An other key point of PIPS is the NEWGEN [7] description
language that is used in PIPS to define all the data structures.
Domains are defined NEWGEN to express complex data structures such
as sums (+ like a C-union) or products (x like a
C-struct) of other domains , lists (*), sets ({}),
array ([]) or maps (->) of domains . These domains
are translated by NEWGEN in type definitions for different languages
(C and Common-Lisp for now) with different methods to deal with this
data: creators, destructors, accessors, readers and writers, etc. The
last ones allow persistence and having PIPS written in different
languages but sharing the same data. In this way, NEWGEN adds an
object oriented layer that abstracts the target language data
structures.
In our dead code elimination example, we do not define any new type
but only use the internal representation of the abstract syntax tree.
An excerpt of its definition is on Figure 7. All
the NEWGEN data structures are defined in the LaTeX files from
$PIPS_DIR/Development/Documentation/newgen that are used to
generate the documentation about the types and of course the target
language specifications.
NEWGEN defines a generic mono- or multi-domain iterator
(gen_recurse and gen_multi_recurse in C) that
recursively applies functions on a list of domain top-down and then a
bottom-up on a data structure. In our example on
Figure 8 in
suppress_dead_code_statement() we iterate on all the statements
of the code and apply on them dead_statement_filter() top-down
to eliminate some dead code, recurse if this function returns TRUE ,
and then dead_statement_rewrite() bottom up to remove some code
that has become useless because of the previous elimination of nested
statements. All the accessors such as statement_instruction()
to get or set an instruction of a statement or the
is_instruction_loop tag to test if an instruction is a loop are
generated by NEWGEN .
Figure: Excerpt of the dead code elimination phase .