The JIT was pessimistically using 64-bit jumps for long branches
or any jump between code that is allocated at different times.
Normally, though, code allocation stays within the same 32-bit
range of the heap, so stick to 32-bit jumps until forced by
allocation addresses to use 64-bit jump targets.
In `(if (pair? x) E1 E2)', convert `(car x)' in E1 to
`(unsafe-car x)', and similarly for `(cdr x)'. Also,
`(begin (car x) (cdr x))' converts to `(begin (car x)
(unsafe-cdr x))' since `(car x)' implies a `pair?' test
on `x'.
More consistent clearing avoids a kind of space unsafety. There's just
one buffer per thread, so it's difficult to turn non-clearing into
a detectable leak (I wasn't abel to construct an example), but it
might be possible. More practically, failing to clear the buffer
can make it difficult to debug memory use.
The scheme_is_multiprocessor() function wasn't the right guard
for whether to use a locking compare-and-swap instruction; any
use of pthread-based futures needs the compare-and-swap.
Merge to v5.3.1
This change doesn't speed up anything, so far. GC performance
of pairs (or anything) is determined almost completely by
its size in bytes, and this change doesn't affect the size of
pairs. At the same time, the change mostly replaces the obsolete
"xtagged" support, and I might have a better idea that builds on
this change, so I'm keeping it for now.
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.
This tracking allows the compiler to treat structure sub-type
declarations as generating constant results, and it also allows
the compiler to recognize an applications of a constructor or
predicate as functional.
The JIT takes advantage of known-constant bindings to avoid the
check that a variable is still bound to a structure predicate,
selector, or mutator; that makes the code short enough to really
inline. The inlined version takes about half the time of the
indirect version.
The compiler does not yet track bindings precisely enough to
recognize constants for sub-type declarations.
Turn use of a finalized ffi callout into a reported error,
instead of a crash. Clarify the existence of the finalizer
in the docs. Fix error logging of the finalizer thread.
Merge to v5.3.1
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
The new argument gets to chaperone/impersonate a guard at
the prompt, and it is applied when the continuation is applied ---
based on a wrapper on th prompt tag of the continuation (as opposed to
the prompt tag of the prompt).
The new argument gets to filter results that come from a
non-composable continuation that replaces one delimited
by a prompt using the chaperoned/impersonated prompt tag.
For simple structure types (no guards, no auto fields, no
procedure property). Inlined allocation makes structure
allocation a little faster; more significantly, it
make structure allocation future-safe.
When thie JIT guesses that an identifier is bound to a
structure predicate, getter, setter, etc., but that guess
turns out to be wrong, and the call is in a tail position,
then preserve tail-call behavior.
(Changes include some setup to inline structure constructors.)
Avoids a common problem with libffi installed by MacPorts
causing problems with a mismatch between an iconv installed
by MacPorts and the system iconv. (When libffi is installed,
then -I/opt/include for the libffi heads also picks up the
iconv headers, but the ordering of the lib flags doesn't
pick up libiconv from /opt/lib. We could try to hack around
this by ordering the flags just right, but it seems better
to avoid the issue.)
JIT-generated doesn't actually conform to the constraints
of the Win64 stack-unwind protocol. In pariticular,
JITted code might move the stack pointer after a "preamble"
that saves non-volatiles, and the frame pointer isn't in
the right place. So, we can't implement the generic unwind
hook --- but the JIT's stack traversal can interleave its own
unwinding with the OS-supplied unwinding interface.
The new parameter (and supporting environment variables and
command-line flags) can bytecode lookup to a tree other than
where a source file resides, so that sources and generated
compiled files can be kept separate. It also supports storing
bytecode files in a version-specific location (either with
the source or elsewhere).
Clients of scheme_apply(), scheme_eval(), etc. (i.e., the variants
without a leading "_") except aborts to continue jumping out, while
a recent change to make them behavior more like a default prompt
handler caused them to return on errors. Changethe handler to behave
like the default, except that after running a result thunk, the
handler effectively aborts again.
The `make-log-receiver' function now includes a logger-name
filter. This filter is implemented as a low enough level that
it affects `log-level?' tests to check whether a log message
needs to be constructed at all.
The -W and -L flags and PLTSTDERR and PLTSYSLOG environment variables
support filters of the form "<level> <level>@<name> ...", where
<level>@<name> specializes filtering of events for a logger whose
name matches <name> to show <level> and higher.
The old `cast' didn't work right for a mismatch between
a pointer GCableness and the source or target types, and
it didn't work right for an GCable pointer with a non-zero
offset. While those pitfalls were documented, the first
of them definitely has been a source of bugs in code that
I wrote.
Also added `cpointer-gcable?'
Using `call-as-atomic' isn't right, because that allows an escape
via `call-as-nonatomic'. Assuming that `call-as-nonatomic' isn't
used, it seems like `call-as-atomic' should be ok, anyway, but
somehow its leads to unbalanced `end-atomic' calls.
Add `file-position*', which can return #f instead of raising
an exception when a port's position is unknown. Change
`make-input-port' and `make-output-port' to accept more
kinds of values as the initial position.
