Incorrect call to SLASET in schkst2stg.f

[serguei-patchkovskii]

This appears to be related to #318 https://github.com/Reference-LAPACK/lapack/issues/318. The problem is not in SLASET but in the calling routine.

TESTING/EIG/schkst2stg.f contains the following code fragment (both in 3.9.0 and in the repo):

1000 * using the 1-stage.
1001 *
1002 CALL SLASET( 'Full', N, 1, ZERO, ZERO, SD, 1 )
1003 CALL SLASET( 'Full', N, 1, ZERO, ZERO, SE, 1 )
1004 CALL SLACPY( "U", N, N, A, LDA, V, LDU )
1005 LH = MAX(1, 4*N)
1006 LW = LWORK - LH

This code is incorrect, and causes a bounds violation in SLASET. The correct code is:

1002 CALL SLASET( 'Full', N, 1, ZERO, ZERO, SD, N )
1003 CALL SLASET( 'Full', N, 1, ZERO, ZERO, SE, N )

[serguei-patchkovskii]

The same issue is present in TESTING/EIG/schksb2stg.f, around the line 673:

671 * using the SSBTRD.
672 *
673 CALL SLASET( 'Full', N, 1, ZERO, ZERO, SD, 1 )
674 CALL SLASET( 'Full', N, 1, ZERO, ZERO, SE, 1 )
675 CALL SLACPY( ' ', K+1, N, A, LDA, U, LDU )
676 LH = MAX(1, 4*N)
677 LW = LWORK - LH

The fix is the same.

[ereisch]

Actually, the fix isn't entirely correct either, since SLASET is expecting a single character for the first argument, and you're instead passing in a string.

[serguei-patchkovskii]

Actually, the fix isn't entirely correct either, since SLASET is expecting a single character for the first argument, and you're instead passing in a string.

As a matter of Fortran pedantry, Fortran association rules explicitly allow passing a longer string of the default character kind, where the dummy argument is a shorter default-kind character type. This situation is covered by the clause 4 of the section 12.5.2.4 of Fortran 2008, ISO/IEC 1539:2010. (you can find an unofficial, believed accurate, copy of the standard at https://wg5-fortran.org/f2008.html).

Because 'Full' has type CHARACTER(LEN=4), it can be associated to a CHARACTER dummy argument in SLASET, and the code is standard-conformant.

Not that it really matters in the grand scheme of things.