[Gnu-arch-users] [REJECT: address@hidden (default naming convention changes)

[Tom Lord]

The gist of the proposed changes to the default naming conventions
are:

      case arch_inventory_backup:
        {
 -        return str_save (0, 
"^.*(~|\\.~[0-9]+~|\\.bak|\\.swp|\\.orig|\\.rej|\\.original|\\.modified|\\.reject)$");
 +        return str_save (0, 
"^.*(~|\\.~[0-9]+~|\\.bak|\\.swp|\\.orig|\\.original|\\.modified|\\.reject)$");
        }
  
      case arch_inventory_junk:
        {
 -        return str_save (0, "^(,.*)$");
 +        return str_save (0, 
"^(,.*)|(.*\\.(o|a|so|core|so(\\.[[:digit:]]+)*)|core)$");
        }
  
      case arch_inventory_unrecognized:
        {
 -        return str_save (0, 
"^(.*\\.(o|a|so|core|so(\\.[[:digit:]]+)*)|core)$");
 +        return str_save (0, "^.*(\\.rej)$");
        }
 



The problems include:

1) .rej is in the same conceptual class as .orig, .original,
   .modified, and .reject --- they should all be handled the same way.


2) .o, .a, .so, .core, and .so* do not belong in junk.
   Those files can be very, very valuable _and_ difficult or
   impossible to reproduce.   "junk", meanwhile, is comprised of files
   that may be casually deleted.

3) .rej does not belong in unrecognized.   For example, suppose I am
   going to replay a few different patches (as I did when
   cherry-picking for this merge before deciding not to commit it).
   I may get conflicts in an early one but keep going anyway.
   Putting .rej in unrecognized would cause problems.

The rationale for these changes was:

    This way someone can use tla out of the box exactly like how
    cvs/svn/whatever are used for typical projects.

which is not enough justification to overcome the problems mentioned
above.

-t