Converting between strings and paths interferes with cross
compilation. This hasn't caused more problems only because cross
compilation has tended to run on Unix platforms, where the generated
paths are acceptable to Windows. But using strings goes wrong when
manipulating a Windows-based build for further cross-build actions on
Unix, and it can go wrong if paths contain bytes that cannot be
encoded in strings.
The `ffi/unsafe/string-list` library is not needed now that `(_list i
_string)` and similar work. This way, existing code that uses `(_list
i _string)`, `(_vector i _string)`, etc., does not needed to be
modified to work with CS.
We don't usually remove libraries, of course, but this one has only
existed for a few days, and it was made to work around a problem that
has been solved in a better way.
Related to #3825
A reference bytevector holds a mixture of addresses within GCable
objects and foreign addresses, where "address" corresponds to the
payload of a bytevector or flvector object. The GC knows to apply a
suitable offset to the reference, so that object counts as reachable
from a reference bytevector, and the reference bytevector is updated
if the object is relocated during a collection.
With this change, the restriction in Racket CS against passing
non-atomic memory to a foreign function can be lifted. For example,
`(_list i _string)` can be useful as the type of a foreign-call
argument.
Making reference bytevectors a subtype of bytevectors is not an
obvious choice, given that writing to a reference bytevector with
byte-level operations can easily corrupt it. But this choice makes
various things simpler and easier.
I'm not sure this is a good idea, but it doesn't apear to be a
particularly bad idea. Exporting symbols just makes the executable
size slightly larger.
Closes#3617
Passing an array of strings or byte strings to a foreign function can
be especially tedious on Racket CS, due to the prohibition against
passing an array of GCable pointers to a foreign function.
Closes#3825
In CS, a pointer to non-atomic memory cannot usefully be passed to a
foreign function. The general foreign-call path checked for that kind
of argument and raised an exception, but the check was missing from
the common-case fast path, so a meaningless argument would be quietly
passed to the foreign function.
Related to #3825
This speeds up `(factorial 1000000)` (using factorial from math/number-theory)
by about 3x, and the conversion of the result to a string by about 2x.
Benchmark:
#lang racket
(require math/number-theory)
(define n (time (factorial 1000000)))
(define s (time (number->string n)))
(string-length s)
Current Racket CS:
cpu time: 19135 real time: 19137 gc time: 372
cpu time: 33416 real time: 33418 gc time: 463
Current Racket BC (GMP is really fast):
cpu time: 1465 real time: 1465 gc time: 51
cpu time: 3661 real time: 3659 gc time: 3
This PR:
cpu time: 6173 real time: 6172 gc time: 168
cpu time: 17846 real time: 17847 gc time: 377
Cutoff between Karatsuba and Toom3 estimated by mflatt.
Cutoff between Toom3 & Toom4 guessed.
The argument type `(_ptr i _string/utf-16)` is not used on CS, because
a pointer to non-atomic memory cannot usefully be passed to a foreign
function from Racket CS.
Closes#3820
When a struct is called as a procedure and the struct is
impersonators, make a method-style `prop:procedure` receive the
impersonated structure as its argument. This change makes a
method-style `prop:procedure` more consistent with a field-index
`prop:procedure. The old behavior, meanwhile, seems to create an
unsoundness in Typed Racket.
Closes#2574
When an inlined `set!` mutates a variable that's bound in the inlined
function, the `set!` target was not updated to refer to the inlined
binding.
Closes#3817
When a function from `define-inline` is applied to the wrong number of
arguments or the wrong keyword arguments, then leave it as a runtime
error (with a compile-time warning) instead of a badly reported
compile-time error.
Closes#3402
This is a backward-incompatible change, but I think it's unlikely that
any code intentionally uses `unsyntax` or `unsyntax-splicing` within a
syntax-quoted box and expects it to stay literal.
Meanwhile, as @rocketnia noted, the documentation for `quasiquote` was
unclear about the espacing positions for `unquote` and
`unquote-splicing`, so this commit impoves that documentation. It
adjusts the documentation for `quasisyntax` to note that a hash table
value position is not an escape position, unlike for `quasiquote`.
(The lack of an escape position within hash tables is consistent with
`syntax`. That's arguably inconsistent with `quasiquote`, but it seems
simpler to leave that alone, and changing `syntax` just might matter
for existing code.)
Closes#3656
I can't see where the ABI pins this down for x86_64, but the default
compiler on Mac OS seems to have started caring that 1-byte and 2-byte
integer arguments are sign-extended in registers. The previous lack of
sign extension would affect only small-structure arguments.
Also, make some adjustments to "Mf-base" so that plain `make`
generates a summary file as before and so that output files are linked
in a way that lets `make root-experr` and `make patches` find them.
* refactor mats to allow different configurations to run in parallel.
The {partial,all,bully}x targets now support being run in parallel should
make decide to do so (e.g., via the -j flag)
* fix mats ignoring "rmg" parameter
* Update travis-ci build scripts to use new partialx target and run multiple
jobs in parallel, based on the number of processors available.
* Add a way to only run particular machines in travis-ci by including
a line that starts with "travis:only:" and lists the machine types in the
commit message.