These changes make it possible to synchronize a port's
position with a `port-commit-peeked' action. It's ugly,
which I think reflect something broken about position
tracking in the port protocol (which seems difficult to fix
without breaking compaibility).
Providing a port instead of a reading or writing procedure
redirects the read/write to the specified port. This shortcut
is kind of a hack, but the run-time system can easily streamline
the redirection when it's exposed this way.
Using the new redirection feature reduces overhead in
`with-output-to-bytes' and `pretty-print'.
Previously, the use of `pretty-print' for `print' by `racket/main'
and the use of `transplant-output-port' by `pretty-print' caused
a flush to happen.
This problem highlights a potential general issue, which is that
`pretty-print' happened to always flush its output before, and
some programs may accidentally depend on that behavior.
We can't disallow the creation of bad mutators without breaking
old code, but we can prevent the JIT from treating them like
good ones.
Closes PR 13062
Since SIGHUP normally means that the output has gone away,
don't try to write to it.
Closes PR 13058 (although it doesn't solve the more general
problem that is noted in the PR)
The bytecode compiler used to convert an expresison like
(<flonum-op> <simple-expr> <complex-expr>)
to
(let ([id <complex-expr>])
(<flonum-op> <simple-expr> id))
to help an older version of the JIT avoid boxing the result of
<simple-expr>, but that transformation isn't needed, since the JIT can
keep unboxed values on the stack.
The transformation was complex and apparently buggy.
The changes also include a repair to the JIT for code that the
bytecode compiler formerly wouldn't generate, but which is allowed
bytecode.
Closes PR 13052
Generalize splitting of `(let-values ([(x ...) (values e ...)]) ....)'
to `(let ([x e] ...) ....)' for any `e', since it's always equivalent.
Right?
(The old requirements on the `e's seem to be needed only for
`letrec-values' splitting and maybe mutable variables.)
Treat unsafe functional operations (which never raise an
exception) as omitable, which means that simple `let-values'
combinations can be split into `let' bindings, etc.
corner of the definitions window, based on the information that check
syntax computes
This commit contains two separate changes to make this work:
- adding a new renderer, based on the text renderer, that
pulls out the contents of the blue boxes and saves them
in the doc/ directories (specifically in the files named
contract-blueboxes.rktd)
- extend check syntax to use and display the information
build by the new renderer
Internally, there's a `prop:method-arity-error' property that is
used for keyword-accepting methods. The same thing could be
accomplished with `procedure->method', but the new property avoids
a wrapper. It might be nice to expose the property from `racket/base',
but that creates trouble for generating arity errors for keyword-
requiring procedures (i.e., when such a procedure is wrapped), so
keep it provate for now.
Closes PR 12982
A progress evt from a close input port must be initially ready,
and the primitive `peek-bytes-avail!' checks a progress evt
before checking whether the port is closed.
These changes resolve a race in `read-bytes-evt' and related evt
constructors.
Fix tchecking for a rest argument to a function that
is lifted by closure conversion so that one of its
arguments is a mutable local variable's location.
Also reject bytecode that would pass too many arguments
to a lifted function, since that would trigger an arity
error that might try to use a location as a value.
Merge to v5.3
When a `port-commit-peeked' succeeds, position information should
(appear to) be updated. This patch synchronizes commits and
position information for primitive ports, but synchronizing
them for user ports remains a problem.
When `guard-evt' or `nack-guard-evt' is used, `sync' didn't
account for the possibility that a channel or semaphore
action might complete during the call to the guard, in which
case it might fail to select the event that has already
completed.
Merge to v5.3
Convert
(hash-ref <hash> <key-expr> (lambda () <literal>))
to
(hash-ref <hash> <key-expr> <literal>)
which is useful for making the `case' expansion fit
Typed Racket.
The "simpliciation" of a syntax object's lexical context was dropping
module contexts that have no bindings, but those contexts now
contribute to the identifty of some bindings. Fix simplification
to replace the full rename rename with a simplified one, instead
of just dropping it.
Merge to v5.3
When a module is loaded from bytecode and then the value of
`use-compiled-file-paths' changes, an attempt to load a submodule
would fail, because source isn't used if the main module is
already declared, and the bytecode code is not used according to
`use-compiled-file-paths'. Make the bytecode path stick when it
is used once, so that submodule loads succeed, and make it work
even with `namespace-module-attach'.
The module-attach part of this protocol requires a change to the
API of a module name resolver: the notification mode gets two
arguments, instead of one, where the second argument is an
environment.
The lambda-lifting transformation needs to iterate to a fixpoint
where each lambda's added arguments and order are known. The
check for whether something changed was formerly just the number
of added arguments, but that's not good enough, because a binding
might get lifted away while another one acquires an extra argument.
The right test is to check the count and original bindings for the
added arguments.
Closes PR 12910
Since the optimizer can lambda-lift a function and adds box arguments
(i.e., locations of mutable variables instead of values), then
the validator must check that calls and declarations are consistent.
But declaration-side information wasn't registered properly.
