Hi,

I'm running into a strange problem with inline methods.

   TCircularBuffer=class
   ...
     Pin:pointer;
     Pout:pointer;
     function IsFull:boolean;inline;
   end;
function TCircularBuffer.IsFull: boolean;inline;
begin
   result:=PIn=POut;
end;

When the method is called from outside the class then it is inlined (x64 linux):

dbcacheservercore.pas:197 if FBuffer[currbuf].IsFull then 0x000000000048e4bf <+47>: mov -0x18(%rbp),%rax 0x000000000048e4c3 <+51>: mov 0x8(%rax),%rdx 0x000000000048e4c7 <+55>: movslq -0x24(%rbp),%rax 0x000000000048e4cb <+59>: mov (%rdx,%rax,8),%rax 0x000000000048e4cf <+63>: mov %rax,-0x30(%rbp) 0x000000000048e4d3 <+67>: mov -0x30(%rbp),%rax 0x000000000048e4d7 <+71>: mov -0x30(%rbp),%rdx 0x000000000048e4db <+75>: mov 0x28(%rax),%rax 0x000000000048e4df <+79>: cmp 0x30(%rdx),%rax 0x000000000048e4e3 <+83>: sete %al
0x000000000048e4e6 <+86>: test %al,%al
0x000000000048e4e8 <+88>: je 0x48e585 <ADD+245>

But when the method is called from inside the class then a call is used:

dbcacheservercore.pas:109 if IsFull then
0x000000000048dfc8 <+24>: mov -0x18(%rbp),%rdi 0x000000000048dfcc <+28>: callq 0x48e170 <ISFULL> 0x000000000048dfd1 <+33>: test %al,%al
0x000000000048dfd3 <+35>: jne 0x48e042 <ADD+146>

Using Self.IsFull results in the same code.

FPC 2.6.0 and 2.7.1 from today.

A bug?

Thanks, Ludo

Ludo Brands wrote

When the method is called from outside the class then it is inlined (x64 linux):

[snip]

But when the method is called from inside the class then a call is used:

[snip]

Are those calls before or after the method body appears in the source code? The compiler cannot inline anything for which it hasn't parsed the source code yet.

Jonas

Does that differ ? If so, then I should mention this in the documentation !

Michael.

Someone else also thought so in the past: http://lists.freepascal.org/lists/fpc-pascal/2009-November/023321.html :)

Jonas

The method was implemented below the code that used it. Moving it up solved it.

Actually the "inline;' in the declaration is not really used. Removing it results in the same code. Perhaps a Hint or Warning when inline is used in the declaration could avoid these mistakes.

Thanks a lot.

Ludo

Actually, the correct behaviour would be that the compiler requires that it is only specified in the interface, or both in the interface and in the implementation (I'm not sure yet what's most appropriate). Inlining is always also enabled across units and hence should have to be specified in the interface.

Jonas

Ludo Brands wrote

Perhaps a Hint or Warning when inline is used in the declaration could avoid these mistakes.

+1 for a hint in those cases.