3 ms·

This used to be unspecified behavior. C99 then codified this wrong behavior. Assume positive b for a moment. We want a * (a/b) + (a%b) = a. If a%b is to alwa

by wnoise 7y ago

This used to be unspecified behavior. C99 then codified this wrong behavior. Assume positive b for a moment. We want a * (a/b) + (a%b) = a. If a%b is to always be within [0..b), then a/b has to round toward -infinity. C99 instead chose round towards 0.

kragen 7y ago
That's right. Why do you suppose they chose that behavior to standardize, rather than Python's? Conceivably it's because nobody on the C99 standards committee had enough technical expertise to make the argument you're making, but can you think of another explanation? Because the prior probability on that one is pretty low.
- wahern 7y ago
  > In C89, division of integers involving negative operands could round upward or downward in an implementation-defined manner; the intent was to avoid incurring overhead in run-time code to check for special cases and enforce specific behavior. In Fortran, however, the result will always truncate toward zero, and the overhead seems to be acceptable to the numeric programming community. Therefore, C99 now requires similar behavior, which should facilitate porting of code from Fortran to C. Source: C99 Rationale v5.10 sec. 6.5.5p10 (http://www.open-std.org/jtc1/sc22/wg14/www/C99RationaleV5.10.pdf http://www.open-std.org/jtc1/sc22/wg14/www/C99RationaleV5.10...)
  - bloak 7y ago
    Thanks for that. Now can anyone tell us where the Fortran behaviour came from? (Haskell has both kinds of integer division, called div and quot, which is nice.)
derriz 7y ago
Don't you mean we want "(a/b) + (a%b) = a"?
- kragen 7y ago
  > Don't you mean we want "(a/b) + (a%b) = a"? Like, 1000/1000 + 1000%1000 = 1000? That sounds like undesirable behavior for division to me.
- a1369209993 7y ago
  > We want a * (a/b) + (a%b) = a. > Don't you mean we want "(a/b) + (a%b) = a"? No, they mean "b * (a/b) + (a%b) = a". (Also I'm not convinced that's actually useful, but that's a different discussion.)
  - derriz 7y ago
    Ha - you’re correct of course. I saw the a^2 term and thought that can’t be right. Note to self - before attempting to correct others, check the “correction”.
wnoise 7y ago
As others have pointed out, I did mean b*(a/b) here.

by wnoise 7y ago

kragen 7y ago

That's right. Why do you suppose they chose that behavior to standardize, rather than Python's? Conceivably it's because nobody on the C99 standards committee had enough technical expertise to make the argument you're making, but can you think of another explanation? Because the prior probability on that one is pretty low.

wahern 7y ago
> In C89, division of integers involving negative operands could round upward or downward in an implementation-defined manner; the intent was to avoid incurring overhead in run-time code to check for special cases and enforce specific behavior. In Fortran, however, the result will always truncate toward zero, and the overhead seems to be acceptable to the numeric programming community. Therefore, C99 now requires similar behavior, which should facilitate porting of code from Fortran to C. Source: C99 Rationale v5.10 sec. 6.5.5p10 (http://www.open-std.org/jtc1/sc22/wg14/www/C99RationaleV5.10.pdf http://www.open-std.org/jtc1/sc22/wg14/www/C99RationaleV5.10...)
- bloak 7y ago
  Thanks for that. Now can anyone tell us where the Fortran behaviour came from? (Haskell has both kinds of integer division, called div and quot, which is nice.)

derriz 7y ago

Don't you mean we want "(a/b) + (a%b) = a"?

kragen 7y ago
> Don't you mean we want "(a/b) + (a%b) = a"? Like, 1000/1000 + 1000%1000 = 1000? That sounds like undesirable behavior for division to me.
a1369209993 7y ago
> We want a * (a/b) + (a%b) = a. > Don't you mean we want "(a/b) + (a%b) = a"? No, they mean "b * (a/b) + (a%b) = a". (Also I'm not convinced that's actually useful, but that's a different discussion.)
- derriz 7y ago
  Ha - you’re correct of course. I saw the a^2 term and thought that can’t be right. Note to self - before attempting to correct others, check the “correction”.

wnoise 7y ago

As others have pointed out, I did mean b*(a/b) here.