Re: [RFA] Extend hashed symbol dictionaries to work with Ada

[Paul Hilfinger <HIlfinger@cs.berkeley.edu>]

On Wed, 2010-10-06 at 16:59 -0700, Doug Evans wrote:

> Hi.  I wouldn't mind having a couple of comments added to this function:
> 

Arghh.  My mistake.  I added a comment to our local version without
submitting it.  Updating:

+/* Produce an unsigned hash value from STRING0 that is consistent
+   with strcmp_iw, strcmp, and, at least on Ada symbols, wild_match.
+   That is, two identifiers equivalent according to any of those three
+   comparison operators hash to the same value.  */
+
+static unsigned int
+dict_hash (const char *string)
+{
+  /* The Ada-encoded version of a name P1.P2...Pn has either the form
+     P1__P2__...Pn<suffix> or _ada_P1__P2__...Pn<suffix> (where the Pi
+     are lower-cased identifiers).  The <suffix> (which can be empty)
+     encodes additional information about the denoted entity.  This
+     routine hashes such names to msymbol_hash_iw(Pn).  It actually
+     does this for a superset of both valid Pi and of <suffix>, but 
+     in other cases it simply returns msymbol_hash_iw(STRING0).  */
...

> > +       case '_':
> > +         if (string[1] == '_')
> > +           {
> > +             if (((c = string[2]) < 'a' || c > 'z') && c != 'O')
> 
> Why does this `if' exist?
> 

Follows from the (newly added) comment above.  That condition indicates
a __ followed by something that can't be an encoded Ada identifier.  We
assume it is part of the suffix (as indicated in the comment, we accept
a superset of the valid suffixes).

> > +               return hash;
> > +             hash = 0;
> 
> Why do we restart calculating the hash here?

As the comments indicate, we are ignoring the prefix of a qualified
name to get wild-matching.

Thanks for your comments.

Paul