Minor changes for nvfortran

[vyu16]

Hello,

When compiling JSON-Fortran with the nvfortran compiler, I ran into the following errors:

1.

NVFORTRAN-S-0026-Unmatched quote (/home/yu/opt/json-fortran/src/json_parameters.F90: 61)

Change: Partly reverted commit 90fee68f02.

2.

NVFORTRAN-S-0188-Argument number 2 to add_to_path: type mismatch (/home/yu/opt/json-fortran/src/json_value_module.F90: 7875)

Change: Removed an optional argument that is not used anyway.

With these changes, JSON-Fortran compiles with nvfortran 21.5 and 21.7.

Feel free to close if support for nvfortran is not of interest.

Thanks!

[jacobwilliams]

note: pushed a minor change just to see if that would make the CI run.

[jacobwilliams]

So, both of these look like nvfortran compiler bugs. (can you report them to nvidia?)

The first one I'm ok with merging. But, I think the one where you remove the CK_'' string is going to break something. Looking at the code, it looks like there is a reason for these blank strings, but I need to look into it more.

[codecov-commenter]

Codecov Report

Merging #496 (1169569) into master (90893ad) will not change coverage. The diff coverage is 75.00%.

:exclamation: Current head 1169569 differs from pull request most recent head bcf8646. Consider uploading reports for the commit bcf8646 to get more accurate results

@@           Coverage Diff           @@
##           master     #496   +/-   ##
=======================================
  Coverage   89.48%   89.48%           
=======================================
  Files           6        6           
  Lines        5353     5353           
=======================================
  Hits         4790     4790           
  Misses        563      563           

[vyu16]

Thanks! I think (hope) the changes wouldn't break anything. In add_to_path, path_sep is only used in case(1_IK). I removed CK_'' from case(3_IK) and case(2_IK).

[vyu16]

Found a workaround for the backslash issue: Use the -Mbackslash flag for nvfortran, and the -qnoescape flag for IBM Fortran. According to Google, backslash was not part of the character set in Fortran until Fortran 2003. The behavior was, and apparently still is, compiler dependent.

[vyu16]

Any update on this pull request?

[vyu16]

Hi @jacobwilliams , anything I can do to get this one merged? Thanks.

[vyu16]

Update: I think the second point has been addressed by commit 9524515119e420e8dbc856031388c17f76d44a6b.

I updated the pull request to only fix the first point. As mentioned above, without the "fix", the -Mbackslash flag must be used to compile the code with nvfortran. Similarly, the -qnoescape flag is needed for xlf.