Re: yet another DLD question

[Andy+Alison Adler]

John W. Eaton wrote:
> On  4-Nov-1998, Andy+Alison Adler <address@hidden> wrote:
> | I've got this bizarre bug:
> |
> | if I put my code
> |
> | [SNIP]
> |    Octave_map As;
> | [SNIP]
> |    ColumnVector coefV=  As["coef"] . vector_value();
> |    double * coef =  coefV . fortran_vec();
> |    for ( int i= 0; i<  nnz; i++ )
> |       printf("a= %f\n", *(coef+i));
> | [SNIP]
> |
> | then it works fine, giving output
> |
> | a= 1.000000
> | a= 2.000000
> | a= 3.000000
> | etc.
> |
> |
> | However, if I try to abbreviate by putting
> |
> | [SNIP]
> |    double * coef =  As["coef"] . vector_value() . fortran_vec();
> |    for ( int i= 0; i<  nnz; i++ )
> |       printf("a= %f\n", *(coef+i));
> | [SNIP]
> |
> | then the first value printed is always wrong
> | a= 39.399666
> | a=  1.000000
> | a=  2.000000
> |
> | (the first value is often 39.??? but is sometimes different)
> |
> | MY QUESTION IS:
> |
> | is As["coef"] . vector_value() . fortran_vec() illegal?
> | or is this a bug?
> | or am I doing something wrong
> 
> I think either approach should work.  I created a simple test example
> and it worked fine.  Perhaps there is another problem in your code
> that causes a memory overwrite.  Can you please send me a complete
> example that demonstrates the problem?

I've gone over my code with a fine tooth(ed?) comb, and
I'm finding that for both

double * x= A["blah"] . vector_value() . data();
double * x= A["blah"] . vector_value() . fortran_vec();

x[0] is often corrupted. Sometimes this code segfaults, also.

I think that vector_value() creates a temprorary copy and
gives the pointer to data(), and then this copy is freed,
corrupting the first element, and invalidating references
to the memory.

Does this make sense?

andy