From: Jens Vollinga <vollinga@thep.physik.uni-mainz.de>
Date: Wed, 12 Oct 2005 18:05:03 +0000 (+0000)
Subject: Changed the conditions for the application of the (1-x)/(1+x) transformation,
X-Git-Tag: release_1-4-0~147
X-Git-Url: https://www.ginac.de/ginac.git//ginac.git?p=ginac.git;a=commitdiff_plain;h=270969d036bd27a8454442501f8eb241fa66c9b2

Changed the conditions for the application of the (1-x)/(1+x) transformation,
which caused an endless recursion for certain arguments around +-I.
---

diff --git a/ginac/inifcns_nstdsums.cpp b/ginac/inifcns_nstdsums.cpp
index 3920bf9e..88f0fa4e 100644
--- a/ginac/inifcns_nstdsums.cpp
+++ b/ginac/inifcns_nstdsums.cpp
@@ -2839,10 +2839,16 @@ static ex H_evalf(const ex& x1, const ex& x2)
 			}
 		}
 
-		// since the transformations produce a lot of terms, they are only efficient for
-		// argument near one.
-		// no transformation needed -> do summation
-		if (cln::abs(x) < 0.95) {
+		// check for the applicability of transformations
+		// 
+		// first condition: since the transformations produce a lot of terms,
+		//   they are only efficient for argument near the boundary |x| = 1, then no
+		//   transformation is needed
+		// second condition: veto for region around +-I to avoid endless recursion
+		//   with the (1-x)/(1+x) transformation. 1.198 is sqrt(1.4142) is the
+		//   boundary of the problematic transformation.
+		//   
+		if (cln::abs(x) < 0.95 || (cln::abs(x) < 1 && cln::abs(x-1) >= 1.198)) {
 			lst m_lst;
 			lst s_lst;
 			ex pf;