* libguile/jit.c (emit_direct_tail_call): Assert self-tail call has
mcode.
(opcodes_seen, bitvector_ref, bitvector_set, compile1): Make the
opcodes_seen set more compact, and log all instruction emissions at
level 3.
(compute_mcode): Don't overwrite mcode if compilation fails.
* lightening/x86.c (reset_abi_arg_iterator): Reserve 32 stack bytes on
64-bit Windows systems, in accordance with ABI. Thanks a million to
Charles Stanhope for the patch and to Mike Gran for testing!
Suggested by Dale P. Smith <dsmich@roadrunner.com>.
* libguile/deprecated.h (SCM_MEMORY_ERROR): New macro.
* doc/ref/api-control.texi (Dynamic Wind): Use 'scm_report_out_of_memory'.
'scm_memory_error' was deprecated in 2014 in commit
c2247b782a and removed in 2017 in commit
c248ea10be. This is a followup.
* libguile/error.h (SCM_MEMORY_ERROR): Remove.
* doc/guile-api.alist: Remove 'scm_memory_error'.
* doc/ref/api-control.texi (Handling Errors): Likewise.
(Dynamic Wind): Use 'scm_misc_error' instead of 'scm_memory_error'.
This significantly speeds up loads that create lots of bignums, like
(language cps slot-allocation) for files with many top-level
definitions. Compiling such a file is typically 2.5 times faster.
See <https://lists.gnu.org/archive/html/guile-devel/2020-02/msg00023.html>.
* libguile/numbers.c (custom_gmp_malloc): Use
'scm_gc_malloc_pointerless' instead of 'scm_malloc'.
(custom_gmp_realloc): Use 'scm_gc_realloc'.
(custom_gmp_free): Remove call to 'free'.
(make_bignum): Use 'scm_gc_malloc' instead of 'scm_gc_malloc_pointerless'.
Call 'scm_i_set_finalizer' only when SCM_INSTALL_GMP_MEMORY_FUNCTIONS is
false.
Fixes <https://bugs.gnu.org/37846>.
Reported by Christopher Baines <mail@cbaines.net>.
* module/texinfo/plain-text.scm (acronym): New procedure.
(tag-handlers): Change 'acro' handle to ACRONYM, and add 'acronym'
handler.
* test-suite/tests/texinfo.plain-text.test ("stexi->plain-text")
["acronym", "recursive acronym"]: New tests.
* module/texinfo/plain-text.scm (*line-width*): New variable.
(wrap*): Honor it.
* doc/ref/texinfo.texi (texinfo plain-text): Document it.
* test-suite/tests/texinfo.plain-text.test: New file.
* test-suite/Makefile.am (SCM_TESTS): Add it.
This lets Guile build on OS X (tested on 10.14) with the system-provided
libreadline.
* acinclude.m4 (GUILE_READLINE): Check for rl_get_keymap_name instead of
rl_get_keymap. The system provided libreadline is an alias to libedit,
which has _keymap but not _keymap_name, and Guile uses both.
* guile-readline/readline.c: Adjust the include guard.
Previously, 'make-stack' calls with the inner and outer cuts arguments
would always return #f because 'call-thunk' wouldn't appear on the
stack.
* module/statprof.scm <top level>: Add self-assignment to 'call-thunk'.
* module/scripts/use2dot.scm: Load (ice-9 getopt-long) with #:use-module
rather than #:autoload. With the previous #:autoload spec, 'option-ref'
would be unbound due to the new autoload semantics.
* module/scripts/frisk.scm: Load (ice-9 getopt-long) with #:use-module
rather than #:autoload. With the previous #:autoload spec, 'option-ref'
would be unbound due to the new autoload semantics.
Fixes <https://bugs.gnu.org/39196>.
Reported by Matt Wette <matt.wette@gmail.com>.
* module/system/repl/command.scm: Augment #:autoload set of bindings.
Fixes <https://bugs.gnu.org/39183>.
These symbols had been inadvertently removed from libguile, even though
they were intended to be public (as they were in 2.2).
* libguile/expand.c: Include "memoize.h".
* GUILE-VERSION (GUILE_MAJOR_VERSION, GUILE_MINOR_VERSION)
(GUILE_MICRO_VERSION): Bump for 3.0.
(LIBGUILE_INTERFACE_CURRENT, LIBGUILE_INTERFACE_REVISION): Bump for
new stable series.
* module/language/tree-il/letrectify.scm (compute-procedures-without-identity):
(letrectify): Only eta-expand lambda references that appear outside
the operator position more than once. This should restore peoples'
expectations that (eqv? f f) without penalizing optimization.
This is largely based on Guix commit
bc3c41ce36349ed4ec758c70b48a7059e363043a and subsequent changes to that
code.
* module/web/client.scm (x509-certificate-directory): New variable.
(set-certificate-credentials-x509-trust-file!*)
(make-credendials-with-ca-trust-files, peer-certificate)
(assert-valid-server-certificate, print-tls-certificate-error): New
procedures.
<top level>: Add call to 'set-exception-printer!'.
(tls-wrap): Add #:verify-certificate? parameter. When it is true, call
'make-credendials-with-ca-trust-files', pass it to
'set-session-credentials!', and call 'assert-valid-server-certificate'.
(open-socket-for-uri): Add #:verify-certificate? parameter and pass it
to 'tls-wrap'.
(http-request): Add #:verify-certificate? parameter and pass it to
'open-socket-for-uri'.
(define-http-verb): Add #:verify-certificate? parameter and pass it to
'http-request'.
* doc/ref/web.texi (Web Client): Update documentation of
'open-socket-for-uri' and 'http-request'. Document
'x509-certificate-directory'.
This is a backport of Guix commit 7b9ac883ea62a816afbfa747c1377dc273c15c20.
* module/web/client.scm (tls-wrap): Catch 'gnutls-error' around
'handshake'. Upon ERROR/WARNING-ALERT-RECEIVED, print a message and
call 'handshake'.
Prior to commit cb14fd2143 (Guile 2.9.7),
autoloading a module would give you access to all its bindings. In
future versions, autoloading a module gives access only to the listed
bindings, as per #:select (see <https://bugs.gnu.org/38895>).
This commit adjusts autoloads to the new semantics, fixing a regression
introduced in cb14fd2143.
* module/web/client.scm <top level>: Remove 'module-autoload!' call.
(gnutls-module, ensure-gnutls): Remove.
(load-gnutls): New procedure.
(tls-wrap): