Parameter doc files are inconsistently formatted with PGI

[adcroft]

gnu and intel produce consistent output but the PGI compiler has started leading to diffs, e.g.:

diff --git a/ice_ocean_SIS2/Baltic/SIS_parameter_doc.all b/ice_ocean_SIS2/Baltic/SIS_parameter_doc.all
index 54c7c2e..e612345 100644
--- a/ice_ocean_SIS2/Baltic/SIS_parameter_doc.all
+++ b/ice_ocean_SIS2/Baltic/SIS_parameter_doc.all
@@ -265,15 +265,13 @@ ICE_THICKNESS_INIT_CONFIG = "data_override" ! default = "data_override"
                                 ! initialized for a new run:
                                 !     file - read sea ice concentrations from a specified file
                                 !     data_override - use the data_override capability or zero everywhere
-                                !     uniform - sea ice has uniform thickness where the concentration is
-                                !       nonzero.
+                                !     uniform - sea ice has uniform thickness where the concentration is nonzero.
 SNOW_THICKNESS_INIT_CONFIG = "data_override" ! default = "data_override"
                                 ! A string that determines how the initial total sea ice concentration is
                                 ! initialized for a new run:
                                 !     file - read sea ice concentrations from a specified file
                                 !     data_override - use the data_override capability or zero everywhere
-                                !     uniform - sea ice has uniform thickness where the concentration is
-                                !       nonzero.
+                                !     uniform - sea ice has uniform thickness where the concentration is nonzero.
 ICE_SALINITY_INIT_CONFIG = "uniform" ! default = "uniform"
                                 ! A string that determines how the sea ice salinity is initialized for a new
                                 ! run:

I have not had a chance to figure when the divergence happened.

[marshallward]

Lots of \t and \n in these line length calculations, and AFAIK it's nonstandard to convert those into ASCII codes. (Gfortran only does it if you enable certain flags).

[Hallberg-NOAA]

The problem seems to be that the len_trim intrinsic function gives different lengths for on tabs on different compilers. We are taking this into account when we replace tabs with a pair of spaces before writing messages in writeMessageAndDesc, at line 526 in MOM_document.F90. But we are not taking this into account when we determine where to automatically set new-lines on lines 502 and 506 of the same file.

For SIS2, this issue of non-reproducing documentation across compiler is being corrected in a PR that will come shortly by explicitly replacing the tabs in the offending parameter descriptions with two spaces, but there should also be modifications made in MOM_document.F90.

[Hallberg-NOAA]

This issue has been addressed by https://github.com/NOAA-GFDL/MOM6/pull/1124.