Using `(local-expand <expr> 'module-begin <stops> def-ctx)` didn't
work right, because definitions added to `def-ctx` were not visible.
(While adding definitions before `module-begin` expansion is an unusual
thing to do, there's no reason that it has to fail.)
Repairs a problem with ce9894c8bf, where a large "inlined" vector
allocation is not actually inlined, but other parts of the JIT
assume that it will behave as inlined --- which implies that the
runstack will be left unchanged after the call.
Closes#1868
In Windows 10 Fall Creators Update, a OneDrive mount looks to Racket like
a reparse point that refers to itself. Make Racket stop trying to resolve
the path further in that case.
Relevant to #1671
Commit 00d438cfbe made an attempt at this,
but this commit does it in a much more careful way, based on manually
emulating how the macroexpander expands module* forms in order to allow
splicing-syntax-parameterize to apply even within #%module-begin forms
introduced by the expander.
Mostly, fix "compiled" versus "expanded" in an error message,
but this variant of the message copied from the new expander
implementation seems better overall.
Repairs a problem reported by Laurent Orseau, where running the
following program with errotrace enabled reports an error:
#lang racket/base
(require racket/class)
(new object%)
(define-namespace-anchor nsa)
(define ns (namespace-anchor->namespace nsa))
(eval '(new object%) ns)
I haven't been able to make a standalone test, yet, but I'll keep
trying.
Normally, it's impossible to generate lots of `eq?`-hashing
collisions, but when the compiler inlines a function, it can duplicate
variables in a way that gives each copy the same `eq?` hash code. The
immutable-hash tree implementation failed when more than 32 collisions
occurred (which triggers a subtree in the collision node).
It's similarly very difficult to generate > 32 values that collide on
`eqv?` hashes but are not `eqv?` (although it must be possible using
exact rationals or complex numbers).
On my machine, this reduces the running time of the sqlite3
tests by about 1/4 (~3.2s to ~2.4s).
Other things I tried that didn't make as big a difference:
- coalesce A regions
- fast path for call-with-lock
Note: quasisyntax has a bug: #`(... (1 2 #,@(list 3) 4)).
Within an escape, no way to express splicing desugaring.
So add a private variant of ?@ that is interpreted even escaped.
Make datum->syntax explicit in guide rather than combined with
constructors like t-cons/x and t-dots (conditional).
This will make datum support easier, later.
For now, it makes it easier to do relocate correctly.
Also, make t-metafun, h-splice inlinable.
Although `raco pkg` doesn't use a package's ring number, it's useful
to preserve for other tools (like the pkg-build service). Adjust `raco
pkg catalog-copy` and `raco pkg catalog-info` to recognize and store a
ring number.
A mistake in the rktio conversion causes a crash if certain functions,
such as `directory-exists?`, are used before certain other functions,
such as `resolve-path`.
Thanks to Alex Harsanyi for the report.
The rktio conversion lost the deregistration of file descriptors in an
internal fd-to-semaphore table building on kqueue/epoll, causing the
wrong semaphore to be checked for a later recycling of the file
descriptor. This bug mainly affects Linux and ports created by
`subprocess`, since kqueue is not used for pipes on Mac OS and BSD
variants. The bug does not affect network sockets (which are the
primary intended clients of epoll/kqueue support), since the relevant
semaphore is deregistered when a socket is closed.
Thanks to James Bornholt for discovering the problem and providing the
repair.
Closes#1769
Fix a problem with compile-time bindings added to a
namespace created by `module->namespace` for a module
that does not have a source file.
Possibly, there's a different fault that should be fixed that caused a
binding to use the module's instantiation-time module path index
instead of its compile-time module path index (which is what happens
when a file is involved). This repair fixes the problem in a general
way, though, and leaves further improvement to the reimplementation of
the expander in Racket (which already does not suffer from the bug).
Thanks to Alexis for providing the example.
In a non-tail position, a JIT-generated application of `apply`
retained the argument list until the called function returned.
Fix it to drop the reference to the list before the function
is called.
