Project

General

Profile

Bug #275

Unhelpful error messages when SmallExponent_t is unsigned char.

Added by John Abbott over 11 years ago. Updated over 6 years ago.

Status:
Closed
Priority:
Normal
Assignee:
John Abbott
Category:
enhancing/improving
Target version:
CoCoA-5.2.4
Start date:
15 Nov 2012
Due date:
% Done:

100%

Estimated time:
3.80 h
Spent time:
3.50 h

Description

I changed config.H so that the type SmallExponent_t is unsigned char (in an attempt to tackle Enrico Carlini's computation). Two CoCoA-5 tests failed: whatiscocoa.cocoa5 and SourceAnna.cocoa5. In both cases the error produced is Unable to convert value.

Investigate why these examples need exponents bigger than 255.

Related issues

Related to CoCoALib - Design #268: Exponent range (in power products)Closed2012-10-18

Related to CoCoALib - Feature #269: PPMonoids: check for exponent overflow in power functionClosed2012-10-18

Related to CoCoALib - Design #683: Module index component in internal compressed representationClosed2015-04-14

History

#1 Updated by Anna Maria Bigatti about 10 years ago

  • Target version set to CoCoA-5.1.0 Easter14

#2 Updated by John Abbott about 10 years ago

  • Target version changed from CoCoA-5.1.0 Easter14 to CoCoA-5.?.?

#3 Updated by John Abbott over 9 years ago

  • Category set to enhancing/improving
  • Status changed from New to In Progress
  • Assignee set to John Abbott
  • % Done changed from 0 to 50

I have just retried the experiment. Several CoCoALib tests fail:

*****  The following CoCoALib tests failed, please tell us about it.
*****   test-GOperations1 test-GOperations2 test-QuotientRing1 test-SparsePolyRing1 test-SqFreeFactor1 test-TmpMorseGraph

In each case the error produced was CoCoA::ERR::ExpTooBig which I think is quite clear.

I compiled CoCoA-5 anyway and 3 tests failed:

*****  The following CoCoA-5 tests failed, please tell us about it.
*****   whatiscocoa.cocoa5 radical.cocoa5 SourceAnna.cocoa5

Again all the errors were "Exponent too big" (or a direct consequence of such an error). These messages are quite satisfactory.

Is it OK for the tests to require such high exponents? I find it a little surprising that so many tests require exponents over 255... perhaps we should take a look?

Addendum also ex-module fails with ExpTooBig, but I find that hard to believe given how simple it is -- this should be checked!!

#4 Updated by John Abbott over 9 years ago

I have just (20140902) checked using typedef unsigned short SmallExponent_t; and everything worked fine.

Reminder: we should still investigate why so many tests (& 1 example) fail with unsigned char! DONE see comment 6 below!

#5 Updated by John Abbott about 9 years ago

I have again encountered the problems (failing tests) with using unsigned char for exponents. I wanted to use very compact exponents because I wanted to compute with polynomials with >10000000 terms in many variables with low degrees (< 10), so it made sense to put exponents in unsigned char (the most compact representation CoCoALib offers).

After a little debugging I found that the problem was an exponent of 10000 in one example; and this high degree came from the "trick" used to represent module elements as polynomials with an extra "module component indet". The module was used to compute GCD between two "multivariate" polys over a finite field.

Hint: GBEnv.H:141 definition of compile time constant myMaxComponentIndex

#6 Updated by John Abbott about 9 years ago

I have just tried all tests with SmallExponent_t set to unsigned char, and myMaxComponentIndex set to 100 (instead of the default 10000). All CoCoALib tests passed. Just 1 CoCoA-5 test failed: radical -- not sure why. All CoCoALib examples ran fine too (except ex-SmallFp1, but that is an unrelated problem).

Note radical.cocoa5 produces an answer with generators of degree 30; according to the debugger at some point it did try to multiply together two polynomials with high degrees (23 and 240ish, if I recall correctly)

#7 Updated by John Abbott almost 9 years ago

  • Target version changed from CoCoA-5.?.? to CoCoA-5.1.3/4 Jan 2016

#8 Updated by John Abbott about 8 years ago

  • Target version changed from CoCoA-5.1.3/4 Jan 2016 to CoCoA-5.?.?

#9 Updated by John Abbott over 6 years ago

  • Status changed from In Progress to Feedback
  • % Done changed from 50 to 90
  • Estimated time set to 2.90 h

After changing some "ugly" code for finding which indets are actually in a poly (in radical.cpkg5 and in BringIn.cpkg5), all CoCoA-5 tests run fine. All CoCoALib tests run fine. All CoCoALib examples run fine.

Just one strange observation: I thought powering did a quick check for exp overflow, but it seems not if the power is 2

J := x^128;
J^2;  --> wrong answer with no warning!
J := x^86;
J^3;  --> error about exponent overflow

UPDATE fixed the problem with squaring (actually there were two bugs)

#10 Updated by John Abbott over 6 years ago

  • Status changed from Feedback to Closed
  • Target version changed from CoCoA-5.?.? to CoCoA-5.2.4
  • % Done changed from 90 to 100
  • Estimated time changed from 2.90 h to 3.80 h

I think I have solved all problems

Also available in: Atom PDF