Only generate dwarf entries if the line & column looks legitimate

[poechsel]

Compilation was failing for catala and beluga:

>> Fatal error: Uint64.of_nonnegative_int_exn: -1 is out of range
Fatal error: exception Misc.Fatal_error
Raised at Misc.fatal_errorf.(fun) in file "ocaml/utils/misc.ml", line 22, characters 14-31
Called from Dwarf_high__Dwarf_attribute_helpers.create_decl_column in file "backend/debug/dwarf/dwarf_high/dwarf_attribute_helpers.ml" (inlined), line 127, characters 15-51
Called from Dwarf_ocaml__Dwarf_concrete_instances.for_fundecl in file "backend/debug/dwarf/dwarf_ocaml/dwarf_concrete_instances.ml", line 58, characters 10-42
Called from Dwarf_ocaml__Dwarf.dwarf_for_fundecl in file "backend/debug/dwarf/dwarf_ocaml/dwarf.ml" (inlined), line 48, characters 2-85
Called from Asmgen.emit_fundecl.(fun) in file "backend/asmgen.ml", line 149, characters 10-47

It's likely that these packages are relying on a ppx that emits a location for which Locatin.is_none returns false but whose column is negative.

[lthls]

For context, there are a few other packages affected and the common factor seems to be a dependency on ppx_sedlex.

[mshinwell]

I'm not sure this is the right fix -- I was thinking we could just not emit the line and column attributes if the line number is nonsensical; and not emit the column number (but still emit the line number, unless it is nonsensical) if the column number is nonsensical.

[poechsel]

I'm not sure this is the right fix -- I was thinking we could just not emit the line and column attributes if the line number is nonsensical; and not emit the column number (but still emit the line number, unless it is nonsensical) if the column number is nonsensical.

Why not going through the same code path a if Location.is_none is true? (that is set the line and column to 0)

[mshinwell]

Hmm, that is a good question. I'm not sure we should set the numbers to zero, it's probably best to leave the attributes out, to show that we genuinely don't know the location.