And unquoted-printing string contains a string to `display` in all
print modes. Although it could be implemented with a structure type
that has a printing function, `raise-arguments-error` further treats
unquoted-printing strings specially by not using the error value
conversion handler, so it reliably produces literal text in the error
message; that way, `raise-arguments-error` can be used to construct
more error messages.
The hack to implement `_union` without help from libffi failed when
the total size of the variants is too large. Try a different approach,
which involves a bet that the total size plus whether the content is
all floating-point numbers will be enough information for most cases.
Relevant to #1351
The switch to rktio swapped the sense of "signal" and "sigset"
incompletely, and "signal" on Solaris is the one that resets
the signal handler to SIG_DFL.
Make `raco setup` propagate the original value of
`use-compiled-file-paths` in case it's reset to avoid loading bytecode
files. Then, `--clean` mode can remove bytecode relevant to that
setting, instead of always clearing "compiled" subdirectories.
There's no external way to initialize `use-compiled-file-paths` right
now, other than forcing it to `null` with the `-c` flag at the
`racket` level, but the current "racket7" implementation uses
different `use-compiled-file-paths` settings for different build
modes, and it seems to make sense in general.
Also, make `--clean` sensitive to `-D` and `-d`, so that it's easy to
clean just bytecode.
The copy was made in c1c427a281 to help get "racket7" started, but
it's no longer needed. Using the copy had an unintended consequence of
duplicating the collection-search cache for phases other than phase 0,
so this change repairs that consequence.
The `racket/private/check` module is dead code in this repo, but
having it simplifies the "racket7" bootstrap when "racket7" goes so
wrong that it can't rebuild itself.
To strip a signature, the old implementation effectively guessed
at the padding that was added to the original linkedit segment
to acommodate 16-byte alignment of the signature. The repaired
stripping works out the actual end of the linkedit segment based
on the various load commands that refer to it.
Add guard, if first argument to `find-relative-path` is not a path,
then don't call `path-convention-type` on it.
Thanks to Jack Firth for reporting:
https://github.com/racket/rackunit/pull/46
The cause is any extra space after a terminating semi-colon in a SQL string passed to sqlite3_prepare_v2 procedure in the prepare1* method of `db` Sqlite3 connection% implementation will result in non-false tail and zero prep-status values being returned. The checking code sees the non-false tail value and raises the "multiple statements given" exception.
The easy fix is to trim any trailing whitespace by calling string-trim on the buffer string before its converted to a byte string in the copy-buffer procedure used by sqlite3_prepare_v2 to initialize the sql-buffer.
See Issue #1702
The same problem is encountered when executing a DDL statement such as "Create Table" more than once after the table had already been created by the first execution.
The cause is when sqlite3_prepare_v2 encounters a bad statement, it aborts the preparation, sets prep-status flag to non-zero and also sets tail? to be non-false.
The previous code ignored the prep-status flag and only checked tail? and raises "multiple statements given" exception when tail? is non-false.
The new code takes prep-status into account and extracts the actual Sqlite3 error message by evaluating get-error-message and raising it as an exception.
See Issue #1702
Also, add a script to extract the "rktio.h" header information to an
S-expression form suitable for FFI use, and refine the "rktio.h"
information to improve that output.
Avoid printing `(unquote @d)` or `(unsyntax @d)` as `,@d` or `#,@d`,
which would mean `(unquote-splicing @d)` or `(unsyntax-splicing @d)`
to the reader, by adding an extra space before the `@`.
When a package to be installed has a module that conflicts with an
existing installed module, but the existing module isn't in a package,
then say so explicitly in the error message. Also report the current
collection and links paths.
Error reporting could do even more work to figure out which path is
relevant, but reporting a list of paths is relatively easy. Listing
paths also seems likely to give a user enough hints for this rare
case, especially if the user just needs to be reminded that collection
links exist.
Additionally this adds macros that distinguish between the chaperone redirects of prop-only vector
chaperones, function chaperones, and structure chaperones since each of these may store a vector
in the redirects field.
Auto fields were incorrectly recorded as immutable in a structure type
that is first generated from the prefab struct key instead of
`make-struct-type`.
Thanks to Deren Dohoda for the report.
