Dear Maude Abusers,

A new alpha release can be accessed at SRI-CSL in:

~eker/public_html/Maude/Alpha95a/

or downloaded from:

http://www.csl.sri.com/~eker/Maude/Alpha95a/

Alpha release site authentication:

User Name: maudeabuser
Password: bughunter

This is a bug fix release to fix a critical bug in the new memo
mechanism reported by Cathy. There is a general problem with hash
consing in that entry into the hash cons table ignores state
information such as reduction flags and sort indices in the dag
nodes. Losing reduction flags is fairly benign but losing sort
information can cause chaos. Here is an example: 

fmod SORT is
  sorts Foo Bar .
  subsort Foo < Bar .

  ops f g h : Foo -> Foo [memo] .
  ops a b c : -> Foo .
  eq b = h(f(a)) .
  eq c = h(f(a)) .
  op i : Foo Foo Foo -> Foo .
endfm

red i(g(a), b, c) .

This version fixes the problem for sorts by a (perhaps overly) cute
trick of bitwise ANDing sort indices each time a hash cons entry is
accessed. Because an unknown sort is recorded as -1 (i.e. all 1s in
binary) and known sorts for the same term should agree, this propagates
known sort information.

This version also supports a new environment variable:
  MAUDE_META_MODULE_CACHE_SIZE
which controls the maximum number of metamodules that are held in any
metalevel LRU cache. The default value if this environment variable is
not set (or set out of range) is 4. The environment variable  can be
set to any value between 1 and 1024. It may be useful to increase the
value for applications that _reuse_ more than 4 different
metamodules. However setting it too high can cause performance and
memory usage issues. Suppose you set the cache size to X. You want to
make sure that no more than X-1 other metamodules are accessed between
any two accesses to a resused metamodule M to ensure that M is not
evicted and then recompiled between accesses.

Currently our Intel Mac server is unavailable so I have only built a
Linux64 binary. Mac users will need to build from source.

Steven