Why CTFE is context-sensitive?

[Ola Fosheim Grøstad" <ola.fosheim.grostad+dlang at gmail.com>]

On Monday, 27 January 2014 at 22:40:19 UTC, Simen Kjærås wrote:
> My computer has 16GB of RAM. You try keeping track of all the 
> possible states.

Wrong reasoning.

The issue is that the proof for the halting problem does assume 
non-finite resources. That means that the proof that you cannot 
prove termination for an arbitrary program does not hold for 
programs that run on your computer. :^)

Anyway, the reasoning does not hold anyway, because there is an 
inifinite number of programs that can be proved to terminate… So 
you just do those instead!

> computers. It's technically true, yes. It also does not matter, 
> for the reason described above.

When people try to be theoretical smartasses they should fess up 
to a theoretical response!

 From a pragmatic position speculative precomputation is quite 
acceptable, works well with profiling on typical datasets and 
whole program analysis.

Even if the compution does not terminate you can run it as far as 
you can and then freeze the result, and use that as a starting 
point.

A cheap version of this has been used for decades to speed up 
program initialization: run the program until it reads input, 
then core dump. Run a program on the core dump that turns it into 
an executable. This is a nice way to speed up the initialization 
of interpreted programs. Java needs this badly!

> Lastly, deadalnix didn't even bring up the halting problem, so 
> why even mention it?

He did implicitly, it was referenced in d.learn. This thread runs 
in two forums.

Ola