Improve manual ports of parson, ptrdist, olden about analyze-conversion HOT 4 OPEN

I'm starting to look into this using bh as an example (though I may soon check the other example programs to see whether I can get more improvement for the work there).

So it's legitimate to change orig to reduce diffs for the purpose of our experiment? Changing whitespace and splitting multi-decls is OK? What about adding parameter types to function declarations? That's required by Checked C and not required by plain C; I think there's consensus that newly written C code should have parameter types, but some of the codebases to which we envision real users applying 3C might be old enough not to have them, so I'm unsure whether adding them to orig to reduce diffs makes our results more or less valid with respect to the target usage of 3C. Can any more general statement be made about what kinds of changes to orig are legitimate?

from analyze-conversion.

Another example from git diff --ignore-all-space --no-index src/Olden/bh/{orig,revert}/newbh.c:

   if (Type(n) == BODY) {
-    Next((bodyptr) n) = bp_free_list;
-    bp_free_list = (bodyptr) n;
+    bodyptr p = 0;
+     { p = (bodyptr)n; }
+    Next(p) = bp_free_list;
+    bp_free_list = p;
   }

Here the extra local variable and block were introduced so that the manual port could add an _Unchecked annotation on the block and an _Assume_bounds_cast inside it. IIUC, currently the diff above is being counted as a refactoring. Is that what we want, or should we apply the change to orig so that this (arguably artificial and formulaic) change is counted as neither a refactoring nor an annotation? Or does it not matter because we'll rely more on the semantic statistics from 3C than on the line diffs?

from analyze-conversion.

So it's legitimate to change orig to reduce diffs for the purpose of our experiment? Changing whitespace and splitting multi-decls is OK? What about adding parameter types to function declarations?

Changing whitespace and splitting multidecls is OK.

The rest, I would say, should be left alone. Part of the porting process is to add types to parameters that don't have them. So we should not change orig to do that.

OTOH, it's also OK to change manual so that it does a better job of porting without making diffs that aren't needed. The bzero and #undef comment I made in the issue description is an example.

from analyze-conversion.

should we apply the change to orig so that this (arguably artificial and formulaic) change is counted as neither a refactoring nor an annotation? Or does it not matter because we'll rely more on the semantic statistics from 3C than on the line diffs?

It should count as a refactoring (in the line diffs) because it was required by Checked C for the program to typecheck. So we should not change it in orig. At the same time, it won't play into the semantic stats, as you've pointed out.

from analyze-conversion.