fixes #20139; hash types based on its path relative to its package path

[ringabout]

fixes #20139

hashTypeSym is used to calculate the hash of a type symbol. To differentiate modules with the same name but in the different paths of a package, we can consider its relative path to its package path.

c &= relativePath(conf.toFullPath(s.info), conf.toFullPath(getPackageSymbol(s).info), '/')

[Araq]

Seems excessive, isn't there an easier way?

[quantimnot]

@ringabout How about including ast.hash(ast.ItemId) into the symbol hashes? I'm not sure of negative consequences, but it seems to work for me. I haven't run the full test suite though.

Here is what I tried:

+template includeIdHash(c: var MD5Context, s: PSym) =
+  # Include the symbol ID to disambiguate symbols with identical names and
+  # identical module names. (bug #20139)
+  c &= hash(s.itemId)

proc hashSym(c: var MD5Context, s: PSym) =
+  includeIdHash(c, s)
  if sfAnon in s.flags or s.kind == skGenericParam:
    c &= ":anon"
  else:
    var it = s
    while it != nil:
      c &= it.name.s
      c &= "."
      it = it.owner

proc hashTypeSym(c: var MD5Context, s: PSym) =
+  includeIdHash(c, s)
  if sfAnon in s.flags or s.kind == skGenericParam:
    c &= ":anon"
  else:
    var it = s
    while it != nil:
      if sfFromGeneric in it.flags and it.kind in routineKinds and
          it.typ != nil:
        hashType c, it.typ, {CoProc}
      c &= it.name.s
      c &= "."
      it = it.owner

[ringabout]

I'm not sure whether it is consistent enough, see also #20564 (comment)

[ringabout]

@Araq Could we use what @quantimnot suggested or is there another way out?

[Araq]

My own "superior" solution ended up being nearly identical... :-)

[arnetheduck]

a backport to 1.6 would be appreciated here - this completely cripples 1.6 both compile-time and runtime depending on the shape of objects

[Araq]

"path relative to its package path" is still wrong or at least inflexible. --> Wait for my change on top of this PR.

[github-actions]

Thanks for your hard work on this PR!
The lines below are statistics of the Nim compiler built from 38d299d

Hint: mm: orc; opt: speed; options: -d:release
166278 lines; 8.249s; 612.445MiB peakmem