Crash problem in 32bit but not in 64bit
lazarus / 62-General

----------------------------------------------------------------------
Tommi · 2018-07-11 12:27

This is my code attempts to convert a byte array to single:
Code: [Select]function bytes2float(var
dd:TMioR;tipo:byte;Precision:byte;aa:Tbytes;Posiz:integer):single;
overload;     //Prcision in byte non in bit
var
  oo:array [0..3] of byte;
begin
  try
    if tipo = 0 then
    begin
      if aa[posiz]<128 then
      begin
           if Precision = 1 then result:=aa[posiz];

           if Precision = 2 then result:=aa[posiz]*256+aa[posiz+1];
      end
      else
      begin
           if Precision = 1 then result:=-(255-aa[posiz]+1);

           if Precision = 2 then
result:=-(65535-(aa[posiz]*256+aa[posiz+1])+1);
      end;
    end;
    if tipo = 1 then
    begin
      oo[3]:=aa[Posiz];
      oo[2]:=aa[Posiz+1];
      oo[1]:=aa[Posiz+2];
      oo[0]:=aa[Posiz+3];
      move(oo,result,4);
      if IsNAN(result) then result:=0;
    end;
  Except
  on E : Exception do
  begin
     result:=0;
     setlength(dd.Errori,length(dd.Errori)+1);
     dd.Errori[Length(dd.Errori)-1].Testo:='Trovato errore generico.';
     dd.Errori[Length(dd.Errori)-1].Codice:=8;
  end;
  end;
end;

When tipo=1, sometime in 32bit this code crashes. In 64 works.

Some idea ?

Thank you

----------------------------------------------------------------------
Gammatester · 2018-07-11 12:49

Quote from: Tommi on July 11, 2018, 02:27:44 pmCode: [Select]     
begin
           if Precision = 1 then result:=-(255-aa[posiz]+1);

           if Precision = 2 then
result:=-(65535-(aa[posiz]*256+aa[posiz+1])+1);
      end;
    end;
Are you sure that  Code: [Select]if Precision = 1 then
result:=-(255-aa[posiz]+1) is correct and not
Code: [Select]if Precision = 1 then result:=-(255-aa[posiz+1])At all
other places the $+1$ is in the index expression.

----------------------------------------------------------------------
Tommi · 2018-07-11 12:50

It isn't there the prblem because it happens when tipo = 1 so when I
am here:
Code: [Select]    if tipo = 1 then
    begin
      oo[3]:=aa[Posiz];
      oo[2]:=aa[Posiz+1];
      oo[1]:=aa[Posiz+2];
      oo[0]:=aa[Posiz+3];
      move(oo,result,4);
      if IsNAN(result) then result:=0;
    end;

----------------------------------------------------------------------
Gammatester · 2018-07-11 13:00

Sorry, you are right. I was confused by the indentation.

----------------------------------------------------------------------
Tommi · 2018-07-11 13:02

specifically the problem borns when :

oo[0]=0
oo[1]=0
oo[2]=0
oo[1]=63

if I put result:=0.5 in the line above if IsNAN(result) then
result:=0;  the crash disappears.

----------------------------------------------------------------------
Tommi · 2018-07-11 13:24

The error is indipendent from the content of the function because if I
just put result:=0; in the first row of the function, the application
crashes there.

----------------------------------------------------------------------
Tommi · 2018-07-11 16:11

Issue solved creating a new project and copying the units. 32bit
version of lazarus cannot open 64bit created projects because it gives
this random errors.

----------------------------------------------------------------------
Thaddy · 2018-07-11 17:10

Quote from: Tommi on July 11, 2018, 06:11:56 pmIssue solved creating a
new project and copying the units. 32bit version of lazarus cannot
open 64bit created projects because it gives this random errors.
That would be a bug.

----------------------------------------------------------------------
Tommi · 2018-07-11 21:17

In few words: my code was working good compiled with 64bit version of
lazarus. I opened the same project with 32bit and I had this problem.
I reopened the same files (already modified with 32 bit version) with
64bit, and I had a lot of issues. GDB crashing and exe project
remaining in memory with no hope to close it from taskmanager, reboot
was necessary.

I created a new project copying the units and everything started to
work fine.

So I tried also to copy lpi, lpr, lps, res files coming from an old 64
bit version (not touched from the 32) in the ruined project and it
came back to work fine (compiling with 64bit).

Yes, I am pretty sure that it is a bug, but it is very difficult to
create a minimal project to reproduce it.