This is another old bug that could have caused validation failures
with flonums, but it showed up with fixnum tracking because fixnums
are more common (e.g., from `string-length').
There were really two bugs: information installed at the
wrong offet in one place, and a failure to detect that information
should be propagated in a different place. Fixing both avoids
a validation problem with `html/sgml-reader'.
Track fixnum results in the same way as flonum results to enable
unboxing, if that turns out to be useful. The intent of the change,
though, is to support other types in the future, such as "extnums".
The output `raco decompile' no longer includes `#%in', `#%flonum',
etc., annotations, which are mostly obvious and difficult to
keep in sync with the implementation. A local-binding name now
reflects a known type, however.
The change includes a bug repair for he bytecode compiler that
is independent of the generalization (i.e., the new test case
triggered the old problem using flonums).
Shape information allows the linker to check the importing
module's compile-time expectation against the run-time
value of its imports. The JIT, in turn, can rely on that
checking to better inline structure-type predicates, etc.,
and to more directy call JIT-generated code across
module boundaries.
In addition to checking the "shape" of an import, the import's
JITted vs. non-JITted state must be consistent. To prevent shifts
in JIT state, the `eval-jit-enabled' parameter is now restricted
in its effect to top-level bindings.
Bytecode changes in two small ways to help the validator:
* a cross-module variable reference preserves the compiler's
annotation on whether the reference is constant, fixed, or other
* lifted procedures now appear in the module body just before the
definitions that use them, instead of at the beginning of the
module body
Inline only trivial functions, such as `(empty? x)' -> `(null? x)',
to avoid generating too much code.
Bytecode includes a new `inline-variant' form, which records a
version of a function that is suitable for cross-module inlining.
Mostly, the variant let the run-time system to retain a copy
of the bytecode while JITting (and dropping the bytecode of)
the main variant, but it may be different from the main variant
in other ways that make it better for inlining (such a less loop
unrolling).
The JIT and bytecode compiler disagreed on the definition of
"constant". Now there are two levels: "constant" means constant across
all instantiations, and "fixed" means constant for a given instantation.
The JIT uses this distinction to generate direct-primitive calls
or not. (Without the distinction, a direct jump to `reverse' could
be wrong, because `racket/base' might get instantiated with the
JIT disabled or not.)
Also, fixed a bug in the JIT's `vector-set!' code in the case that
the target vector is a top-/module-level reference that is ready,
fixed, or constant.
- the `lam' structure from `compiler/zo-struct' changed to include a
`toplevel-map' field
This change helps solve a finalization problem in `racket/draw',
which in turn sigificantly reduces the peak memory use of `raco setup'
during the doc-building phase (because some documents load `racket/draw'
to render images, and multiple copies of `racket/draw' were retained
before finalization was fixed).
The change is an extreme way to solve a specific finalization
problem, but it's a kind of space-safety improvement; space safety
almost never matters, but when it does, then working around a lack of
space safety is practically impossible. In this case, it's not clear
how to otherwise solve the `racket/draw' finalization problem.
The improvement doesn't change the representation of closures, but it
requires special cooperation with the GC. All closures in a module
continue to share the same array of globals (plus syntax objects);
that is, instead of completely flat closures, Racket uses a two-level
environment where top-/module-level variables are grouped
together. The code half of a closure now records which
top-/module-level variables the body code actually uses, and the mark
phase of GC consults this information to retain only parts of the
top-/module-level environment frame that are actually used by some
closure (or all of the frame if it is accessible through some other
route). In other words, the GC supports a kind of "dependent
reference" to an array that is indexed by positions into the array
--- except that the code is more in the "Racket" directory instead of
the "GC" directory, since it's so specific to the closure
representation.
