UD-chains for fields

  I wonder if there has been any research done on the behavior of records in
dataflow analysis. What I mean is that the algorithms for the building of
ud-chains or alias analysis that I have read about consider only simple
variables and arrays. But field selection introduces significant
complexity. For example:


      s1: y = x;
              ...
      s2: y->a = v;
              ...
      s3: return (x->a.b);


  There seems to be no definition of x->a.b reaching s3, though s2 is such a
definition, due to aliasing of x to y in s1, even if the field b does not
appear explicitly in s2.


  I'd be really grateful if somebody could direct me to literature on this
problem.
[It seems to me that there's no extra complexity here beyond the problems
already introduced by pointers. Record offsets are all known at compile
time, and can be handled the same way as constant subscripts. Am I missing
something here? -John]
--

Return to the comp.compilers page.
Search the comp.compilers archives again.