Re: [boost] [Review] Boost.Endian mini-review

Subject: Re: [boost] [Review] Boost.Endian mini-review
From: Peter Dimov (lists_at_[hidden])
Date: 2015-01-26 10:11:20

Beman Dawes wrote:

> My expectation is that the class templates will only be used directly is 
> in highly generic code where specifying the integer type is not a problem.

Your expectation is incorrect. Some file formats support both little and big 
endian variants (to make it easier for writers, at readers' expense). It's 
typical to do

read( signature );

if( signature == 0x12345678 )
{
    read_header<little_endian>( h );
}
else if( signature == 0x78563412 )
{
    read_header<big_endian>( h );
}
else
{
    // not a valid file
}

which is "lowly generic".

That said, I support your decision to have both the type and the size as 
template parameters; as I demonstrated in my example, the type is decided by 
the program, and the size is decided by the file format, so in general, the 
two do not necessarily match.

However, this allows me to sneak in another comment. Currently, if I try to 
put 0x123456 into a 16 bit buffer, it silently cuts it to 0x3456 without 
missing a beat. This is sometimes, but very rarely, what we want. Typically, 
what one wants in this case is an exception, because this usually happens 
when writing a file, and if one tries to write 0x123456 and only 0x3456 
comes out, it's likely that the result will be corrupted and unreadable.