Comment author: @gasche

The (startchar >= 0) test is already used to inhibit printing of column position in presence of garbage information. Maybe you could simply change the in_file function so that it also use (-1) as the line number, and extend the print_loc tests to not print negative line numbers.

There may be backward-compatibility issues with modifying Location.in_file, but I had a quick look at the various occurrences in the compiler and it looks like they would benefit of this change -- to be checked more carefully again.

Comment author: @xavierleroy

If I remember correctly, the spurious "line 1" was added so that Emacs would deign parse the error and at least open the incriminated file -- it would not do this with just a file name but no line number. Maybe Emacs's behavior changed since. But please let's avoid a regression here.

Also, I feel the patch would be less invasive if new functions print_file_location and print_file_error were introduced instead of this pesky optional argument.

Comment author: rp

New, less-intrusive patch.

The omission of (dummy) line number is only for error-enabled warnings. The (correct) location information is printed (and parsed by emacs) with the warning message.

Comment author: @mshinwell

@rp Can you please submit a Github pull request if you would like to move this forward?

Comment author: @damiendoligez

This will very probably be obsoleted by #948