#106 Syntax error with square root of rational expression

[chakravala]

Greetings, looks like one of my users (of Reduce.jl in Julia) found a bug in REDUCE:

reduce> (1/(f+t)^2)^(1/2);
Error showing value of type RExpr:
ERROR: Reduce: 

***** Syntax error: f(
+++ Error attempt to take car of an atom: 1

Link for the issue on github: https://github.com/chakravala/Reduce.jl/issues/27

[Arthur Norman]

On Tue, 1 Oct 2019, chakravala wrote:

** [bugs:#106] Syntax error with square root of rational expression**
Greetings, looks like one of my users (of Reduce.jl in Julia) found a
bug in REDUCE:
```
reduce> (1/(f+t)^2)^(1/2);
Error showing value of type RExpr:
ERROR: Reduce:

* Syntax error: f(
+++ Error attempt to take car of an atom: 1
```
Link for the issue on github:
https://github.com/chakravala/Reduce.jl/issues/27

When I try a first check using Reduce directly I see

$O/bin/redcsl -w
Reduce (Free CSL version, revision 5144), 30-Sep-19 ...

1: (1/(f+t)^2)^(1/2);

  1

---

abs(f + t)

so there are several things that we can ask you about:
(1) Exactly which release of Reduce has gone through all the various
transformation processes so as to be sued within Julia? Ie if the current
version on sourceforge built in simple ways doe snot fail is that because
something has changed since the version you use?

(2) I also tried using the Java "jlisp" with the most recent reduce.jar I
happened to have made for myself. If I understand that you just sent
"*1/(f+t)^2)^(1/2)" to the Reduce reader this takes things at least a bit
closer to you. When starting up Reduce or earlier in the session with it
what has happened? Are there flags set or changed or other things
involved?

java -jar reduce.jar -w -v
Image file defaulting to in-store data
Reduce (Free CSL version, revision 4976), 04-May-2019 20:49:43 ...
Jlisp 0.97 ... 04/05/19 20:49
Sym = 7475
Cons = 61750
String = 676

1: (1/(f+t)^2)^(1/2);

  1

---

abs(f + t)

(3) Please try the example using the names a and b rather than f and t,
since at the Lisp level T is a variable that has itself as its value and
that has been used since the early 1960s to denote "true", and to a lesser
extent sometimes f had the value nil and was used for false. That OUGHT
not to intrude on algebraic mode usage but lets rule out any issues of
that sort.

(4) Can you go "tr simpsqrt;" and indeed "tr xxx;" for many functions
within Reduce used to get to where you see this and see Reduce's trace
output? I do not know yet what would be most useful to trace! Can you go

(5) Can you find a way to reproduce the bad behaviour using the PSL or CSL
"ordinary" versions of Reduce?

(6) If all the above fail how feasible would it be to get a kit to us so
that we could both build and rebuild trick versions of Reduce with extra
print statements if we needed and them reproduce the issue you see? Note
that I have perhaps never touched Javascript and so anything that involves
it would need to be in a form that gave me very very simple push button
schemes to do things.

THose at least maybe provide s starting point, even though they do not
resolve your issue!

Arthur

[Arthur Norman]

Oh one more idea to help debug. "on backtrace;" in Reduce can lead to a
(sometimes cryptic!) indication of where things were when a low-level mess
was seen, so switching that on can help diagnose.

However the errorset function can make errors silent, so soimethimes I
hack thatimplementation of errorset so that it is as if its args 2 and 3
were always non-NIL so that backtraces do not get supressed!!

Arthur

[chakravala]

Thanks for your help, the solution was to upgrade to a newer version of REDUCE, I found that using revision 5122, 5129, and 5151 are all not erroring on this input.

I have updated the build scripts to use the updated versions of REDUCE from september now.

This issue can be closed now.