de-parse of CASE block containing non-ascii char fails under Python 2

[arporter]

Again, the non-ASCII char is in a string:

     CASE(  30  )       !==  fixed 3D shape  ==!
        IF(lwp) WRITE(numout,*) '   ==>>>   eddy viscosity = F( latitude, longitude, depth )'
        IF(lwp) WRITE(numout,*) '           maximum reachable coefficient (at the Equator) = ', zah_max, cl_Units, '  for e1=1°)'

[arporter]

This case fails at line 5571 of Fortran2003.py:

def tofortran(self, tab='', isfix=None):
    tmp = []
    start = self.content[0]
    end = self.content[-1]
    tmp.append(start.tofortran(tab=tab, isfix=isfix))
    for item in self.content[1:-1]:
        if isinstance(item, Case_Stmt):
            tmp.append(item.tofortran(tab=tab, isfix=isfix))
        else:
            tmp.append(item.tofortran(tab=tab + '  ', isfix=isfix))
    tmp.append(end.tofortran(tab=tab, isfix=isfix))
    return '\n'.join(tmp)

[arporter]

This is similar to the problem fixed in #217

[arporter]

BlockBase.tofortran() has the same problem where it does a "\n".join(tmp). (Line 653 of utils.py.)

[arporter]

Strangely, I can only reproduce these errors in a pytest test when I use the FortranFileReader with ignore_comments=False. If I set it to True then all is fine.

[arporter]

With comments, the list of strings being joined contains some (the comments) that are unicode while the code itself is just str. Without comments, the list of things being joined does not contain any unicode strings and thus (presumably) no conversion is attempted.

[arporter]

The original problem is exercised if there is a comment *inside" a select case block. The problem in BlockBase is exercised if there is a comment just in the program body. I've added two tests to illustrate these.

[arporter]

I started down the road of making sure everything was unicode but that got rapidly out of hand and, theoretically, Python 2 end-of-life is only a month away. I've therefore gone for the minimal fix of ensuring that all strings are consistently encoded before we do a join on them (when producing Fortran).

[rupertford]

PR #227 has been merged to master. Closing this issue.