Since ijp's work, lower-cps is run before it reaches this stage.
* module/language/cps/compile-js.scm (lower-cps): Remove import.
(compile-js): Remove call to lower-cps.
* module/language/cps/effects-analysis.scm (compute-known-allocations):
(compute-clobber-map): Add "conts" parameter, and use it to compute
primcalls that access known allocations. A write to a known allocation
only clobbers a read to a known allocation if they are the same.
* module/language/cps/cse.scm (eliminate-common-subexpressions-in-fun):
Pass conts also to compute-clobber-map.
* module/language/cps/utils.scm (compute-var-representations): Use
'arg-representations from metadata for arg representations.
* module/language/tree-il/compile-cps.scm (sanitize-meta):
(convert): Make sure incoming terms have no arg representations.
* module/language/cps/graphs.scm (rename-keys, rename-intset)
(rename-graph, compute-reverse-control-flow-order)
(compute-live-variables): Move here from slot-allocation.
* module/language/cps/utils.scm: Remove duplicate compute-idoms
definition.
(compute-defs-and-uses, compute-var-representations): Move here from
slot-allocation.
* module/language/cps/slot-allocation.scm: Move routines out to utils
and graphs.
* module/ice-9/match.upstream.scm (match-gen-ellipsis): Instead of just
binding the identifier when matching (x ...), go through match-one so
that if the id is already bound, we unify instead.
* test-suite/tests/match.test ("unify in list patterns"): Add test.
* test-suite/tests/match.test.upstream: Add additional tests from
upstream.
See commit 05c546e38 in Chibi Scheme. Thanks to Alex Shinn for help
here!
Reported by Marius Bakke <marius@gnu.org>
at <https://issues.guix.gnu.org/50696>.
Previously, the baseline compiler would incorrectly emit a right shift
when for, say, (ash x 2), and a left shift for (ash x -2).
* module/language/tree-il/compile-bytecode.scm (canonicalize): When Y is
negative, emit 'rsh', not 'lsh'.
* test-suite/tests/numbers.test ("ash at -O1"): New test.
Previously the pattern would only match when the two 'src' values were
the same, which is not the case for example when running on the
interpreter.
* module/language/tree-il/compile-bytecode.scm (canonicalize): In 'ash
primcall pattern, rename second 'src' to 'src*'.
This flag was set, but never used in Guile, and there was no documented
API to access it.
To check if an array is contiguous, use (array-contents <> #t).
* libguile/arrays.h (scm_i_raw_array): New function.
SCM_I_ARRAY_CONTIGUOUS, SCM_SET_ARRAY_CONTIGUOUS_FLAG,
SCM_CLR_ARRAY_CONTIGUOUS_FLAG, SCM_I_ARRAY_CONTP: Remove.
scm_t_array_dim: Declare here, not in array-handle.h.
SCM_I_ARRAY_NDIM: Shift by one bit since the contp flag isn't there
anymore.
* module/syste/vm/assembler.scm: Match removal of contp flag.
* libguile/arrays.c (scm_i_make_array): Reuse scm_i_raw_array.
(scm_i_ra_set_contp): Remove.
(scm_transpose_array): Don't set or clear the contp flag.
(scm_make_shared_array): Don't set or clear the contp flag.
(scm_make_typed_array): Don't set the contp flag.
* libguile/array-map.c (scm_i_array_rebase): Reuse scm_i_raw_array.
This is up to 20%-30% faster than the previous versions in (scheme base) or
(srfi srfi-43) that used vector-move-left!/vector-move-right!.
* libguile/vectors.h:
* libguile/vectors.c: As stated.
* doc/ref/api-data.texi (vector-copy!): Document the new function.
(vector-fill!): Document optional arguments.
(vector-copy): Document optional arguments.
* module/scheme/base.scm: Reuse core vector-copy!.
* module/srfi/srfi-43.scm: Reuse core vector-copy!.
This is a regression since Guile 3.0.2 and breaks compilation
of a Guile library.
