Hi All,

I try to port a program from C to Pascal, see the following example:

#include <stdio.h>
#include <sys/socket.h>
#include <linux/if.h>
void main() {
    printf("ifmap=%ld\n", sizeof(struct ifmap));
}

I converted it to:

est.c test.pas

       X
program test;
{$mode objfpc}{$PACKRECORDS C}
type
  ifmap = record
    mem_start: dword;
    mem_end: dword;
    base_addr: word;
    irq: byte;
    dma: byte;
    port: byte;
  end;
begin
  WriteLn('ifmap=', SizeOf(ifmap));
end.

The C struct's size is 24, but pascal record is 16. The ifmap struct is defined as:

struct ifmap {
    unsigned long mem_start;
    unsigned long mem_end;
    unsigned short base_addr;
    unsigned char irq;
    unsigned char dma;
    unsigned char port;
    /* 3 bytes spare */
};

How can I instruct FPC to align records exactly same as GCC?

Thanks!

It seems that $packrecord does not did some research and found this document:

http://www.freepascal.org/docs-html/ref/refsu19.html

compiled and ran the example on that page, I got:

​​Size Trec1 : 4 Offset B : 2
Size Trec2 : 3 Offset B : 1
Size Trec3 : 2 Offset B : 1
Size Trec4 : 2 Offset B : 1
Size Trec5 : 5 Offset B : 1 Offset C : 4
Size Trec6 : 5 Offset B : 1 Offset C : 4
Size Trec7 : 9 Offset B : 1 Offset C : 8
Size Trec8 : 9 Offset B : 1 Offset C : 8

While the expected output, as stated in the document above, should be:

Size Trec1 : 4 Offset B : 2
Size Trec2 : 3 Offset B : 1
Size Trec3 : 2 Offset B : 1
Size Trec4 : 2 Offset B : 1
Size Trec5 : 8 Offset B : 4 Offset C : 7
Size Trec6 : 8 Offset B : 4 Offset C : 7
Size Trec7 : 12 Offset B : 4 Offset C : 11
Size Trec8 : 16 Offset B : 8 Offset C : 15

2015-08-07 10:48 GMT+08:00 Xiangrong Fang :

By using the correct types for the fields:
http://www.freepascal.org/docs-html/rtl/ctypes/index-3.html

E.g., "unsigned long" is not (always) the same as "dword".

Jonas

The documentation seems to be outdated regarding the alignment of arrays. Arrays probably used to be aligned based on the size of the entire array at one point (it's before recorded history, i.e. before September 2000), or maybe the record field insertion code used to contain some special (broken) logic for arrays that ignored their alignment and instead used their size.

In any case, the alignment requirement of an array is and has been for quite some time the alignment requirement of its individual elements, and the alignment of a field's type is what's used to determine its offset in a record (with a maximum of the current packrecords setting).

Jonas

conversion program. :-) I will use ctypes to test again.

However, another issue: if I use this :

​{$CODEALIGN RECORDMIN=4}

It worked. i.e. output size same as C version. I wonder what's the difference between $CODEALIGN and $A/$PACKRECORDS?

Also, if I use "packed records", alignment will never take place, regardless of $A settings, right?

Thanks!
​Xiangrong​

Xiangrong Fang wrote

​OK, as a matter of fact, the stuff was copied from output of h2pas conversion program. :-)

Then it seems that h2pas is pretty broken.

No, it didn't. The record definition was still wrong.

i.e. output size same as C version.

That's coincidence.

I wonder what's the

difference between $CODEALIGN and $A/$PACKRECORDS?

Packrecords sets the maximum field alignment. As the name implies,
{$codealign recordmin} sets the minimum field alignment.

Also, if I use "packed records", alignment will never take place, regardless of $A settings, right?

Indeed.

Jonas

"Xiangrong Fang" wrote

> > By using the correct types for the fields:

http://www.freepascal.org/docs-html/rtl/ctypes/index-3.html

conversion program. :-) I will use ctypes to test again.

If I remember correctly there's a parameter you to have h2pas use the ctypes unit and its types as well.

Regards,
Sven

Jonas Maebe wrote

... Then it seems that h2pas is pretty broken. ...

FWIW, I had better results in the past using a tool called ToPas http://sourceforge.net/projects/topas/

It will run under wine in linux. However, I can't get it to pass standard #include <foo> style directives, unless the headers are copied to the current folder and the removed.

Peter

Peter wrote

However, I can't get it to pass ....

Should have been "get it to parse" of course.

Xiangrong Fang wrote

It seems that $packrecord does not always explicitly use PACKED Record as in

/T//ype////ifmap = PACKED Record/
/ mem_start: dword;/
/ mem_end: dword;/
///b//ase_addr: word;/
/ irq: byte;/
/ dma: byte;/
/ port: byte;/
/ end;//
/
and at least in/up to FPC 2.6.4, this works for me just fine

Ralf

---
This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus

This depends however on the code you're interfacing with. Also the record you mentioned as an example shouldn't differ with and without packed anyway...

Regards,
Sven

El 08/08/15 a les 09:30, Sven Barth ha escrit:

With packed the above record is 13 bytes, without is 16 (the bytes are word aligned).
That's with fpc 2.6.4 x86_64 (though it should be the same with a 32 bits compiler).
Just to be sure, in such cases I also use "packed array" and most of the times bitpacked records.

Bye

I just copied and pasted the code snippet from the OP, to show how to use the PACKED Record statement explicitly for a specific record. I am using this for cross-platform/cross-compiler/cross-language software for at least 30 years now and probably for +20 years in FreePascal (when it was still called FPK and was an attempt of creating a 32bit compiler compatible with 16bit Turbo/Borland Pascal at that time)...

Beside, on the above record, the layout in fact SHOULD matter as it has 3 single bytes at the end and there for needs to be aligned on the byte, not on any word/dword/etc boundary...

Ralf

Ralf

---
This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus