pjd.c:975:7: error: ‘ritzi’ undeclared

[Drew Parsons]

Building SLEPc 3.10.0 (real numbers) hits an error:

/build/slepc/src/pep/impls/jd/pjd.c: In function ‘PEPSolve_JD’:
/build/slepc/src/pep/impls/jd/pjd.c:975:7: error: ‘ritzi’ undeclared (first use in this function); did you mean ‘ritz’?
       ritzi = pep->eigi[idx];
       ^~~~~

ritzi is used inside a preprocessor test for PETSC_USE_COMPLEX, so it looks like you've been using SLEPc only with complex number support.

[Jose E. Roman]

This is very strange, I cannot reproduce it. This file should not be built in the case of real scalars, only with complex scalars. Yes, having code fragments with PETSC_USE_COMPLEX in this case is silly, but this is part of work in progress (see branch carmen/pjd-real). However, this should be harmless, because of the line

#requiresscalar    complex

in file src/pep/impls/jd/makefile. The gmakegen.py step at the beginning of make should remove this file from the build. Now the question is: why in your case the file src/pep/impls/jd/pjd.c is being built if scalar-type=real? Are you using the default build mode or one of the other (cmake or legacy)?

[Drew Parsons]

hmm, looks like there might be something weirdly wrong with my installation. The deb file has the complex configuration intact (petscconf.h etc), but after installation I find a real number configuration in the complex dir. My directories seems to have gotten tangled.

[Drew Parsons]

that's it. I was experimenting with saving space by sharing header files, but went too far with it. Fixed now, slepc now builds. Thanks Jose.

[Drew Parsons]

petsc installation fixed