@Markiyan Hirnyk 

Of course. But this could happen with any other algorithm - this is the nature of the problem!

On the other side, your example with RandomSolid will be solved at once!

Edit.
P.S. Nothing works in (too) higher dims.

Unfortunately there are these problems:

1. LinearMultivariateSystem seems to work properly only for relatively simple systems
2. isolve is very slow and gives "Warning, solutions may have been lost"

I would recommend the following algorithm:

a.) Given the polyhedral P, find a minimal cube a1..b1, ..., an..bn  containing P.
ak, bk can be found using simplex:

ak = eval(xk, simplex[minimize](xk, P));
bk = eval(xk, simplex[maximize](xk, P));

b.) Use  cartprod for the intervals ceil(ak) .. floor(bk)
and return the elements which are in P.

It was a joke, see the :-) sign.

What I don't understand is why convert,exp acts like this.

@acer 

@Bendesarts 

For f defined in [a,b) use

f_per:=x -> f(x - floor((x-a)/(b-a)) * (b-a));

Should simplify to 0 even for complex s,t   :-)

[evalc assumes s,t real]

@acer

Your f@p is not periodic; try it for -8..8

@Christian Wolinski

@Bendesarts

You cannot have an infinity of branches in piecewise. And you don't need that.

If f is defined in [-1,1) instead of [0,2) then use:

f := x -> exp(-x);   #  -1 <= x  <  1
f_persym:= x -> f( x - 2*floor(x/2+1/2) );

# for a period T>0, use   f( x - T*floor(x/T+1/2) )  

Probably "product" means "componentwise product", *~  in Maple.

f is probably assumed to be linear on the span of the vectors; but the vectors not being assumed linearly independent, the definition of f could be incorrect.

For example all the vectors could be 0; or all <1,1,...,1>.

Conclusion: problem not doable as it is.

@Carl Love 

I am not a native English speaker (probaby this is obvious). In my native language (Romanian, btw) numeral=number.

So, is it incorrect to say that that "the number 123 has three distinct digits"? Is it mandatory to use numeral?

@nMaple 

Carl's solutions works; probably you have an older version.

In Rouben's solution you will have to consider the cases t <= -2 and t>= 10 which are not covered.

@H-R 

It's not so complicated; it is in this way that Maple and maths works.

You can't ignore the complex numbers even if you are working in R. E.g. you will not be able to completely understand why the power series of 1/(1+x^2) converges only in (-1,1).

RealDomain is an acceptable (somehow) compromise.

@Christopher2222 

1 - 9^3/10^3;

    271/1000

@H-R 

You must think mathematically and anticipate the difficulties that may arise.
Here is a short list of problems.

- The entire CAS works over the complex field because this is the best option.
- It is well known that choosing the convenient branch of a multivalued complex function could be a hard task, Maple is not working with Riemann surfaces.
As a consequence, several definite integrals could be wrongly computed or indefinite integrals could be discontinuous. It would be almost imposible to avoid all such problems; there are scientific papers addressing them.
- It is not unusual to have intermediate complex expressions even if the final result is real.
- The assume facility is far from perfect and RealDomains relies on it.

Example:

Why not simply:

Heaviside(0):=0: Heaviside(0.0):=0.0:

?

@Axel Vogt 

I have also obtained a symbolic result using some conversions such as
Expand(%); Change(%,xp=t^(1/3));
subsindets(%,'specfunc(anything,int)', Split, [infinity]);

It's a pitty that one has to do such acrobatics instead of a simple convert(%,GAMMA).