Issues when compiling user4martin branch on Delphi 7

Alexey-T commented 1 year ago

@user4martin

Missed type SizeInt. we have PtrInt specially defined in regexpr.pas so use PtrInt
Missed func StrLComp for PWideChars. it requires PChars.

I wait for these fixes in your branch look-around-new

Alexey-T commented 1 year ago

More

Exit(False) is not supported

Alexey-T commented 1 year ago

Operator is not applicable to operand type for:

scan := AlignToPtr(scan + 1) + SizeOf(TRENextOff);

can we change it to:

scan := AlignToPtr(scan + 1 + SizeOf(TRENextOff));

User4martin commented 1 year ago

PtrInt fix comming.

I can't find the PChar ?

function StrLPos(str1,str2 : PRegExprChar; len1, len2: PtrInt) : PRegExprChar;

It will be the same kind of char, than the rest of the regex.

And the call is also ensuring all params are the correct type (regMustString: RegExprString;)

      if StrLPos(fInputStart, PRegExprChar(regMustString), fInputEnd - fInputStart, length(regMustString)) = nil then

Alexey-T commented 1 year ago

PChar/PWideChar. to fix it, copy StrLComp from FPC to regexpr.pas. adjust types to RegExpr types.

function strlcomp(str1,str2 : pwidechar;l : SizeInt) : SizeInt;
  var
    counter: sizeint;
    c1, c2: widechar;
  begin
    if l = 0 then
      begin
        strlcomp := 0;
        exit;
      end;
    counter:=0;
    repeat
      c1:=str1[counter];
      c2:=str2[counter];
      inc(counter);
   until (c1<>c2) or (counter>=l) or
         (c1=#0) or (c2=#0);
    strlcomp:=ord(c1)-ord(c2);
  end;

Alexey-T commented 1 year ago

Bug

function TRegExpr.IsFixedLength result may be undefined (not Delphi error, general error)

User4martin commented 1 year ago

Operator is not applicable to operand type for:

It will be RENextOffSz since then it's pointer math. Anyway that part is ok.

However it isn't the same. => It only currently would be the same (by accident)

Currently TRENextOff = PtrInt;

So aligning to pointer firs or last, will lead to the same end result.

But if ever the above changes to TRENextOff = LongInt; (on a platform with 8 byte ptr, and 4 byte longint) then it will break.

Say scan is at 0x...1

alignToPtr => 0x...8
add 4 => 0x...C

But when adding first

add 4 => 0x...5
AlignToPtr => 0x...8

Since the align is part of EmitNode, and the data of size(TRENextOff) was therefore added after the AlignToPtr, the first calculation is correct.

I don't know all the Delphi details, but it should have directives on what you can do with pointers.

If not

TypeCast to PByte
Change the order, and willingly introduce the bug => since the entire AlignToPtr is currently broken. (Afaik)

Alexey-T commented 1 year ago

https://www.freepascal.org/docs-html/prog/progsu112.html

By default, pointer arithmetic is enabled in FPC/OBJFPC modes and disabled in DELPHI mode. The directive $POINTERMATH ON/OFF can be used to toggle pointer arithmetic support

so I cannot solve it - $POINTERMATH is not supported in D7. what to do?

Alexey-T commented 1 year ago

Martin, here are my changes, buggy of course because of Align-changes. please see which lines I must change regarding AlignToPtr. can you change these Aligns on ur side?

and StrLComp types are not adjusted...

diff --git a/src/regexpr.pas b/src/regexpr.pas
index fef6cb5..d04f187 100644
--- a/src/regexpr.pas
+++ b/src/regexpr.pas
@@ -136,6 +136,7 @@ type
     PtrInt = Int64;
     PtrUInt = UInt64;
     {$IFEND}
+    SizeInt = PtrInt;
   {$ENDIF}

   {$IFDEF UnicodeRE}
@@ -1044,6 +1045,26 @@ Begin
 end;

