon as you don't
need it any more. But sometimes it is easier just to free a memory-pool
after a bunch of allocations.
Nevertheless, you should not use this feature if you are not sure when
the memory-pool will be deleted. Imagine a program like this (do not try
to compile it; it won't):
#define <exec/types.h>
#define <exec/memory.h>
#define <dos/dos.h>
int main(int argc, char *argv[])
{
APTR pool;
APTR mem;
/* Create our memory pool and test, if it was successful. */
pool = CreatePool(MEMF_ANY, 50*1024, 25*1024);
if (pool)
{
/* Just a dummy function. Image that this function will open
a window, with two buttons "Do Action" and "Quit".
*/
open_our_window();
for(;;)
{
/* Another dummy function that returns one of the
definitions below.
*/
switch(get_action())
{
/* This is returned, if the button "Do Action" was released. */
case DOACTION:
mem = AllocPooled(pool, 10*1024);
if (mem)
{
/* Another dummy function that uses our memory. */
silly_function(mem);
}
break;
/* This is returned, if the button "Quit" was released. */
case QUIT:
return RETURN_OK;
}
}
/* Close the window, we have opened above. */
close_our_window();
/* Delete our pool. */
DeletePool(pool);
}
}
Each time the button Do Action is released, some memory is allocated.
This memory is freed at the end of the program, when DeletePool() is
called. Of course, the longer the program is used, the more memory will
be in use. That is why it would be much better to free the memory after
use. This is done by replacing the part between case DOACTION: and
case QUIT: by:
mem = AllocPooled(pool, 10*1024);
if (mem)
{
silly_function(mem);
FreePooled(pool, mem, 10*1024);
}
break;
Memory-pools are managed with struct MemHeader's. If you have a pointer
to such a structure, you may try to allocate some memory from its pool:
#include <proto/exec.h>
void *Allocate( struct MemHeader *mh, ULONG size );
Apart from the pointer to the struct MemHeader passed in as mh, you
have to supply the size of the memory-block you want to allocate. This
function returns either a pointer to the first memory-block found or
NULL if no matching block was found.
You must free every memory-block allocated with Allocate() with:
#include <proto/exec.h>
void Deallocate( struct MemHeader *mh, APTR mem, ULONG size );
You have to pass the same mh and size to Deallocate() that you
passed to Allocate() and additionally the pointer it returned.
intuition.library provides another way to handle memory pools with the
functions AllocRemember() and FreeRemember(). Note, though, that these
are obsolete. You should use the normal pool-functions of exec.library,
instead.
Under very rare circumstances you may need to allocate memory at a specific
memory address. This performed by using:
#include <proto/exec.h>
void *AllocAbs( ULONG size, APTR address );
This function tries to allocate size bytes at address. If this is
successful, a pointer to the requested address is returned. If some memory of
the requested block is already allocated or is not available in the system,
NULL is returned, instead.
Avis
You should not write to the beginning of the memory-block!
The beginning of the returned memory block will have been used by exec
to store its node-data (the exact size is calculated by
(2*sizeof (void *)) ). Therefore, this area will not be available to
you. Either don't write there, or if there's memory before the address
you need, request a slightly larger block, starting far enough before
the intended start-address to make room for exec's data.
Because of these obstacles, you should not use AllocAbs(), except in
case you really need it.
Memory allocated with AllocAbs() must also be freed using FreeMem().
To get the size of available memory, use the function:
#include <proto/exec.h>
ULONG AvailMem( ULONG type );
The type parameter consists of some of the following flags (or-ed),
as defined in exec/memory.h:
- MEMF_ANY
- Return the size of all free memory in the system.
- MEMF_CHIP
- Return the size of memory, which is accessible by graphics and sound
chips.
- MEMF_FAST
- Return the size of memory that is not accessible by graphics and sound
chips.
- MEMF_LARGEST
- Return only the largest block, instead of all memory of the type
specified.
You may also specify other MEMF_ flags, however, they will simply be
ignored.
Note
Note well that the returned memory-size need not reflect the real size
of the memory available, as in a multitasking system this may change at
any moment, even while AvailMem() is being executed.
A program to list memory available in the system:
#include <stdio.h>
#include <exec/memory.h>
int main(int argc, char *argv[])
{
printf("Total free memory: %h, largest block: %h\n",
AvailMem(MEMF_ANY), AvailMem(MEMF_ANY|MEMF_LARGEST));
printf("Free chip memory: %h, largest block: %h\n",
AvailMem(MEMF_CHIP), AvailMem(MEMF_CHIP|MEMF_LARGEST));
printf("Free fast memory: %h, largest block: %h\n",
AvailMem(MEMF_FAST), AvailMem(MEMF_FAST|MEMF_LARGEST));
}
This chapter is only of concern to you, if you want to write a
hardware-driver for a piece of hardware which adds memory to the system.
This exec function adds memory to the list of free memory in the system:
#include <proto/exec.h>
void AddMemList
(
ULONG size, ULONG type, LONG priority,
APTR address, STRPTR name
);
You have supply the address and the size of the memory to add. The
type has to be set to at least one of the MEMF_ flags, which are
defined in exec/memory.h:
- MEMF_FAST
- Your memory must not be accessed by graphics or sound chips.
- MEMF_CHIP
- Your memory is reachable for graphics and sound chips.
You can provide a priority with which your memory will be added to the
memory list. The general rule is: The quicker your memory, the higher the
priority should be. If you don't know, what to supply here, supply 0.
Finally, you can provide a name, with which your memory can be identified
by the system and its users. You may provide NULL instead of a name, but
giving your memory a name is recommended.
Once your memory is added to the list of free memory,
it can't be removed again.
FIXME: AddMemHandler()/RemMemHandler()
An AROS-specific feature for tasks is that each Task gets an array of IPTRs
that can be used for Task-specific data. It can, for example, be used by
shared libraries to easily associate data with a task or by compilers to
make a program pure.
Slots can be allocated with AllocTaskStorageSlot() and freed with
FreeTaskStorageSlot(). After a slot has been allocated the data associated
with the slot can be manipulated by accessing
tc_UnionETask.tc_TaskStorage[slot] from a struct Task.
Generally, when a new task is run, the contents of all the slots of the
parent Task are copied into the new task. This holds for AddTask(),
NewAddTask() and CreateNewProc(). When RunCommand() is used, however, the
parent TaskStorage is reused in the child; changes made in the child to
TaskStorage will be visible in parent after the child has exited.
If these default behaviour is not acceptable, user code has to implement
checks and override it.
Functions:
|
|