Hello!
I have uVision that compiles fine with the C51 v7.03 compiler and the related package, but not complete with the 8.06. I used two different Keil installations. All files are in the same folder.
In the 8.06 I get linker errors like "object does not fit in to pdata page" and "0080H". This looks like the compiler was thinking the PDATA was only 128 bytes, but it is set to 256 bytes in the startup.a51. Any clue what's different in the newer Keil package?
Also there is a warning in 8.06 (which does not show in 7.03) "converting non-pointer to pointer" on this
ptr_xdata = sPtr_obj->Adresse;
while the vars are set like this:
uchar uc_set_obj( uchar pdata *ptr_Set) { uchar i; uchar xdata *ptr_xdata; struct stOBJADR code *sPtr_obj; sPtr_obj=&Obj[*ptr_Set]; . . . ptr_xdata = sPtr_obj->Adresse; }
The struct stOBJADR has a member "uint Adresse;"
I can see no wrong use of the pointers. I just want to be sure that the warning does not affect the code to not work correctly.
Well, you say it here: stObj->Adresse is a uint, and ptr_xdata is a pointer. The compiler will do the assignment, but issue a warning due to potentially incompatible data types. If you want the warning to go away, you need to cast the uint explicitly to a pointer.
Hmm, that'll do. But, incompatible data types? A pointer's address is always uint, so the compiler should not complain.
Thanks!
A pointer's address is always uint, so the compiler should not complain.
Maybe this is true for (today's) PC compilers, but it's definitely not true for the C51 compiler.
The C51 manual has a detailes explanation on how pointer work:
http://www.keil.com/support/man/docs/c51/c51_le_ptrs.htm
http://www.keil.com/support/man/docs/c51/c51_le_genptrs.htm
www.keil.com/.../c51_le_memspecificptrs.htm
Is it?!
A Generic Pointer in C51 is 3 bytes - so not a uint at all!
Anyhow, you're setting the pointer's value not its address!
Given that you haven't shown your definition of "uint", I s'pose you could have defined it such that this is true...
A pointers address?
The address of a pointer is another pointer.
What you where thinking about is the value of a pointer, not the address of a pointer.
You may decide that you would like to see the value of a pointer as an integer - after all, the memory can bee seen as a large data array with an integer index.
However, to the language, an integer and a pointer are two separate data types, just as char and int are two separate data types.
Almost all compilers will complain in almost all situations if you treat a pointers as an integer or an integer as a pointer.
By adding a typecast, you can inform the compiler that you are the master, and that you specifically want the integer to be known as a pointer value. You are allowed to do it, but it is up to you to make sure that integer value represents a valid pointer.
But once more: The value of a pointer may be an integer value, but it is not correct to assume a pointer value and an int (or unsigned int, long, ...) to be equivalent. Because they are not equivalent, the compiler will warn you.
It is only when programming in assembler, where you can play freely with addresses, since an assember doesn't have strict type checking.
Although the 8051 specifically has three distinct arrays (four, if you include the SFRs...)
Guys, please don't be so picky. A pointer's address for me is the address it points to. And this address is always uint, means two bytes. So if you're setting a pointer address, you should always use an uint value, while a char value would also do. That's all I meant. Not how many bytes a pointer actually consumes when compiled to machine code.
"Guys, please don't be so picky."
This is not picky - you have some very basic misunderstandings here! It's just as well that the compiler did warn you - otherwise you're heading for some really big traps!
"A pointer's address for me is the address it points to."
That is simply incorrect!
A pointer is a "variable" like any other; as such, it has a value, and it has an address - and they are in no way related!
"And this address (sic) is always uint, means two bytes."
That is not true in general, and not true for the 8051 specifically:
A pointer to DATA, IDATA or PDATA space is just a single byte; A pointer to CODE or XDATA is, indeed, 2 bytes; A Generic pointer - able to address any of the above - is three bytes
"So if you're setting a pointer address (sic), you should always use an uint"
Only in cases where the size of a uint (whatever that means) matches the size of the pointer.
"while a char value would also do"
Not necessarily: in general, you can only say that a char value might also do.
"Not how many bytes a pointer actually consumes when compiled to machine code."
But that is the value that actually matters - because that is the value that the code will actually be using!
Exactly. There is no guarantee of a one-to-one mapping of all aspects of a pointer when storing it's value in an integer.
A pointer may also have explicit rules about what values are allowed. For example a huge pointer on 16-bit x86 requires that the value is normalized, i.e. that the offset part is in the range 0 to 15, while the segment part can take the full range 0 to 65535.
The address of a pointer (a pointer's address) is a pointer to a pointer, and the concept is called dual indirection.
You need to use dual indirection when you program in C and want a C function to take a pointer as parameter and be able to modify the parameter on return.
It is also a common concept on hardware platforms that does not have a Memory Management Unit (MMU) and where you need to move allocated memory blocks around without having to know what parts of the code that have a reference to the memory blocks.
But no, it isn't picky to very clearly separate the value and the address of a pointer. There are a large number of programming situations where you must keep track of the number of indirections - and the compiler do keep track of the data type of pointers, i.e. a pointer to an int or a pointer to a pointer to an int are two very different data types, resulting in a warning if you mismatch your assign. The indirection of a pointer to a pointer to int is a pointer to int. The indirection of a pointer to int is the int, i.e. no longer an address.
