Re: Ada Shootout program for K-Nucleotide (patches)

[Ole-Hjalmar Kristensen <ole-hjalmar.kristensen@substitute_employer_here.com>]

>>>>> "GB" == Georg Bauhaus <rm.tsoh.plus-bug.bauhaus@maps.futureapps.de> writes:

    GB> Ole-Hjalmar Kristensen wrote:
    >>>>>>> "GB" == Georg Bauhaus <rm.dash-bauhaus@futureapps.de> writes:
    >> 
    >> I think the problem is not the explicit copy which happens when we hit
    >> the end of the buffer (easily verified, just increase the buffer size
    >> 10 times, the copying will happen 1/10 of the time, but execution time
    >> stays the same), but the implicit copy which seems to happen in the statement
    >> return In_Buf(Start..I-1); I do not know if this copy is unavoidable,
    >> or if the compiler could have optimized it away.

    GB> It looks like copying In_Buf(Start..I-1) be circumvented.
    GB> I have added a solution to Line_IO.

    GB> read() performance vs. Stream_IO.Read performance seems
    GB> to depend on the OS.  On a Mac here, read() is faster.
    GB> On Ubuntu, Stream_IO.Read is faster.

Strange. I cannot really imagine how Stream_IO.Read is faster since it has to
do the system call to do the actual read anyway. One possibility is
that In_Buf is not properly aligned, so that read() into In_Buf is
slower than the equivalent read() into the C stdio buffers. The system
call should be the same in any case. 

    GB> There is an issue with GNATs and End_File (GNATs other
    GB> than I guess the one you have). When comparing N <= 0,
    GB> ``operator for type "System.Crtl.size_t" is not
    GB> directly visible''.  To work around this, I wrote

    GB>    function End_File return Boolean is
    GB>       type Size_T is mod 2**Standard'Address_Size;
    GB>       N0 : constant Size_T := Size_T(N);
    GB>       Zero : constant Size_T := 0;
    GB>    begin
    GB>       return N0 <= Zero;
    GB>    end End_File;

Yes. I was using an old version of GNAT.

    GB> Finally, I think Get_Line_Test can be made more portable among Unix
    GB> compilers by changing GNAT specific Interfaces.C_Streams
    GB> into Interfaces.C; Could be slightly less convenient to
    GB> write, though?

I only included Interfaces.C_Streams because it was a convenient place
to get hold of the types needed to import the read() function, so if
you find them in Interfaces.C, that's just fine.

-- 
   C++: The power, elegance and simplicity of a hand grenade.