#2936 stack overflow in integrate

[Robert Dodier]

Maxima 5.36.0 + Clisp. I don't have a smaller example. EDIT: smaller example in comment below.

(%i1) foo : exp(-(log(x) - MU)*(log(x) - MU)/(2*SIGMA*SIGMA))/(x*SIGMA*sqrt(2*%pi)) $
(%i2) bar : (log(B) - log(x*SIGMA) + ((x-A)*(x-A)/(2*B*B) - (log(x) -MU)*(log(x) -MU)/(2*SIGMA*SIGMA))) $
(%i3) assume (SIGMA>0, x>A, B>0) $
(%i4) foobar : expand (foo*bar) $
(%i5) integrate (foobar, x, A, inf);
Is A positive, negative or zero?

p;
Is A - 1 positive, negative or zero?

p;
     MU
Is %e   - A positive, negative or zero?

p;

*** - Program stack overflow. RESET

A stack trace obtained with SBCL seems to show (among others) SIMP-GAMMA-INCOMPLETE, MEQP, ZEROP1, and SRATSIMP on the call stack. Not sure who called whom.

[Robert Dodier]

Simpler:

integrate (%e^((-log(x)^2)-1)*log(x),x);
trigsimp (%);

triggers the stack overflow.

In the original example, integrate calls DINTLOG calls ANTIDERIV and then INTSUBS calls trigsimp on the antiderivative.

trigsimp ((%i*gamma_incomplete(1,(1-2*log(x))^2/4)*(1-2*log(x))^2)
           /(2*log(x)-1)^2);

is enough to cause a stack overflow.

Even just

trigsimp (gamma_incomplete(1,log(x)));

is enough.

[Robert Dodier]

Problem was caused by trigsimp formal argument x appearing in input expression. Renaming x to x% works around it; lexical scope is a better solution. Fixed by commit [1780b2f], Closing this report.