+function strlcomp(str1,str2 : pwidechar;l : SizeInt) : SizeInt;
+  var
+    counter: sizeint;
+    c1, c2: widechar;
+  begin
+    if l = 0 then
+      begin
+        strlcomp := 0;
+        exit;
+      end;
+    counter:=0;
+    repeat
+      c1:=str1[counter];
+      c2:=str2[counter];
+      inc(counter);
+   until (c1<>c2) or (counter>=l) or
+         (c1=#0) or (c2=#0);
+    strlcomp:=ord(c1)-ord(c2);
+  end;
+        
 function StrLPos(str1,str2 : PRegExprChar; len1, len2: SizeInt) : PRegExprChar;
 var
   p : PRegExprChar;
@@ -1235,7 +1256,10 @@ begin
     case APtr[i] of
       '-':
         if not IsOn then
-          Exit(False)
+        begin
+          result:= False;
+          Exit;
+        end
         else
           IsOn := False;
       'I', 'i':
@@ -5413,7 +5437,7 @@ begin
           LookAroundInfoList := @Local.LookAroundInfo;
           fInputCurrentEnd := fInputEnd;

-          scan := AlignToPtr(scan + 1) + SizeOf(TRENextOff);
+          scan := AlignToPtr(scan + 1 + SizeOf(TRENextOff));
           Result := MatchPrim(scan);

           if Local.LookAroundInfo.IsBackTracking then
@@ -5421,11 +5445,11 @@ begin
           LookAroundInfoList := Local.LookAroundInfo.OuterInfo;
           fInputCurrentEnd := Local.LookAroundInfo.savedInputCurrentEnd;

-          next := AlignToPtr(next + 1) + SizeOf(TRENextOff);
+          next := AlignToPtr(next + 1 + SizeOf(TRENextOff));
           if Local.IsNegativeLook then begin
             Result := (next^ = OP_LOOKAROUND_OPTIONAL);
             if Result then
-              next := AlignToPtr(next + 1) + SizeOf(TRENextOff)
+              next := AlignToPtr(next + 1 + SizeOf(TRENextOff))
             else
               Result := (not Local.LookAroundInfo.HasMatchedToEnd);
             if Result then begin
@@ -5436,7 +5460,7 @@ begin
           end
           else
           if (next^ = OP_LOOKAROUND_OPTIONAL) then begin
-            next := AlignToPtr(next + 1) + SizeOf(TRENextOff);
+            next := AlignToPtr(next + 1 + SizeOf(TRENextOff));
             if not Local.LookAroundInfo.HasMatchedToEnd then begin
               regInput := Local.LookAroundInfo.InputPos;
               Result := MatchPrim(next);
@@ -5453,7 +5477,7 @@ begin
       OP_LOOKBEHIND, OP_LOOKBEHIND_NEG:
         begin
           Local.IsNegativeLook := (scan^ = OP_LOOKBEHIND_NEG);
-          scan := AlignToPtr(scan + 1) + SizeOf(TRENextOff);
+          scan := AlignToPtr(scan + 1 + SizeOf(TRENextOff));
           Local.IsGreedy := PReOpLookBehindOptions(scan)^.IsGreedy;

           Local.LookAroundInfo.InputPos := regInput;
@@ -5497,11 +5521,11 @@ begin
           LookAroundInfoList := Local.LookAroundInfo.OuterInfo;
           fInputCurrentEnd := Local.LookAroundInfo.savedInputCurrentEnd;

-          next := AlignToPtr(next + 1) + SizeOf(TRENextOff);
+          next := AlignToPtr(next + 1 + SizeOf(TRENextOff));
           if Local.IsNegativeLook then begin
             Result := (next^ = OP_LOOKAROUND_OPTIONAL);
             if Result then
-              next := AlignToPtr(next + 1) + SizeOf(TRENextOff)
+              next := AlignToPtr(next + 1 + SizeOf(TRENextOff))
             else
               Result := not Local.LookAroundInfo.HasMatchedToEnd;
             if Result then begin
@@ -5512,7 +5536,7 @@ begin
           end
           else
           if (next^ = OP_LOOKAROUND_OPTIONAL) then begin
-            next := AlignToPtr(next + 1) + SizeOf(TRENextOff);
+            next := AlignToPtr(next + 1 + SizeOf(TRENextOff));
             if not Local.LookAroundInfo.HasMatchedToEnd then begin
               regInput := Local.LookAroundInfo.InputPos;
               Result := MatchPrim(next);
@@ -5538,7 +5562,7 @@ begin
             LookAroundInfoList := Local.LookAroundInfoPtr^.OuterInfo;

