Re: [lmi] Reading an empty file into a string

[Greg Chicares]

On 2008-09-21 16:26Z, Vadim Zeitlin wrote:
> 
>  Otherwise replacing the in_avail() check with "is.rdbuf()->sgetc() !=
> Traits::eof()" should work too: sgetc() will result in an attempt to read
> from the file and if it fails for any reason other than the file being
> empty, it should set the failbit or badbit. IOW I propose the following
> patch:

I'll integrate that soon, as well as your other patch. I had just
finished an update of my own, so I committed that to get it out
of the way, but I'll merge yours, too. Thanks--I thought it'd be
good to add a C benchmark, and I was trying to remember how to do
file IO in C. It's been so long that I would have forgotten BUFSIZ
or whatever it is.

>  FWIW here is the result under Linux with g++ 4.1.2:
> 
>   Method 0: 2.758e-03 s =    2757620 ns, mean of 100 iterations
>   Method 1: 3.529e-05 s =      35289 ns, mean of 284 iterations
> 
> I'm not quite sure if everything is correct here (why 284 iterations for
> the second method?)

That's because of my 20080915T1414Z change:
  "Improve timing accuracy for operations that take very little time."
I wanted to time something in 'expression_template_0_test.cpp'
that was so fast that one hundred iterations took almost no time,
so successive timings varied a great deal. To increase stability,
I changed the timing algorithm so that it never spends less than
one percent of the maximum time allowed, even if the iteration
counter has reached one hundred.