• yaneurao <yaneurao_member pathlink.com> Jan 28 2004
• "Serge K" <skarebo programmer.net> Jan 28 2004
• yaneurao <yaneurao_member pathlink.com> Jan 29 2004
• "Robert" <no spam.ne.jp> Jan 29 2004
• "Walter" <walter digitalmars.com> Jan 29 2004
• yaneurao <yaneurao_member pathlink.com> Jan 29 2004
• "C" <dont respond.com> Jan 29 2004

yaneurao <yaneurao_member pathlink.com> writes:

I report some bugs.
if you understand this problems , neglect me , Mr.Walter.

1. GC disable/enable are not implemented.

it needs to write in fullCollect() in gcx.d
if (gcx.disabled > 0) return 0;

2. object.d(14): alias size_t conflicts with time.size_t at std\c\time.d(10)
eg.
import std.c.time
private import std.c.time;
size_t a; // compile error

3.a bug in dchar decode(wchar[] s, inout uint idx) in std.utf.toUTF32
at std.utf.toUTF32(222) , ' else { i++; }' is needed.

4.private import in a class is not 'private'.
eg. in std.stream.d(51) :
class Stream
{
	private import std.string, crc32, std.c.stdlib, std.c.stdio;


We can access std.c.stdlib via Stream.

5.
I've examined the bug about :
D/22628
D/21999

the access violation code is :

004042B8  |. FF50 14        CALL DWORD PTR DS:[EAX+14]
(where EAX = 409f88 , [EAX+14] = e8004996)

the call stack :
ESP ==>  > 004042BB  RETURN to dmd.004042BB
ESP+4    > 003AAC90  ASCII "x€I"
ESP+8    > 003AB444
ESP+C    > 003AE770
ESP+10   > 003AE770
ESP+14   > 00403FB6  RETURN to dmd.00403FB6 from dmd.004042AC

I guess the address is DsymbolTable::lookup in dsymbol.c(519)
    sv = tab->lookup(ident->toChars(),strlen(ident->toChars()));


ident->toChars() makes access violation when 'ident' is something wrong ,
for example , a struct name is the same as a module name.

6.pre-declaration of class makes compiler crash.

eg.
class foo;

I've tested in WindowsXP.
perhaps , 6. is the same cause as 5.

7.a variable use in naked function

class Test {
void	test(){
m = 1;
asm {
naked;
ret;
}
}
int m;
void print() { printf("%d\n",m); }
}

int main() {

Test t = new Test;
t.test();
t.print();
return 0;
}

result : 0 ( I assume it should be 1.)
(compile with -debug option)

compiler generates the following code :
 mov [EAX + 8], 1
 ret


I assume the next code would be right when compile with -debug option.

push EAX;
mov [EAX + 8],1
pop EAX
ret

yaneurao.

"Serge K" <skarebo programmer.net> writes:

 compiler generates the following code :
 mov [EAX + 8], 1
 ret


 I assume the next code would be right when compile with -debug option.

 push EAX;
 mov [EAX + 8],1
 pop EAX
 ret


??? >8-/
There is no need to preserve EAX around this "mov", since it does not modify
any register.

yaneurao <yaneurao_member pathlink.com> writes:

In article <bv9st6$2no5$1 digitaldaemon.com>, Serge K says...
 I assume the next code would be right when compile with -debug option.
 push EAX;
 mov [EAX + 8],1
 pop EAX
 ret


There is no need to preserve EAX around this "mov", since it does not modify
any register.


sorry for my bad explanation. compiler generates ,

mov EAX,[EBP+4]
mov [EAX + 8], 1
ret

compiler assumes that [EBP+4] is 'this' pointer ,
but it doesn't so in a naked function compiled with -debug option.
and EAX is 'this' pointer by chance in a naked function.

therefore , it doesn't need to generate  mov EAX,[EBP+4] ,
or it is better to preserve EAX to access member variables.

yaneurao.

"Robert" <no spam.ne.jp> writes:

"yaneurao" <yaneurao_member pathlink.com> wrote in message
news:bvap19$15qi$1 digitaldaemon.com...
 In article <bv9st6$2no5$1 digitaldaemon.com>, Serge K says...
 I assume the next code would be right when compile with -debug option.
 push EAX;
 mov [EAX + 8],1
 pop EAX
 ret


There is no need to preserve EAX around this "mov", since it does not




 sorry for my bad explanation. compiler generates ,

 mov EAX,[EBP+4]
 mov [EAX + 8], 1
 ret

 compiler assumes that [EBP+4] is 'this' pointer ,
 but it doesn't so in a naked function compiled with -debug option.
 and EAX is 'this' pointer by chance in a naked function.

 therefore , it doesn't need to generate  mov EAX,[EBP+4] ,
 or it is better to preserve EAX to access member variables.

 yaneurao.


