RE: [boost] Re: CRC library error

From: Björn Karlsson (Bjorn.Karlsson_at_[hidden])
Date: 2002-09-26 01:38:13

> From: David Abrahams [mailto:dave_at_[hidden]]
> 
> Volodya,
> 
> You might want to sick Bjorn Karlsson on him. Bjorn's the maintenance
> wizard.
> 
> -Dave

Indeed :-), this issue is present on my map. However, we should give the
author/maintainer a fair chance (read: a few days) to respond, and take
appropriate action. In this particular case (and AFAICT), the patch is a
workaround for a compiler - this is arithmetic with unsigned integral types,
which is guaranteed not to overflow; the compiler is doing something it
shouldn't, and the code is perfectly legal (this doesn't mean that it's less
annoying for the user, but it does matter). 

So, a few questions are seeking answers:

* Should the patch be applied for specific compilers (there may be more than
one)?
* Should 0xFF be used (at least in theory, not portable), or UCHAR_MAX, or
std::numeric_limits<unsigned char>::max(),...?
* Is there a (measured, I'm sure) penalty for the extra operation, i.e, it
won't be optimized?
* Does this need a fix in more places (the reported problem arises when
using crc_optimal)?

It ain't wizardry...but I do think that Daryle holds the best (and correct!)
answers to these questions.

Bjorn Karlsson