token logic error? about f90nml

[marshallward]

Sure, that looks like a bug, I will try to look into it. Thanks for the report.

Things are quite busy for me workwise at the moment, and then I go on one-month leave, so it could potentially take quite a long time. In the meantime, you could try putting string delimiters around it ('%'), which seems to work for me.

[greyltc]

Thanks for the quick response.
Yep, that's exactly what I'm doing for now.
I'm just a bit worried about other token combos biting me later!

[marshallward]

Following up on here.. (mostly for myself)

Problem appears to be L357 of parse_variable. We rely on the appearance of those tokens (%, (, =) to identify the end of the current variable and the beginning of the next variable. (This is primarily for when multiple variables, including vectors, are on a single line.)

This may be failing because it sees the % and thinks that it's finished, and then sets the value of units to None.

This will probably be a bit harder to fix. For example, this namelist:

&nml x=y%z=c /

could have two interpretations:

&nml
    x = 'y%'
    z = 'c'
/

&nml
    x = 
    y%z = 'c'
/

But there might be a way out of this particular situation. Your example only has one interpretation, since % would need a variable name to be interpreted as a derived type dereference.

[marshallward]

Actually I could be wrong about that example... spaces may be able to resolve the case that I mentioned, I'd need to check it out.

[marshallward]

Sorry for the long delay.

I've testd this case in gfortran and Intel fortran, and the results are similar to #43 in that gfortran rejects the namelist, and Intel fortran parses it differently (specifically it sets units to an empty string).

Given that these non-delimited tokens don't appear to produce the strings that you need, I'm probably not likely to support these.

From what you mentioned in #43, it may be that SDDS is producing very old namelist files that don't produce meaningful output on any actively maintained compiler, so I'm inclined to close these issues as not relevant. It might be more constructive to lean on SDDS to produce syntatically correct namelists rather than supporting incorrect namelists.

I could possibly support this case with a control flag or something that relaxes the parsing rules, but these are very hard edge cases to support, and I don't think I would be able to prioritise them.

[marshallward]

I'm going to close this ticket, due to the lack of support for these values in gfortran and Intel fortran, the only compilers available to me. If you can find a compiler which supports these values, or just feel like this is an urgent and unsolvable problem for you, then feel free to re-open this and we can talk about a plan. But I probably won't personally be able to devote much time to this one, sorry.

[marshallward]

Reopening, but explicitly noting this is an SDDS issue.

[marshallward]

As in #43 this is probably too big a challenge to implement unless its strongly driven by user feedback, so will close for now. Please re-open in the future if necessary.