Not -debug.
-g is right.

See D/20418

"Walter" <walter digitalmars.com> writes:

"yaneurao" <yaneurao_member pathlink.com> wrote in message
news:bvap19$15qi$1 digitaldaemon.com...
 In article <bv9st6$2no5$1 digitaldaemon.com>, Serge K says...
 I assume the next code would be right when compile with -debug option.
 push EAX;
 mov [EAX + 8],1
 pop EAX
 ret


There is no need to preserve EAX around this "mov", since it does not




 sorry for my bad explanation. compiler generates ,

 mov EAX,[EBP+4]
 mov [EAX + 8], 1
 ret

 compiler assumes that [EBP+4] is 'this' pointer ,
 but it doesn't so in a naked function compiled with -debug option.
 and EAX is 'this' pointer by chance in a naked function.

 therefore , it doesn't need to generate  mov EAX,[EBP+4] ,
 or it is better to preserve EAX to access member variables.


With a 'naked' function, the programmer must set up the stack frame, as the
'naked' tells the code generator to not do it automatically. Realistically,
with 'naked' functions, you should write the entire function in assembler.

In other words, this is working as intended.

yaneurao <yaneurao_member pathlink.com> writes:

In article <bvbmn8$2nld$1 digitaldaemon.com>, Walter says...
With a 'naked' function, the programmer must set up the stack frame, as the
'naked' tells the code generator to not do it automatically. Realistically,
with 'naked' functions, you should write the entire function in assembler.


I agree. I think it is better to transcribe it.

yaneurao.

"C" <dont respond.com> writes:

Good reports.  On a side note the new .79 has broken my code , and I finnaly
ran into the dreaded import trouble.  How does changing from module level
imports to class level fix this , and why is this a problem ?  Also we
should warn new users of it.

C
"yaneurao" <yaneurao_member pathlink.com> wrote in message
news:bv9gf2$24bl$1 digitaldaemon.com...
 I report some bugs.
 if you understand this problems , neglect me , Mr.Walter.

 1. GC disable/enable are not implemented.

 it needs to write in fullCollect() in gcx.d
 if (gcx.disabled > 0) return 0;

 2. object.d(14): alias size_t conflicts with time.size_t at


 eg.
 import std.c.time
 private import std.c.time;
 size_t a; // compile error

 3.a bug in dchar decode(wchar[] s, inout uint idx) in std.utf.toUTF32
 at std.utf.toUTF32(222) , ' else { i++; }' is needed.

 4.private import in a class is not 'private'.
 eg. in std.stream.d(51) :
class Stream
{
 private import std.string, crc32, std.c.stdlib, std.c.stdio;


 We can access std.c.stdlib via Stream.

 5.
 I've examined the bug about :
 D/22628
 D/21999

 the access violation code is :

 004042B8  |. FF50 14        CALL DWORD PTR DS:[EAX+14]
 (where EAX = 409f88 , [EAX+14] = e8004996)

 the call stack :
 ESP ==>  > 004042BB  RETURN to dmd.004042BB
 ESP+4    > 003AAC90  ASCII "x?I"
 ESP+8    > 003AB444
 ESP+C    > 003AE770
 ESP+10   > 003AE770
 ESP+14   > 00403FB6  RETURN to dmd.00403FB6 from dmd.004042AC

 I guess the address is DsymbolTable::lookup in dsymbol.c(519)
    sv = tab->lookup(ident->toChars(),strlen(ident->toChars()));


 ident->toChars() makes access violation when 'ident' is something wrong ,
 for example , a struct name is the same as a module name.

 6.pre-declaration of class makes compiler crash.

 eg.
 class foo;

 I've tested in WindowsXP.
 perhaps , 6. is the same cause as 5.

 7.a variable use in naked function

 class Test {
 void test(){
 m = 1;
 asm {
 naked;
 ret;
 }
 }
 int m;
 void print() { printf("%d\n",m); }
 }

 int main() {

 Test t = new Test;
 t.test();
 t.print();
 return 0;
 }

 result : 0 ( I assume it should be 1.)
 (compile with -debug option)

 compiler generates the following code :
 mov [EAX + 8], 1
 ret


 I assume the next code would be right when compile with -debug option.

 push EAX;
 mov [EAX + 8],1
 pop EAX
 ret

 yaneurao.