#1608 Real-valued expressions do not type coerce sub-expressions

[Mark Foster]

The example below shows that despite the clear intention of these both being evaluated as real expressions by the use of the 3.0 constant in the front, the (3/2) sub-expression is still evaluated as integer. This is counter-intuitive and contrary to virtually all modern interpreted languages where use of a real-valued expression type-coerces the entire expression and all sub-expressions to real-value. Integer evaluation should only be used if integer valued components are used through-out an entire expression:

gnuplot> print 3.0**(3/2)
3.0
gnuplot> print 3.0**(3./2)
5.19615242270663

The gnuplot FAQ does warn about integer vs real valuation, but does not state that it applies at the sub-expression level and that no type-coercion is performed.

This bug is in both 4.6 and 5.0.

[Hans-Bernhard Broeker]

This is not a bug, it's documented behaviour.

gnuplot clearly states that its arithmetic is modelled after C, not after whatever the latest "new" language happens to do. Oh, and just FTR, by your logic, Python is not a modern interpreted language. Well either that, or you completey misunderstood what type-coercion means.

[Clark Gaylord]

Agreed 3/2 = 1

If you don't know what an integer is, use Excel.

[Mark Foster]

Why, yes, we could consider python to be the exception that proves the rule ;-)
And, citing C rules (with it's strong-typing warnings at compile-time) somehow doesn't seem really useful in a silently-interpreted environment, even if literally true.

Well, perhaps consider then as a feature request, as the inadvertent embedding of integer sub-expressions in an otherwise real-valued expression almost always silently leads to unintended consequences.

[Ethan Merritt]

Ruby also pokes a hole in your expectations:

chauvet [2] irb
irb(main):001:0> 3.0 ** (3/2)
=> 3.0
irb(main):002:0> 3.0 ** (3.0/2)
=> 5.196152422706632
irb(main):003:0> 

So what languages were you thinking of?

[Ethan Merritt]

Gnuplot is clearly not a "modern interpreted language", as it was designed 30 years ago. Please think of math in gnuplot as working like math in C/C++.

That said, I think your expectations about "virtually all modern interpreted languages" may be overly broad.

[1] python
Python 2.7.6 (default, Sep 29 2014, 16:37:37) 
[GCC 4.7.2] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> print 3.0**(3/2)
3.0
>>> >>> print 3.0**(3./2)
5.19615242271