Dear Feng,

Thanks for reporting this (rather severe) issue. I am attaching a patch to fix it. It also includes additional GiNaC self-check.

Best wishes,
Vladimir

PS. The previous not yet committed bugfix is attached as well.
 --
Vladimir V. Kisil                 http://www.maths.leeds.ac.uk/~kisilv/
  Book:      Geometry of Mobius Maps       https://doi.org/10.1142/p835
  Soft:      Geometry of cycles         http://moebinv.sourceforge.net/
  Jupyter notebooks:       https://github.com/vvkisil/MoebInv-notebooks

________________________________________
From: GiNaC-list <ginac-list-bounces@ginac.de> on behalf of Feng Feng <f.feng@outlook.com>
Sent: 27 July 2020 02:23
To: ginac-list@ginac.de
Subject: [GiNaC-list] 回复: issue about series involving positive symbol

I got a more compact case for this issue, by changing the in1 and in2 to

ex in1 = reader("t^2/(3*m^2-t*m^2)/2");
ex in2 = reader("-(m^(-2)*(-3+t)^(-1)*t^2/2)");

sometimes we get the output:
in1-in2 = 0
diff = 0
while, for some other times, we get the output:
in1-in2 = 0
diff = 96*log(1/48)

it seems that the sign of the log term has been changed.

Best regards!
Feng