Submission Date: 03 Dec 93
Submittor: WG14
Source: Clive Feather
Question
Item 18 - left shift operator
The result of the left shift operator E1 < E2, when E1
is signed, is defined (subclause 6.3.7) as E1 left-shifted by
E2 bits, with vacated bits filled with zeros. But what exactly
does this mean?
The C Standard defines a bit (subclause 3.3) only as a unit of data
storage. Bits are related to the value of an object
only in subclause 6.1.2.5,
which specifies the representation of certain types. It may therefore
be claimed that the left shift operator must act on representations,
which are of fixed length. In this interpretation, the left E2
bits (including the sign bit) are lost, as they would be if E1
was unsigned; the sign bit of the result is taken from a bit in E1,
E2 places to the right of the sign bit and, provided that the
resultant bit pattern actually represents a value of the result type,
an exception is impossible.
On the other hand, it may also be claimed that the whole of subclause
6.3 specifies the meaning of operations in abstract mathematical terms,
subject to the general note about exceptions. In this view, the bit
sequence representing the non-sign part of a signed integer is converted
by the shift operation to a bit sequence of indefinite length, and,
to avoid an exception due to overflow, this bit sequence must fit
back in the non-sign part without the loss at the left of anything
but copies of the sign bit.
Which of these two views is correct?
If the answer to (a) is the first view, does undefined behavior
occur if the resulting bit pattern is not the representation of an
integer?
The following questions apply only if the answer to (a) is that the
second view is correct.
If E1 is positive, and E1 times 2 to the power E2
is less than or equal to INT_MAX (or LONG_MAX), is the
result always E1 times 2 to the power E2?
Under what circumstances is the result undefined?
Response
Subclause 6.3 states that the binary operator < ,
among others, has implementation-defined aspects for signed types.
Therefore, the answer to ``What does it mean to left shift a signed
value?'' is that it is implementation-defined.
Previous Defect Report
< - >
Next Defect Report