You are all right, but I think this code is still not generally wrong. It wasn't written by me and it always worked. In order to distinguish the addresses a pointer is related to, I call the address the pointer points to the pointer address. The address where the pointer itself is located or managed, I call base address. But this address is commonly not used in consideration. When I'm writing C, then it's painful enough to fulfill the microprocessor's picky demands, so I don't want to care about if the pointer's base address consists of two or three bytes. Only if I wouldn't have enough RAM.
Back to my problem: Putting a cast in front of the particular line did not succeed. It still reports warning C289, which is not described in the C51 help file.
"I call the address the pointer points to the pointer address."
In your own little world, you can call it what you like. You just have to understand that it is completely contrary to what the rest of the world understands by that term. It also suggests that you have not fully grasped what's going on.
"The address where the pointer itself is located or managed, I call base address"
That is also contrary to the generally-accepted meaning of that term. A "base" address is generally accepted to mean the datum (or reference) from which an offset or relative address is calculated.
" But this address is commonly not used in consideration"
Oh yes it is!
As Per said, having a pointer to a pointer is quite common; in fact, the C51 compiler supports up to 19 levels of indirection! http://www.keil.com/support/man/docs/c51/c51_xd.htm
It is also common to need to know the address (ie, location) of a pointer when debugging...
"When I'm writing C, then it's painful enough to fulfill the microprocessor's picky demands"
Yes - you can get away with sloppy terminology when talking to other people, but not when talking to a compiler!
If you think 'C' is picky, you should try writing in Pascal or Ada...
Assembler may seem less "picky", but that just puts even more responsibility on the programmer to ensure that they have everything right!
"I don't want to care about if the pointer's base address (sic) consists of two or three bytes"
In general, you don't need to - it's only when you start trying to assign a non-pointer value (such as a uint) to a pointer that you need to be really sure of what you are doing...
"Only if I wouldn't have enough RAM."
No, that has nothing to do with it! If you mess up a pointer, it can point to absolutely anywhere in memory, and do untold damage.
That's why pointers in 'C' are both so powerful and so dangerous!
Which is why it is so important that you have a really firm understanding of what's going on.
Which is why compilers tend to give warnings when there's any doubt...
"Back to my problem: Putting a cast in front of the particular line did not succeed"
Show exactly what cast you used!
But you do have to learn to start thinking and talking about a pointers value, not a pointers address.
You will get into trouble if you do not see it as a variable. And variables have values. Sometimes they also have memory addresses.
A base address is something completely different, so please do not continue thinking about the address of a pointer as a base address. The start address of an array could be seen as a base pointer but a pointer and an array are not identical.
char buf[1000]; char* p = buf;
Now you think that the address of p is a base address. Why? The address of p has nothing to do with buf, and isn't the base of anything meaningful.
The reason it is painful for you to "fulfill the microprocessor's picky demans" is that you have not spent any time on figuring out why. If you know why something is as it is, you will start to see a pattern. And when you see the pattern, it will not feel like any picky demands anymore. People have spent huge amounts of time thinking about logical (not picky) solutions to the language design, the processor instruction design, the processor peripherials etc.
Everything you choose to treat as black magic will continue to be black magic. And black magic is something we humans are not comfortable with. It is up to you to decide if you want to take control, or if you cant to continue to be confused. As long as you decide to invent your own terminology, you will never be able to buy books and read. You will just be confused, since the books will - in your mind - be picky and invent the wrong terminology...
You say you don't want to have to care about if a pointer's base address (sic!) is two or three bytes. You don't have to know the size of a pointer variable, as long as you stop playing with addresses in integers. For small chips it may still be good to know the size of a pointer variable, since you have limited memory in the chip. But that is a completely different issue. Right now, your problem is that you are writing code where you need to make a lot of assumptions. This specifically forces you to have the knowledge to be able to figure out if the assumptions are correct or not.
You still have a warning?
Maybe you should post the full set of data declarations, and your rewritten source line - including the type cast. Only then will people on this forum be able to tell if your cast is correct or not.
LOL.
You won Andy. But there are a lot of similarities in our answers ;)
So if you're setting a pointer address, you should always use an uint value
Not true. intptr_t is the integer type wide enough to hold a pointer; ptrdiff_t is the type wide enough to hold a difference of two pointers. Neither one is necessarily the same as an unsigned int.
Compilers are allowed to warn about whatever they feel like. Given the complexity of the 8051 addresses spaces, as mentioned above, along with the variable width of pointers in the C51 implementation, I think it's a good thing for C51 to be sensitive in this regard. You're always free to disable or ignore the warning if you think it's silly, but in this case I'd recommended taking it to heart and putting in the proper casts where needed to convert the data type explicitly, rather than relying on default, implicit, conversions -- even if those did work in version 7.
char buf[1000]; char* p = buf; Now you think that the address of p is a base address. Why? The address of p has nothing to do with buf, and isn't the base of anything meaningful. This is what I consider as pointer address.
But we better stop here. I'm programming since a few years now and never had such problems. I think I don't need to read a bible about pointers, just to write some code. That's not the meaning of a computer language. To be more complicated than its inventor.
My conclusion is: if I set a pointer address in XDATA with a number I read from a struct and as long as this number is 2 bytes long and thus valid, then the pointer has to be set. Period. By assigning a pointer's address I'm not doing a pointer conversion, because the pointer only changes the byte it points to. Nothing more, nothing less.