I know very little about Linker --just vaguely what it does, and that's all. I googled for it, but could not find much either.

Could someone give me some pointers to read about the subject.

The reason why I was interested is, it seems a Linker is more difficult to write than a compiler --after all we do have FPC, but not a linker in Pascal.

Is it true that a linker is harder to write? If not, why hasn't anyone done one yet?

TVMIA

No.

If not, why hasn't anyone done one yet?

Other things were important so far?

Ever heard Google? The second result with a search for 'linker' has a lot of information.

A linker is much easier than a compiler. The hardest part is the analysing of bugs, because you need to debug the generated result and debug that at assembly level.

> If not, > why hasn't anyone done one yet? Other things were important so far?

There is already a start of a linker, but there was no time to finish it. We first wanted release a stable compiler.

Peter,

That was harsh. I did of course google. But, from
the links, I did not get much digestable info.

The second link for 'linker' on Google is this.

http://enproject.codelib.co.kr/

How is it relevant?....

You see, if you know little about a subject, there's little chance you can locate useful information...

That is why I asked here.

Pardon my ignorance, but why does it have to be debugged at assembler level?

Any links to it?

Thank you for your time.

Cheers,
Ray

Below is my google output. So you can see that the linkers and loaders is the second one.

Het Internet Resultaten 1 - 10 van circa 4.470.000 voor linker (0,33 seconden)

Gesponsorde Koppelingen
Linker
Alles over Linker.
Registreer op eBay.nl nu!
eBay.nl

Linker.ch - Das Schweizer Portal - direkt zum guten Link Linker.ch - Das Schweizer Portal mit den besten Links im deutschsprachigen Raum ( Linksammlung )
www.linker.ch/ - 2k - In cache - Gelijkwaardige pagina's

Linkers and Loaders
You don't need to tell me I'ma lousy artist. I already know that. Please send comments and questions to Thanks! ...
www.iecc.com/linker/ - 8k - 18 juli 2005 - In cache - Gelijkwaardige pagina's

add some const and base2power func
my change marks with
//==Okulov==start
...my changes...
//==Okulov==stop

listmember wrote

You see, if you know little about a subject, there's little chance you can locate useful information...

I wonder how we got our information about this subject :)

Okulov wrote

add some const

The constants are too unprecise (float is actually an extended).

and base2power func

Is this really useful? I mean when I see
code like
i:=base2power(j);
i:=1 shl j;
I understand quicker what is meant in the second case :)

Ok. You can change it to single

PI consists more than 200 items, but we use only 9 of them isn't it, and if you are using a cycle for example in OpenGL render (like me) it's cool speed up!

Ofcource you understand... but some people can use it.

Anyway if you don't wish to insert it, it is not necessary. My business is to offer =)

| > and base2power func
|

If you know how to write it, you can put this function in an external unit of your own.. doesn't have to be in the original unit, if others don't want it there. I use an IP address converter function when using sockets that isn't in the sockets.pp, which is just a wrapper around shl. Handy in this situation, because Copy and Pasting ip addresses is not easy when you always have to convert to shl.

In 2.0 afaik there are routines to convert IPs in string form to a value, in an endian correct way. htons and strtoaddr or so.

I found out that for operators which are declared to be inline and have asm code inside the inlining option is ignored (I assume the same would be true for procs/funcs).

Is that intentional? Does it have a reason? Would be great if that could be stopped.

Gerhard

Yes - needs complex parsing to determine if the asm is dangerous (e.g. EBP usage) - enhanced support not expect anytime soon due to complexity

Yes, inlining assembler is much harder than inlining Pascal code (and even harder to do it efficiently).

Jonas

I would suggest an "experimental" solution {$inline unsafe ?} which does the inline without checks, and maybe gives a warning at compiletime: Warning: inlined func/proc might be unsafe.

Could save some code and some exection time.

I do not understand the argument with e.g. EBP usage. I would say this is dangerous in not inlined code also, isn't it?

Gerhard

I would suggest an "experimental" solution {$inline unsafe ?} which does the inline without checks, and maybe gives a warning at compiletime: Warning: inlined func/proc might be unsafe.

The used registers still have to be known. For pascal code this is info that is gained during the compilation. For asm this is afaik not kept.

However if sb would work on this and submit a patch (scanning assembler blocks to build a register use), it would probably help.

Could save some code and some exection time. I do not understand the argument with e.g. EBP usage. I would say this is dangerous in not inlined code also, isn't it?

procedure x(a:integer);assembler; stdcall;

asm
movl 8(%ebp),%eax
end;

Will load parameter in eax. However when inlined it will fail.

Also a lot of assembler procedures (coming from delphi) expect the parameters in specific registers. The compiler then firsts needs to load the values in these registers like the register calling convention describes. The result is in a lot of cases not what was expected and it makes the compiler a lot more complex regarding inlining. So for now the inlining of assembler code is disabled to keep the code in the compiler simple. Also note that this behaviour is compatible with delphi 2005.

Peter