When `--update-deps` (which is implied by `--all`) was used on
user-scope packages that depend on installation-scope packages, the
installation-scope dependencies were treated as missing, which forced
an update of the user-scope packages. Check both scopes for
dependencies.
Closes#1730
The `file/cache` relies on matching the text of a message, and the
text had changed to have less information on Unix, including whether
the problem was with the source or destination file. Add a notion of
error step to rktio and use it to improve the `copy-file` message.
Report a proper error when path decoding fails, instead of
synthesizing a path hat shouldn't exist. The rktio conversion
made it much easier to report the error at the Racket level
like other filesystem errors.
In the process of extracting minimal Windows path encoding for rktio,
I noticed a decoding issue with a path that ends with an unpaired
high-surrogate value. Add a suitable tests and fix the old decoder
(although it will probably go away).
Closes#1721, which points out that the `for/list` expansion
introduced in commit 5e94a906cd interacts badly with a body
that captures a continuation plus Racket's current implemenation
of continuations.
When Racket one day gets a better implementation of continuations,
this change could be considered again, but the general question is
whether programs can detect or be affected by the size of the
continuation (when the programs don't directly control the
continuation creation --- otherwise continuation marks obviously
expose the size).
Provide a compatibility layer to bridge the old (not completely
documented) `scheme_fd` interface and rktio. The old interface is
used by the Gtk implementation of `racket/gui`, for example.
Process handling relies on SIGCHLD being blocked, but signals are
blocked per-thread in Linux, so SIGCHLD needs to be blocked before new
threads are created.
For `open-output-file`:
Move 'replace mode handling out of rktio and into the client.
Formerly, on Windows, 'replace mode was just 'truncate with a fallback
for permission problems, so this change makes it delete an existing
file and replace it, which is more consistent with how 'replace has
always worked on Unix.
In 'truncate/replace mode, if a truncating open fails due to a
permission error and the file exists, then try again as a 'replace.
That's how it worked on Windows before, and now it's how Unix works.
The world doesn't need yet another cross-platform I/O library, but
it's getting one. This one has exactly the things that Racket needs,
and pulling it out will make it reusable from other VMs while
improving the Racket code organization.
This first step just gets started.
For a term
(lambda (arg-id ...) (define def-id _rhs) ... (arg-id def-id) ...)
the expander could take quadratic time in the number of `def-id`s
due to walking an environment to remove use-site scopes. (The
variant of the expander rewritten in Racket didn't have this
problem.)
This commit add mzrt_sema_try_wait to the functions that operate
on semaphores. The existing ones are:
* int mzrt_sema_create(mzrt_sema **sema, int init);
* int mzrt_sema_post(mzrt_sema *sema);
* int mzrt_sema_wait(mzrt_sema *sema);
* int mzrt_sema_destroy(mzrt_sema *sema);
This make-c-id allows an author to specify a convention for how
to connect and identifier defined with define-ffi-definer and
the actual symbol in the file.
* Adds docs.
* Adds tests.
* Adds history.
Accessing unsafe functionality through the FFI seemed like a good way
to avoid writing C code, but it made things more complicated instead
of easier, and it interacts badly with a more agressive shift away
from C (such as porting to Chez Scheme). So, add functions to the
primitive `#%unsafe` module, instead.
An authentic structure type is one whose instances cannot be
impersonated or chaperoned. The intended use of `prop:authentic` is to
annotate a library-private data structure where impersonators are
never needed internally for the data structure, and the declaration
lets the compiler produce less code and fewer branches by omitting
impersonator support.
In non-cross mode, `-C` needs to go after `-G` and `-X` when setting
up a "bundle" directory to turn into an installer, because that mode
needs to use foreign libraries (such as SQLite) at build time, and it
can use the instances that are being set up for the installer.
Meanwhile, improve the advice for setting `PLAIN_RACKET` to use `-C`
for a cross-platform build mode, even though things tend to work
anyway without it.
When comparing a part of a hamt that is a collision node versus a
subtree node, a "hash code" was extracted from the collision node ---
but that's really a code for an integer key is that used for the
collision element. The comparison should instead use a code extracted
from the reference to the collision node (which is the code that is
common to all colliding keys).
