[cfe-commits] [PATCH] Use the new APFloat::convertToInt(APSInt) function (issue4728042)

[Eli Friedman]

On Thu, Jul 14, 2011 at 6:06 PM, Jeffrey Yasskin <jyasskin at google.com> wrote:
> On Thu, Jul 14, 2011 at 5:30 PM, Eli Friedman <eli.friedman at gmail.com> wrote:
>> On Thu, Jul 14, 2011 at 5:06 PM,  <jyasskin at gmail.com> wrote:
>>> Reviewers: cfe-commits_cs.uiuc.edu,
>>>
>>> Message:
>>> Initial diff at
>>> http://codereview.appspot.com/download/issue4728042_1.diff.
>>>
>>> Description:
>>> Use the new APFloat::convertToInt(APSInt) function to simplify uses of
>>> convertToInt(integerParts*) and make them more reliable.
>>>
>>> Relies on http://codereview.appspot.com/4738043.
>>>
>>> Please review this at http://codereview.appspot.com/4728042/
>>>
>>> Affected files:
>>>   M lib/AST/ExprConstant.cpp
>>>   M lib/Sema/SemaChecking.cpp
>>>
>>>
>>> Index: lib/AST/ExprConstant.cpp
>>> diff --git a/lib/AST/ExprConstant.cpp b/lib/AST/ExprConstant.cpp
>>> index
>>> 9943222b1dd30d9761731fca984d20d6b578f788..03c0747efba9e8ed9451607b6a33c8b1c9716196
>>> 100644
>>> --- a/lib/AST/ExprConstant.cpp
>>> +++ b/lib/AST/ExprConstant.cpp
>>> @@ -223,12 +223,10 @@ static APSInt HandleFloatToIntCast(QualType DestType,
>>> QualType SrcType,
>>>    // Determine whether we are converting to unsigned or signed.
>>>    bool DestSigned = DestType->isSignedIntegerOrEnumerationType();
>>>
>>> -  // FIXME: Warning for overflow.
>>> -  uint64_t Space[4];
>>> +  APSInt Result(DestWidth, !DestSigned);
>>>    bool ignored;
>>> -  (void)Value.convertToInteger(Space, DestWidth, DestSigned,
>>> -                               llvm::APFloat::rmTowardZero, &ignored);
>>> -  return APSInt(llvm::APInt(DestWidth, 4, Space), !DestSigned);
>>> +  (void)Value.convertToInteger(Result, llvm::APFloat::rmTowardZero,
>>> &ignored);
>>> +  return Result;
>>>  }
>>
>> Why are you removing the "FIXME: Warning for overflow" comment?  The
>> issue still exists...
>
> Oops. I thought that was talking about overflowing the 4 slots in
> Space. I've put it back.
>
>>>  static APFloat HandleFloatToFloatCast(QualType DestType, QualType SrcType,
>>> Index: lib/Sema/SemaChecking.cpp
>>> diff --git a/lib/Sema/SemaChecking.cpp b/lib/Sema/SemaChecking.cpp
>>> index
>>> 267cda8f4016ca935873eac6d22bb82906201051..d1990f3061f353f3152a33d450b971b445d95a10
>>> 100644
>>> --- a/lib/Sema/SemaChecking.cpp
>>> +++ b/lib/Sema/SemaChecking.cpp
>>> @@ -2990,17 +2990,15 @@ void DiagnoseFloatingLiteralImpCast(Sema &S,
>>> FloatingLiteral *FL, QualType T,
>>>    if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble)
>>>      return;
>>>
>>> -  // Try to convert this exactly to an 64-bit integer. FIXME: It would be
>>> -  // nice to support arbitrarily large integers here.
>>> +  // Try to convert this exactly to an integer.
>>>    bool isExact = false;
>>> -  uint64_t IntegerPart;
>>> -  if (Value.convertToInteger(&IntegerPart, 64, /*isSigned=*/true,
>>> +  llvm::APSInt IntegerValue(S.Context.getIntWidth(T),
>>> +                            /*isUnsigned=*/false);
>>> +  if (Value.convertToInteger(IntegerValue,
>>>                               llvm::APFloat::rmTowardZero, &isExact)
>>>        != llvm::APFloat::opOK || !isExact)
>>>      return;
>>>
>>> -  llvm::APInt IntegerValue(64, IntegerPart, /*isSigned=*/true);
>>> -
>>>    std::string LiteralValue = IntegerValue.toString(10, /*isSigned=*/true);
>>>    S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer)
>>>      << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue);
>>
>> I was going to ask why you're forcing this to be signed, but then I
>> realized it's impossible to write a negative floating-point literal.
>> :)
>
> More because the previously-existing code made it signed. I didn't
> think hard enough to notice that negative floating literals are
> impossible. Using signed integers may actually be a bug, since if the
> target is a uint64, and the floating literal doesn't fit in an int64,
> this will fail the conversion even though the result would fit.

Hmm... might actually be a regression for something like "unsigned
short x = 60000.;" ?  Should be trivial to call isSignedIntegerType()
here, though.

Otherwise looks fine.

-Eli