Re: [[PATCH RFC 2] 02/63] Y2038: add function __difftime64

[Paul Eggert <eggert@cs.ucla.edu>]

On 05/03/2018 10:30 AM, Albert ARIBAUD wrote:
> the simpler code in difftime.c which 64-bit signed integer
> needs is enclosed in run-time conditionals, not compile-time
> conditionals


These run-time conditionals are optimized away, and have zero run-time 
overhead.

Really, you should use difftime.c and see what it does (look at the 
machine code that is generated if you like). You will see that it does 
the right thing and generates the same instructions that your copy 
would. Similarly for many of the other routines that you incorrectly 
think we need to make source-code copies of.


> I am choosing not to assume how clever compilers can be, and I consider
> that putting a run-time test in some source code will be costlier in
> execution time than not putting it there.

This is completely, 100%, backwards. It is routine in high-quality 
software nowadays to assume decent optimizing compilers; for example, 
that's why we now use functions rather than macros, since simple 
functions are routinely inlined. The glibc implementation assumes GCC, 
it is not our job to second-guess GCC, and we should not add complexity 
to glibc when GCC does a perfectly good job of optimizing simpler code.