Sergei Gorelkin wrote:
Attached is a one-line fix that makes programs a couple of bytes smaller
by preventing creation of empty threadvar tables when threadvars are not
actually used.
The condition "Assigned(TAsmList.First)" always evaluates to True
because the list contains at least one marker item.
Whoops, not that easy. It appears that
Tresourcestrings.CreateResourceStringData
does not create new section in current_asmdata.asmlists[al_const] and
therefore it
depends on the state in which that asmlist was left by the previously
executed code. Hence, compiling a unit that does not have threadvars but
has resourcestrings ends up in an IE.
Attached a corrected patch. The section prefix 'RESSTRTABLE' is my wild
guess, so it may need to be changed.
Regards,
Sergei
Index: cresstr.pas
===================================================================
--- cresstr.pas (revision 13026)
+++ cresstr.pas (working copy)
@@ -162,6 +162,7 @@
makes the linking too dependent on the linker script requiring a SORT(*) for
the data sections }
maybe_new_object_file(current_asmdata.asmlists[al_const]);
+ new_section(current_asmdata.asmlists[al_const],sec_data,make_mangledname('RESSTRTABLE',current_module.localsymtable,''),sizeof(pint));
maybe_new_object_file(current_asmdata.asmlists[al_resourcestrings]);
new_section(current_asmdata.asmlists[al_resourcestrings],sec_data,make_mangledname('RESSTR',current_module.localsymtable,'1_START'),sizeof(pint));
current_asmdata.AsmLists[al_resourcestrings].concat(tai_symbol.createname_global(
Index: pmodules.pas
===================================================================
--- pmodules.pas (revision 13026)
+++ pmodules.pas (working copy)
@@ -219,7 +219,7 @@
if assigned(current_module.globalsymtable) then
current_module.globalsymtable.SymList.ForEachCall(@AddToThreadvarList,ltvTable);
current_module.localsymtable.SymList.ForEachCall(@AddToThreadvarList,ltvTable);
- if ltvTable.first<>nil then
+ if not ltvTable.empty then
begin
s:=make_mangledname('THREADVARLIST',current_module.localsymtable,'');
{ end of the list marker }
_______________________________________________
fpc-devel maillist - fpc-devel@lists.freepascal.org
http://lists.freepascal.org/mailman/listinfo/fpc-devel