             if (next^ = OP_LOOKAROUND_OPTIONAL) then
-              next := AlignToPtr(next + 1) + SizeOf(TRENextOff);
+              next := AlignToPtr(next + 1 + SizeOf(TRENextOff));
             Result := MatchPrim(next);
             LookAroundInfoList := Local.LookAroundInfoPtr;
           end;
@@ -5567,7 +5591,7 @@ begin
             LookAroundInfoList := Local.LookAroundInfoPtr^.OuterInfo;

             if (next^ = OP_LOOKAROUND_OPTIONAL) then
-              next := AlignToPtr(next + 1) + SizeOf(TRENextOff);
+              next := AlignToPtr(next + 1 + SizeOf(TRENextOff));
             Result := MatchPrim(next);
             LookAroundInfoList := Local.LookAroundInfoPtr;
           end;
@@ -6625,7 +6649,7 @@ begin
       OP_BACK:
         begin
           // No point to rescan the code again
-          Next := AlignToPtr(scan + 1) + SizeOf(TRENextOff);
+          Next := AlignToPtr(scan + 1 + SizeOf(TRENextOff));
         end;

       OP_OPEN_FIRST .. OP_OPEN_LAST:

User4martin commented 1 year ago

Afaik it isn't PointerMath => PointerMath is the one that is working

SomePWideChar := SomePWideChar + 1 adds 2 to the address, because a WideChar has 2 bytes.

Without PointerMath, you need to add 2, to advance to the next char.

AlignToPtr return Pointer => which does not have a type. Adding to a pointer is the most basic thing... I wonder if Delphi really has an issue, or if it is something else that causes this.

But anyway, I can't test it, I don't have Delphi.

I will change the AlignToPtr Of course the SizeOf(TRENextOff) must be replaced by TRENextOffSz otherwise it adds twice the value.

Or well before, could you test:

 Next := PRegExprChar(AlignToPtr(scan + 1) + SizeOf(TRENextOff));

It probably isn't the add, but the assign.

User4martin commented 1 year ago

Ok I added a PR for the stuff that relates to already merged changes

StrInt / StrLComp

And the rename of TRegEx[pr]Anchor.

I'll go through the rest, but I believe they are all for the upcoming PR.

Alexey-T commented 1 year ago

Please update ur branch 'look-around-new' with last fixes. i will pull it and test on D7.

User4martin commented 1 year ago

Please update ur branch 'look-around-new' with last fixes. i will pull it and test on D7.

on the way.

Did you check if the Next := PRegExprChar(AlignToPtr(scan + 1) + SizeOf(TRENextOff)); works? Or do you rather not have the pointer?

Because Next := AlignToPtr(scan + 1 + SizeOf(TRENextOff)); or actually Next := AlignToPtr(scan + 1 + TRENextOffSz);

Is using AlignToPtr wrong => in that case, do what half of the other code does, and remove it? Next := scan + 1 + TRENextOffSz; ?

User4martin commented 1 year ago

AlignToPtr is still pending. The rest should be updated.

AlignToPtr => I go with whatever option you like. Just wanted the different options on the table/considered, before making the change. (see my last comment please.)

User4martin commented 1 year ago

SHould probably be scan := PRegExprChar(AlignToPtr(scan + 1)) + RENextOffSz;

That way the cast happens before the addition, the addition uses the adjusted value

Alexey-T commented 1 year ago

Yes! I also just found it! this works: temp := PRegExprChar(AlignToPtr(scan + 1)) + SizeOf(TRENextOff);

User4martin commented 1 year ago

good, give me 5 minutes

User4martin commented 1 year ago

OK, pushed

Alexey-T commented 1 year ago

It compiles. Tks.

andgineer / TRegExpr

Issues when compiling user4martin branch on Delphi 7 #314