Detangle the target and host DLL and library directories by
making `get-lib-search-dirs` and `get-dll-dir` report the
host system's directories, and add `get-cross-lib-search-dirs`
and `get-cross-dll-dir`.
A new `-C`/`--cross` flag causes `racket` to save a host config and
collection directory and make them available via `(find-system-path
'host-{config,collects}-dir)`, while plus `(system-type 'cross)`
reports whether `-C` mode is in effect. Besides making the host paths
available, this change allows a same-platform build to run in
corss-platform mode.
The immediate problem to solve was the creation of Windows installers
on Windows, where recent changes to support 'gui-bin-dir configuration
need a clear distinction between the host Racket and the target Racket
being built, even if they're the same platform. (The "GRacket.exe"
executable didn't work, for example.)
The changes in this commit are more than needed for the immediate
problem, but they naturally build on the necessary `-C` flag, and they
support cross-platform package setup where native libraries are needed
during setup.
Avoid the well-known possibility of quadratic handling of ephemeron
chains, where all ephemerons are immediately known, no keys are
immediately known, and each link in the chain has a value that refers
to the next link's key.
To aviod quadratic behior, attach a list of ephemerons to each page of
allocated objecst, where marking any object on the page triggers a
rescan of the ephemerons without waiting to rescan all ephemerons.
Make `log` in `racket/base` optionally accept a second argument.
The second argument is the log `base`. The docs also recommend
`fllogb` when precision is important.
* Error message when base is 1
* Added docs.
* Add tests.
Store relative paths in "info-cache.rktd" (which corresponds,
roughly, to packages) in a platform-independent form, instead
of using the current platform's convention.
Using the current platform's covention works badly when
cross-compiling for Windows on Unix, since relative paths are used as
keys in the "info-cache.rktd" table. For example, updating a
pre-installed package on Windows mangles the mapping if the installer
is created from a cross-compiled installation.
Optimization to convert `(hash-ref <ht> <key> (lambda () <constant>))`
to `(hash-ref <ht> <key> <constant>)` didn't check that the `lambda`
for had zero argument.
Closes#1648
* generics: optional scope arg for private macros
To make them more friendly to macros that expand to generics
* add tests for generic-method-table macro
and/c, and between/c (which implies <=/c and >=/c) so that they turn
themselves into integer-in when appropriate
for example, (contract-stronger? (integer-in 0 4) (and/c natural? (<=/c 4)))
returns #t
- positive-integer?
- negative-integer?
- nonpositive-integer?
- nonnegative-integer?
These are like their exact-* counterparts provided by racket/base,
but they work for inexact numbers, and not just exact ones.
Fix problem with once-use tracking and delayed variable-use marking
that is performed for local function bodies. A delayed variable-use
registration might happen after a once-used variable is replaced by
its use.
This scenario is difficult to provoke, because the optimizer has to
first decide not to move a once-use function, and in a latter pass
decide to move it after all. There's not enough information to
retract the tentative use plus its transitive implications.
The solution is to avoid the generic once-use layer for `lambda` forms
whose uses are delayed (and that likely has a good effect on inlining
anyway). The other half of the solution is to avoid transitive use
marking on a once-used variable whose expression has been moved (and
there are no transitive things to skip, because that expression isn't
a `lambda` form).
It appears that Mac OS wants `RTLD_LOCAL` in the dlopen() call,
otherwise dlsym() searches through all previously open shared objects
– even though dlopen() is given a specific library handle.
The expr/c syntax class, as well as its underlying implementation
function, wrap-expr/c, previously produced misleading error messages.
The main purpose of these tools is to ensure a user-provided expression
conforms to a macro-provided contract. However, contract errors produced
by these forms were consistent with situations where both value and
contract were provided by the same party.
This fixes the discrepancy by changing how these forms assign blame to
emulate contract errors that arise from improper function arguments,
since most expressions provided to macros are semantically similar to
function arguments. All examples within the documentation itself
reflect this use case.
These changes alter the contents of error messages raised by expr/c and
wrap-expr/c, which could theoretically break some test suites, but it’s
extremely unlikely that any non-test code would depend on the precise
wording of contract error messages, and the interface is otherwise
completely backwards-compatible.
fixes#1412
When the second argument to `bytes-set!` is a reference to a
module-level variable that is definitely defined but not a known
constant, then an incorrect reordering was used that would cause
the third argument value to get overwritten before the call.