This is an old bug, but it was exposed by recent improvements
in unboxing.
Also, fix JIT implementation of explicitly decremented fuel on
a 64-bit platform, plus some other code clean-up.
Each future thread has its own x87 flags, apparently reset to the
default (at least on Mac OS X) rather than inherited from the
creating thread, so reset the x87 configuration in each new future
thread.
To support module caching, module path indexes must be cloned
for each use of the cached module, so that path resolutions
don't collide. Cloning was previously implemented at the point
of shifting the modidx based on the module instantiation name,
but now its cloned at declaration time. This result in better
sharing of module-path resolutions, which in turn speeds up
compile-time instantiation of modules, which in turn speeds up
interactions & examples in documentation (as much as 10% for the
Guide, for example).
Furthermore, the reverse cache within a modidx may have been
used improperly during cloning, and that's not a problem in
the new implementation.
The JIT wants to detect calls to functions bound to module-level
variables that are always instantiated to a procedure of a particular
arity, in which case it can avoid some checks. The problem is that
bytecode and JITted code can be shared via the module cache across
namespaces that use different modules to implement a particular module
path or that enable the JIT differently. In particular, starting with
one that has a procedure binding and then using one (in a different
namespace) that has a non-procedure binding could lead to a crash.
Defend against this possibility by never treating imported variables
as constant in that sense. The JIT detects imported variables through
a new import map in the prefix.
This change may slow code. My guess is that it will have no
performance impact in practice, due to cross-module inlining,
although I can construct a microbenchmark that slows by 20%.
The optimizer can no longer reduce
(with-continuation-mark _id _v-expr _expr)
to just
_expr
when _v-expr and _expr are simple enough, because _id
might be bound to a continuation mark key with an
impersonator that checks the result of _v-expr.
The loss of an optimization can have a significant affect on
errortrace of microbenchmarks, such as
(for ([i (in-range 10000000)])
i)
A arity-reduced procedure-valued `prop:procedure' was not handled
correctly, for example.
A good candidate for random testing? I had the right kind of test in
place, but only for an arity of 1. It turns out that testing any other
arity would have exposed the problem, so now there are tests with
arity 0. If I had randomly generated procedures instead of manually
constructing `f0' through `f1:+' in "procs.rktl", then maybe I would
have more naturally generalized the arity testing, too. Then again,
I did already have relevant inputs, and it was the testing of inputs
that was too specific.
Closes PR 12870
For all currently supported platforms, the result was already
portable, despite the documentation's hedging.
Also fixed up the documentation in other ways, such as the fact
that `seconds->date' returns a `date*'.
Add extra intitial-message lines, use "..." on a field name
to indicate that it could reasonably be hidden by default,
and refine some existing messages.
Includes the addition of 'overflow and 'start-overflow-work
events, whcih are effectively specializations of 'sync and
'start-work to expose overflow handling.
Also, fix a bug related to a potential GC during mark-stack
restore from a lightweight continuation.
By itself, this change won't help anything, because tail-call
allocation is triggered for something that can't be called directly.
This change sets up part of an improvement for future-local recovery
from stack overflow, though.
(I had trouble constructing a test that would trigger the new code.
Fortunately, the existing tests trigger it.)
When a "synchronized" operation is handled for a future thread
during a `touch' on the future, then still report the handling
as "synchronized" insteda of "blocked".
Also, fix FFI procedures to preserve names: change `ptr-ref' with
`_fpointer' on an `ffi-obj' value to return the `ffi-obj'
value, so that the name in the `ffi-obj' value can be used
by `_cprocedure'.
Closes PR 12645
The new predicates are `progress-evt?' `thread-cell-values?',
`prefab-key?', `semaphore-peek-evt?', and `channel-put-evt?'.
These were used internally, and now they appear in contract
error messages.
When supplying an accessor to redirect, either the corresponding field
must be accessible through the current inspector, or a mutator for
the same field must be redirected, too.
Stevie realized that we need this constraint; otherwise, impersonators
can implement mutator-like behavior even when the mutator is otherwise
secret.
Add `raise-argument-error', `raise-result-error', `raise-arguments-error',
and `raise-range-error'.
The old convention was designed for reporting on a single (sometimes very
long line). The new convention is
<name>: <short message>
<field>: <detail>
...
If <detail> is long or itself spans multiple lines, then it may
also use the form
<field>:
<detail>
where each line of <detail> is indented by 3 spaces.
Backtrace information is shown as a multi-line "context" field.
For example, if definitions have two unmarked `x's that originate
from different modules, make them correspond to different bindings.
This improvement will be used by `scribble/srcdoc', which will
rely on module context to connect `for-doc' requires to documentation
code that appears in the same module --- which is needed, for example,
if a macro expands to documentation code and the macro is used in
a different module.
These primitives atomically update a box to a new value, as long
as the current value is the same as a provided value. They also
are future-safe.
When futures are enabled, they use low-level hardware instructions
to perform the change atomically.
The expander no longer needs to generate certain phase-N
identifiers, since we now have a comparison operation that
uses different phases for each of two identifiers.
