Hi.

I discovered some Delphi incompatibility. Call AllocMem(0) do not return nil. Fpc from svn.

In SysGetMem are some comments about it:

  if size=0 then
    { we always need to allocate something, using heapend is not possible,
      because heappend can be changed by growheap (PFV) }
    size := 1;

Is this behaviour essential, or is possible make it compatible with Delphi?

Petr

Current behaviour is compatible with TP7 that gave a valid pointer back pointing to heapend. See also the comment.

IMHO doing an AllocMem(0) is an error in your program.

Peter

Peter Vreman wrote

Current behaviour is compatible with TP7 that gave a valid pointer back pointing to heapend. See also the comment.

I thought so too, but I tested and TP7 stores nil in p when doing getmem(p,0) (under DosBox).

Jonas

But AllocMem(0)?

Vincent

AllocMem does not exist in TP7. And the code posted by the original poster comes from SysGetMem, not from AllocMem (which is just a wrapper for getmem+fillchar).

Jonas

Yes. I tested GetMem and AllocMem in TP7.0, Delphi1, Delphi7, Kylix and all compillers returns nil when size is zero.

This patch solves it.

Index: inc/heap.inc
=================================================================== --- inc/heap.inc (revision 10368)
+++ inc/heap.inc (working copy)
@@ -991,10 +991,11 @@
function SysGetMem(size : ptruint):pointer;
begin
{ Something to allocate ? }
- if size=0 then
- { we always need to allocate something, using heapend is not possible,
- because heappend can be changed by growheap (PFV) }
- size := 1;
+ if size<=0 then
+ begin
+ result := nil;
+ exit;
+ end;
{ calc to multiple of 16 after adding the needed bytes for memchunk header }

   if size <= (maxblocksize - sizeof(tmemchunk_fixed_hdr)) then
     begin

But heaptrc will show unfreeed blocks with zero size. And this helps.

Index: inc/heaptrc.pp
===================================================================
--- inc/heaptrc.pp      (revision 10368)
+++ inc/heaptrc.pp      (working copy)
@@ -406,6 +406,11 @@
   pp : pheap_mem_info;
   loc_info: pheap_info;
 begin
+  if size<=0 then
+    begin
+      TraceGetMem:=nil;
+      exit;
+    end;
   loc_info := @heap_info;
   try_finish_heap_free_todo_list(loc_info);
   inc(loc_info^.getmem_size,size);

Are these patches acceptable?

Petr

Well, at the same time it breaks backwards compatibility with FPC. I really don't remember anymore why we did things this way, but I'm pretty sure there was a reason for it. Does anyone else remember?

Jonas