GitLab

Reset used_ccalls and requires_file_io in initialize_code to only repeat the warning when needed in case the bytecode generator is called multiple times

More than one file as input leads to an incorrect symbol table. This is
because enter_label overwrites labels from the previously seen module,
which means the old labels (from the previously seen module) are never
printed.

This was not a problem for the old bytecode generator by John (which did
not output a symbol table), but it has been a problem ever since we added
proper relocations.

Fix global_label_string_count for multiple arguments of bcgen

See merge request !128

Setting label_module_n=-1 prevented the label name length to be counted
towards the global_label_string_count when we would find a .export
directive later on. This happened for instance when calling bcgen on
some module's ABC and _system.abc: the module refers to _cycle_in_spine
and _print_graph, these are then found in _system.abc, but the string
length was never counted, leading to a memory corruption in the bytecode
parser.

- Prevent overriding start warning when bcgen main function is called
  several times
- We must have several code_dummy versions for the types to match

Fix replace_r and update_r

See merge request !127

Rename src files included by Clean to avoid conflicts

See merge request !126

Pass -fno-addrsig to clang to avoid .addrsig and .addrsig_sym pseudo-ops that are not known to our version of GNU as yet

Avoid explicit mention of stdin, stdout, and stderr in instructions for file I/O when building with LINK_CLEAN_RUNTIME; this avoids __acrt_iob_func which has changed in recent Windows versions