Closes#1601
The primitive `read` uses a shortcut --- a private "ungetc"
implementation --- that did not count position correctly for
non-ASCII characters.
Closes#1599
The documentation says that it should work on any output port,
although there's special treatment of ports that originate
from `pretty-print` itself.
Closes#1579.
Allow the directory for GUI executables to be specified as different
from console executables. The defaults for those two are different
on Mac OS, and configuring them differently might be useful to
address #1575.
Although there is probably no demand on Windows or Unix for splitting
the console and GUI bin directories, this patch tries to make things
work sensible there. On Windows, there's a corner case where a
launcher that starts GRacket (especially with `-z`) is intended to be
a console executable. The launcher creator can be told that via a
`subsystem` option, but a new `#:console?` argument was needed for
`make-gracket-launcher-path` lets the path selector know.
After some reductions, the new rator advance less the effect
clocks than the original rator. For example in
(equal? x 7) ==> (eq? x 7)
(my-struct? x) ==> #t or #f
The lambdas can be marked as single valued and/or mark preserving.
With this information is possible to remove unnecessary wrapping
like the `values` in
(let ([f (lambda () '(1))])
(display f f)
(values (f)))
or in reductions like
(car (list (f))) ==> (values (f)) ==> (f)
Moreover, this is useful to test that the optimizer has marked
correctly the function f as single valued and mark preserving.
If a module has any sort of complex bindings, such as a definition of
a macor-introduced identifiers, then `module->namespace` and variants
(like `variable-reference->namespace`) need to recreate suitable
bindings. Make sure that the module-path index for recreated bindings
is the run-time one, not the compile-time one.
Closes#1584
To avoid moving expressions that may have a side effect, the optimizer must
recognize that in this position this will cause an error and advance
the virtual clock.
Currently the only primitive that is flagged as SCHEME_PRIM_IS_OMITABLE and
may have multiple return values is `values`.
Thanks to Robby for finding the original version of the test.
When a hash table or other special value appears immediately on the
right-hand side of `define-values`, it needs to be protected by an
explicit quote when writing to bytecode.
Closes#1580
* byte-regexp? values should not be considered 3D syntax.
* hash? values are now allowed in serialized syntax properties with (template … #:properties (…))
* marshalling properties which were prefab structs called map on the result of struct->vector, changed it to struct->list as the struct "name" is always serializable.
Continuing the saga that includes 8190a7730d and d1ba9fbb6e, it turns
out that a 0-binding clause as the last one isn't so special after
all. A little later in the optimizer, now that we're sometimes moving
an error to the body, we can't assume that the body can be discard
if an error was detected.
Set up bindings and shift phases as needed to make
`variable-reference->namespace` work in a run-time position when the
enclosing module is instantiated at a phase other than 0.
Thanks to Rohin Shah for the bug report.
Support an external implementation of `read-syntax` by exposing
functionality that is currently internal to `read-syntax`: a srcloc
argument to a "special"-producing port function and wrapping special
results to reliably distinguish them from characters.
When Racket is run with stdout or stderr redirected to a file,
then it must be treated as a regular file, otherwise flushing
and position counting doesn't work right.
Merge to v6.8
In e59f888, new GCs no longer inherit the `avoid_collection` value set by
PLTDISABLEGC, and so that setting is lost as soon as the master place spawns
(due to `GC_switch_out_master_gc`).
When multiple-binding `let-values` form is split into a single-binding
form on the grounds that the right-hand side will definitely error,
the optimizer's effect clocks were advance incorrectly.
Closes#1552
Instead of duplicating a context line, show "[repeats <n> times]".
This improvement particularly helps avoid showing less context
now that `for/list` creates a non-tail recursion to build up
the result list.
Specifically, when it sees these contracts:
(and/c real? negative?)
(and/c real? positive?)
(and/c real? (not/c positive?))
(and/c real? (not/c negative?))
it generates the corresponding use of >=/c, <=/c, </c, or >/c, but
those contracts have also been adjusted to report their names as
(and/c real? ...).
