Compiling with Cray Fortran 14 about toml-f

[awvwgk]

This seems to fail in the tomlf_ser module, which provides the implementation to turn a data structure into a string.

You could try to comment out part of the visitor implementation to see which routine exactly breaks the compiler (they will probably appreciate a minimal reproducer for this).

The module is pretty separated from the rest of TOML Fortran, you could remove it for the time being to see whether Cray can pass the unit and decoder tests or there are more issues.

[awvwgk]

I don't have access to Cray compilers for checking this unfortunately. The error message however looks like an internal compiler error rather than an issue with TOML Fortran. Maybe it is possible to reduce the tomlf_diagnostic module to isolate the exact problem. This way it is easier to find a workaround or/and send a support request to Cray.

[awvwgk]

Looks like it is caused by this lines:

toml-f/src/tomlf/diagnostic.f90

Lines 253 to 255 in 5a19921

	associate(first => token%first)
	shift = first(line) - 1
	end associate

[timofeymukha]

Trying to understand the code a bit. We have

   type :: line_token
      integer :: first, last
   end type line_token

Then we have in failing function

type(line_token), allocatable :: token(:)
! ...
associate(first => token%first)
      shift = first(line) - 1
end associate

The code is a bit unclear to me, token is an array, so is the idea here that token%first will be a corresponding array of all first attributes? Does it work that way? Can we do

associate(first => token%first(line))
      shift = first - 1
end associate

instead?

[awvwgk]

See #137 for an explanation:

token is an array of toml_token type, token%first provides an array of integers from all members first in the types, with associate a reference first to this (strided) array is created, finally the element line of this array is selected. Seems super complicated for just fetching an integer value, previously I was doing more with this array but removed most of it now.

I still think Cray should handle this correctly, but it is easier to remove the usage here since it is not needed in this context anymore.

[timofeymukha]

The module compiles now, but unfortunately there is more trouble down the road with

FAILED: CMakeFiles/toml-f-lib.dir/src/tomlf/ser.f90.o include/tomlf_ser.mod
/opt/cray/pe/craype/2.7.16/bin/ftn -I/cfs/klemming/projects/snic/kthmech/tmu/Software/toml-f/src/tomlf -I/cfs/klemming/projects/snic/kthmech/tmu/Software/toml-f/_build/include -em -Jinclude -h PIC -c CMakeFiles/toml-f-lib.dir/src/tomlf/ser.f90-pp.f90 -o CMakeFiles/toml-f-lib.dir/src/tomlf/ser.f90.o
ftn-2116 ftn: INTERNAL
  "/opt/cray/pe/cce/14.0.1/cce/x86_64/bin/ftnfe" was terminated due to receipt of signal 013:  Segmentation fault.

This does not seem helpful at all :(. Perhaps one can engage Cray somehow to look at this.

[timofeymukha]

By making the ser.f90 empty and removing the "compliance" subdir from the tests, I could compile.
Unfortunately, the unit tests fail

  Starting lexer-from-sequential ... (108/108)
       ... lexer-from-sequential [FAILED]
  Message:
# Testing: parser
  Starting root-invalid ... (1/39)
       ... root-invalid [EXPECTED FAIL]
  Message: error: Invalid syntax
 --> mocked:1:1
  |
1 | }
  | ^ unexpected opening brace
  |
  Starting table-body ... (2/39)
       ... table-body [PASSED]
  Starting table-header-long ... (3/39)
Segmentation fault (core dumped)

Sad times.

[awvwgk]

Can we get a better backtrace for the segmention fault? If it fails in the table header then there might be an issue with the way the internal stack for the TOML parser is handled.

Also not happy to see a failure in the lexing... if you can give me more details I might figure out a workaround, but ultimately these are bugs in the Cray compiler which should be reported and fixed.

[timofeymukha]

I will also try with version 15 of the compiler. I guess we can't report bugs on an older version anyway.