New attribute scfinfo for microiteration data.

For the Molden writer, energies after each scf convergence step are required.

It is the TOTAL ENERGY column in RHF SCF CALCULATION section in GAMESS outputs.

 ITER EX DEM     TOTAL ENERGY        E CHANGE  DENSITY CHANGE    DIIS ERROR
   1  0  0      -74.7981539269   -74.7981539269   0.585814622   0.000000000
   2  1  0      -74.9499878493    -0.1518339224   0.180197673   0.000000000
   3  2  0      -74.9626905270    -0.0127026777   0.060203035   0.000000000
   4  3  0      -74.9640834596    -0.0013929326   0.020782027   0.000000000
   5  4  0      -74.9642853920    -0.0002019324   0.007719362   0.000000000
   6  0  0      -74.9643205282    -0.0000351362   0.005106732   0.000000000
   7  1  0      -74.9643287842    -0.0000082560   0.000126896   0.000000000
   8  2  0      -74.9643287911    -0.0000000069   0.000045747   0.000000000
   9  3  0      -74.9643287923    -0.0000000011   0.000017697   0.000000000
  10  4  0      -74.9643287925    -0.0000000002   0.000007301   0.000000000
  11  5  0      -74.9643287925    -0.0000000000   0.000003237   0.000000000
  12  6  0      -74.9643287925    -0.0000000000   0.000001437   0.000000000
  13  7  0      -74.9643287925    -0.0000000000   0.000000639   0.000000000
                       ^                ^              ^
                    Needed;      Could be used;     Parsed
                  not parsed      not parsed      as scfvalues

(from data/GAMESS/basicGAMESS-US2014/water_mp2.out). These were initially omitted from the GAMESS parser because it uses density change for convergence.

Adam and Karol have proposed parsing these into a new attribute scfinfo.

[langner]

What about turning scfvalues into a list of named tuples? For existing values, we could maintain the order so that current code doesn’t break when accessing with indices but there will now be a “key” to give it dict-like behavior where desired.

I really like that, and I’ll call it option 3. The only bad thing I can think of is that we currently don’t have any named tuples as attributes… but there needs to be a first. Makes me think whether in the long run (version 2.x) we should have custom classes for attributes. That would make many things easier.

[ATenderholt]

What about turning scfvalues into a list of named tuples? For existing values, we could maintain the order so that current code doesn’t break when accessing with indices but there will now be a “key” to give it dict-like behavior where desired.