[ferret_users] Compilation of pyferret with gcc > 4.8

[Martin Schmidt <martin.schmidt@xxxxxxxxxxxxxxxxx>]

Dear ferret developers!

ferret (version 731) incorporates many other libraries under rapid 
development and this motivates to compile pyferret with a gnu compiler 
version > 4.8.

All test  are carried out on Suse Linux os 42.3 but on machines with 
different processors. .

With gcc 4.8.5, i.e. the systems default compiler I get running code. It 
is compiled on one machine and works on others with the same system but 
different processor too. But compiling hdf5 (1.10.1) and netcdf (4.6.0) 
warnings show up that the compiler does not fully support some c99 
standard, which motivates to use a more recent gcc version. I have 
checked gcc-5 and gcc-7 and the related gfortran available with opensude 
42.3.

Compiling .  with  gcc 5.3.1, ferret fails for example with

yes? go multi_view 1, 2, 0.38, 0.07, 0.1, 0.8, 0.15, 0.02, "V", 0.05,0.05
 **ERROR: invalid subcommand: mv_r    = 1
defiymbol mv_r    = 1
Command file, command group, or REPEAT execution aborted

One may write simpler scripts with similar errors - just the script 
parsing does not work properly. Images can be produced and many ferret 
elements work as they should.

Remarkably, on another machine this binary works well. The installed 
system packages "should" be the same on all machines and I could not 
identify a significant difference in the results of ldd applied on the 
pyferret-library.. It does not matter, on which machine the code is 
compiled, it runs on the some distinct machines and on others it does not.

Compiling with gcc-7 the binary fails on all machines when started. The 
error comes from

PyFerret-master/fmt/src/tm_fmt

around line 300. The code in question is

       sig_needed = sig_needed + need_adjust
       nright = MIN( sig_needed-pow10, slen-(nleft+1) )
       str_len = nleft + nright + 1
       WRITE ( fmt, 3200 ) str_len, nright
 3200       FORMAT ( '(F',I2,'.',I2,')' )
       WRITE ( outstring, fmt ) val

 "fmt" contains only "(F" and nothing else, so the last write fails. 
With the "print-debugger" I have checked that  str_len and nright are 
filled with correct values, even writing to stdout instead of fmt 
reveals a resonable format string. Introducing more prints and trying to 
introduce intermediate variables gives strange results like for a 
corrupted memory.

hdf5/netcdf are build with the same compiler like ferret is. Many other 
libraries (cairo, pangoo, python) come from the system.

My questions:

- did anybody compile pyferret with gcc > 4.8?

- any idea what could be wrong?

Best,

Martin


--
Dr. Martin Schmidt/PyFerret-master/fmt/src/tm_fmt
Leibniz-Institute for Baltic Sea Research
Seestrasse 15
D18119 Rostock
Germany