[bug #66026] gsl_linalg_LU_decomp using uninitialized memory

[Andrew Davison]

Follow-up Comment #3, bug #66026 (group gsl):

Here's what my user who found this problem, says:

fyi
More info for if those gsl guys start a discussion:

1.) ./a.out 128 128 or 256 256 or 512 512 or 1024 1024 does not crash,
    so you don't always get the fault.

2.) The problem might be the status of a singular matrix:

status = LU_decomp_L3 (&AL.matrix, ipiv);

If not singular then status = 0 and all fields of vector ipiv are valid(have
been processed).

if singular then status = number of fields which have been processed.

For example: singular 127x127 status = 16
means that this is wrong:
for (i = 0; i < minMN; ++i) {  # the segfault area

Better is this:
size_t damage = minMN;
if (status) damage = status; 
for (i = 0; i < damage; ++i) {
}  

Even better (perhaps)
if (status)  { ; /* singular, do nothing */
} else { 
   for (i = 0; i < minMN; ++i) { 
   : 
}
gsl_vector_uint_free(ipiv);
return status;

Performance back: no calloc and in case of singular dismiss signum as
valueless.



    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?66026>

_______________________________________________
Message sent via Savannah
https://savannah.gnu.org/

signature.asc
Description: PGP signature