Project

General

Profile

Bug #1794

Strange bug: demo-Osaka2015

Added by John Abbott about 1 month ago. Updated 8 days ago.

Status:
Resolved
Priority:
Normal
Assignee:
-
Category:
bug
Target version:
CoCoA-5.4.2
Start date:
16 Mar 2024
Due date:
% Done:

70%

Estimated time:
Spent time:
1.00 h

Description

I observe differing output with the test demo-Osaka2015 when compiled using clang and with debugging active.
The problem seems not to arise with g++. Platform is Linux (Ubuntu 23.04).

History

#1 Updated by John Abbott about 1 month ago

My g++ reports version info gcc version 11.4.0 (Ubuntu 11.4.0-1ubuntu1~22.04)
My clang reports version info Ubuntu clang version 14.0.0-1ubuntu1.1.

Is this bug reproducible on other platforms?

My current guess is that a random number generator behaves differently... but I've not yet looked at the code!

#2 Updated by John Abbott about 1 month ago

  • Status changed from New to In Progress
  • % Done changed from 0 to 10

The cause is not too surprising: demo-Osaka2015.cocoa5 uses a random example (on line 72).
Evidently, with debugging active, some random numbers are used internally, and this affects the publicly visible sequence.

BTW I have verified that the problem exists also when compiling with g++ in debug mode. Curiously, the .cocoa5 files contains a comment saying what the random values are... but the comment is not the same as what is actually generated...

Ideally we would replace the randomness by specific "random" numbers.

#3 Updated by Anna Maria Bigatti about 1 month ago

John Abbott wrote:

The cause is not too surprising: demo-Osaka2015.cocoa5 uses a random example (on line 72).
Evidently, with debugging active, some random numbers are used internally, and this affects the publicly visible sequence.

BTW I have verified that the problem exists also when compiling with g++ in debug mode. Curiously, the .cocoa5 files contains a comment saying what the random values are... but the comment is not the same as what is actually generated...

For some reason (I cannot remember, but I think I knew it when it first happened, recently) the random values I'm getting now differ from the original version of the test (in the comment). I didn't think it depended on clang, so I changed the expected output.

Ideally we would replace the randomness by specific "random" numbers.

OK, I agree, especially if that changes with debugging on/off. Should I do it?

#4 Updated by John Abbott 8 days ago

  • Status changed from In Progress to Resolved
  • % Done changed from 10 to 70

random is called harmlessly in exbugs.cocoa5.
We should check ErrMesg.cocoa5 and test-ApproxSolve.cocoa5; it is not clear to me that random is used harmlessly in these files!!

Also available in: Atom PDF