* module/ice-9/read.scm
(%read)[read-parenthesized]: When SAW-BRACE? is #t but CH isn't
#\#, don't eat CH.
* test-suite/tests/reader.test
("#{}#): Add four test cases.
In 3.0.7 (after 0cc7991855 "Ensure
that (syntax ()) results in ("), the use of #nil in syntax-rules
expansions like this:
(define-syntax foo
(syntax-rules ()
((_ x) (eq? #nil x))))
(foo #t)
could cause a crash that looks like this:
ice-9/psyntax.scm:2795:12: In procedure syntax-violation:
Syntax error:
unknown location: unexpected syntax in form ()
To fix it, add another special case (the commit mentioned above
special-cased the empty list) to preserve #nil
* module/ice-9/psyntax.scm (gen-syntax): Preserve #nil.
* test-suite/tests/syntax.test: Test #nil in syntax expansions.
Closes: 49305
* module/language/cps/cse.scm (compute-out-edges): Only propagate
constant to successor if successor not kf.
* test-suite/tests/compiler.test ("cse auxiliary definitions"):
("closure conversion"): Refactor.
("constant propagation"): New test.
Fixes#48368.
* am/bootstrap.am (SOURCES):
* module/Makefile.am (SOURCES):
* module/language/tree-il/optimize.scm (make-optimizer): Wire up the new
pass.
* module/language/tree-il/resolve-free-vars.scm: New pass.
* module/system/base/optimize.scm (available-optimizations): Enable new
pass at -O1.
* module/language/tree-il/letrectify.scm (letrectify): Inline "let"
bindings inside residualized "letrec*" forms, to allow the dominator
relationship to be reflected in the scope tree. Also, detect
"define-module*" invocations, and add these to the mod-vars set, so that
residualized "module-ensure-local-variable!" primcalls can clearly
denote their module without having to use "current-module".
Well, I think this is one of the necessary steps anyway.
* module/language/cps/compile-js.scm (compile-exp*): Switch use
of $closure to use the newer name, $const-fun.
However, it probably isn't correct. Right now it just always chooses
"bytecode"... how to get it to correctly/intelligently choose the right
path?
* module/language/cps/spec.scm (choose-compiler): New variable.
(cps): Add it as #:compiler-chooser target.
This enables the compilation from "manually" written Tree-IL to
bytecode. See also <https://bugs.gnu.org/45131>.
* system/base/compile.scm (read-and-compile)[(joint #f)]<? eof-object?>:
Join exps using the default joiner for to.
<exp>: Compute compiler for to.
* test-suite/test/compiler.test ("read-and-compile tree-il"): New test.
* testing.scm (%test-begin, %test-end): When (test-runner-current) is
not set, create a new one like before but also add a finalizer that
will remove it after the test is finished. Previously the test runner
was getting unset unconditionally.
See https://srfi-email.schemers.org/srfi-64/msg/16468240/
Fixes <https://bugs.gnu.org/47031>.
* module/language/tree-il/analyze.scm (make-use-before-definition-analysis)
[resolve]: Add case for when NAME is re-exported by MOD.
* test-suite/tests/tree-il.test ("warnings")["re-exported binding"]: New test.
* module/ice-9/r7rs-libraries.scm (define-library): Fix bug handling
include-library-declarations and cond-expand subforms. Thanks to Adam
Nelson for the fix! Fixes#40252.
* module/language/tree-il/effects.scm (make-effects-analyzer): The body
of a prompt is an expression only for escape-only prompts, and the
handler is always a lambda. Fix bug where a prompt could be incorrectly
marked effect-free.
* test-suite/tests/tree-il.test ("optimize"): Add test for bug 48098.
Fixes bug 48098.
* module/ice-9/psyntax.scm (syntax-violation): Pass source as an alist.
A sourcev would be better but it's incompatible.
* module/ice-9/psyntax-pp.scm: Regenerate.
* module/ice-9/read.scm (%read): Add 1 to both line and column: the
former to make 1-based user lines instead of 0-based Guile lines, and
the latter because apparently that's how we always did it :) Fixes