This mostly is an improvement for contract-stronger, but also make it
so that (between/c -inf.0 +inf.0) just uses the real? predicate
directly, instead of a more complex function
Although "macOS" is the correct name for Apple's current desktop OS,
we've decided to go with "Mac OS" to cover all of Apple's Unix-like
desktop OS versions. The label "Mac OS" is more readable, clear in
context (i.e., unlikely to be confused with the Mac OSes that
proceeded Mac OS X), and as likely to match Apple's future OS names
as anything.
For example, an `unsafe-unbox` call should not be moved past the
call to an unknown function that might change a box's content.
Thanks to Sergey Pinaev for the report.
Using `--disable-jit` causes futures to be disabled, and places are
always disabled for CGC; in that case thread-local variables are
not needed. Meanwhile, the 3m build still has places, so a "gmp.c"
compiled without thread-local support is broken.
The objective of lookup_constant_proc and the first part of
optimize_for_inline was to find out if the value of an expression was a
procedure and get it to analyze its properties or try to inline it. Both
were called together in a few places, because each one had some special
cases that were missing in the other.
So, move the lookup and special cases from optimize_for_inline to
lookup_constant_proc, and keep only the code relevant to inlinig in
optimize_for_inline.
If an OS-level thread other than a Racket thread logs a message, then
the message needs to be queued instead of handled immediately.
If multiple places are running, then the right handler thread is not
clear, so just queue to the main place's thread.
Closesracket/gui#66Closesracket/drracket#77
Implement POSIX.1-2001/pax and GNU extensions for long paths and links
in `untar` and `tar`. Add a `#:format` argument to `tar` to select
among POSIX.1-2001/pax, GNU, or error encoding for long paths.
it doesn't apply it the second time (since we know that the
only difference for indy blame is in the negative position
and we know that flat contracts never assign negative blame)
This commit combined with the two previous (2b9d855 and 003e8c7) do
not seem to have a significant effect on the performance of ->i
contract checking. In particular, I see a 50% slowdown between the
version before and the version after these commits on the third `time`
expression below, but no significant difference on the first two.
(without the improvement to flat-contract?, these commits are
a significant slowdown to `g`)
#lang racket
(require profile)
(define f
(contract (->i ([y () integer?]
[x (y) integer?])
(values [a () integer?]
[b (a) integer?]))
values
'pos 'neg))
(define g
(contract (->i ([y () (<=/c 10)]
[x (y) (>=/c y)])
(values [a () (<=/c 10)]
[b (a) (>=/c a)]))
values
'pos 'neg))
(define (slow-predicate n)
(cond
[(zero? n) #t]
[else (slow-predicate (- n 1))]))
(define h
(contract (->i ([y () slow-predicate]
[x (y) slow-predicate])
(values [a () slow-predicate]
[b (a) slow-predicate]))
values
'pos 'neg))
(time
(for ([x (in-range 100000)])
(f 1 2) (f 1 2) (f 1 2)
(f 1 2) (f 1 2) (f 1 2)
(f 1 2) (f 1 2) (f 1 2)))
(time
(for ([x (in-range 100000)])
(g 1 2) (g 1 2) (g 1 2)
(g 1 2) (g 1 2) (g 1 2)
(g 1 2) (g 1 2) (g 1 2)))
(time
(for ([x (in-range 10000)])
(h 50000 50000)))
When the main interpreter loop is called for an application where the
argument array coincides with the current runstack pointer, then the
protocol is that the callee gets to modify that space --- and it
should modify that space as arguments become unused. The interpreter
was always copying arguments to a fresh space, though.
Both function have a similar purpose and implementation, so merge them to consider
all the special cases for both uses.
In particular, detect that:
(if x (error 'e) (void)) is single-valued
(with-continuation-mark <chaperone-key> <val> <omittable>) is not tail sensitive.
Also, as ensure_single_value was checking also that the expression was has not a
continuation mark in tail position, it added in some cases an unnecessary
wrapper. Now ensure_single_value checks only that the expression produces
a single vale and a new function ensure_single_value_noncm checks both
properties like the old function.
