symbolic variable names should be valid identifiers, or ridiculousness follows

[williamstein]

WTF?

sage: var('1')
1
sage: var('1')+1
1 + 1
sage: expand((var('2')+var('2'))^2)
4*2^2

When this ticket is closed, #9724 can also be closed.

---

Apply attachment: trac_7496_check_symbolic_variable_names.patch and attachment: trac_7496-reviewer.2.patch.

Component: symbolics

Keywords: sd31

Author: Volker Braun, Karl-Dieter Crisman

Reviewer: Karl-Dieter Crisman, Volker Braun

Merged: sage-4.7.2.alpha0

Issue created by migration from https://trac.sagemath.org/ticket/7496

[burcin]

comment:1

This would be really easy to implement once we have python 3. Apparently, valid identifiers got extended:

http://www.python.org/dev/peps/pep-3131/

String literals also got an isidentifier() method.

I don't know how to do this efficiently, while still allowing people to just use α (alpha) as a variable name.

[simon-king-jena]

comment:2

As long as we don't have Python 3, I would try to find a regular expression that does what we need. It is of course easy to write a regular expression that works for ascii strings. But as soon as 'ä' or other language-specific letters are supposed to be considered a variable name, things will become difficult.

I tried

sage: import re
sage: _identifiers = re.compile("(?!\d)\w*\Z", re.LOCALE|re.UNICODE)

It works for simple cases:

sage: _identifiers.match('k_1')
<_sre.SRE_Match object at 0x50d1030>
sage: _identifiers.match('1k')
sage: _identifiers.match('_1k')
<_sre.SRE_Match object at 0x50d1238>

But it ignores other letters:

sage: print _identifiers.match('ä')
None

Perhaps I have misunderstood the effect of re.LOCALE? What value should re.LOCALE have in order to work with accented letters?

[robertwb]

comment:3

Whatever we do, let's be sure to avoid locality dependance on what a valid symbol is. Let's look into back-porting whatever Python3 does, and then falling back to that once we make the transition.

[vbraun]

Changed keywords from none to sd31

[vbraun]

Author: Volker Braun

[sagetrac-mariah]

Description changed:

--- 
+++ 
@@ -8,3 +8,5 @@
 sage: expand((var('2')+var('2'))^2)
 4*2^2
 ```
+
+When this ticket is closed, #9724 can also be closed.

[kcrisman]

Reviewer: Karl-Dieter Crisman

[kcrisman]

comment:6

This code should be correct - nice to learn something about the parser and bytecode. Is this how it's implemented in Python 3 (presumably that's written in C, though)? I couldn't find it, anyway.

This needs doctests just to document that var('3') and var(' ') (see #9724) fail, though of course the tests already here document this indirectly. I'll post a reviewer patch momentarily, assuming I didn't miss something and the patch doesn't actually work.

[kcrisman]

comment:7

Ok, I think this is still ok, though I am a little concerned about both of the following being bad:

sage: var(' x')
(, x)

not good because an empty string is a variable

sage: var(' x')
---------------------------------------------------------------------------
ValueError: The name "" is not a valid Python identifier.

not good because the intent is clear to make precisely x the variable.

So is this breaking incorrect but usable behavior?

sage: var("x y  z")
(x, y, , z)
sage: 

is similar.

Anyway, I withhold judgment on this. Reviewer patch attached, but 'needs info' on this. At the least it seems reasonable to open a new ticket to allow the above behavior - one could easily remove empty strings from the list names_list and then complain if there are none left, for instance.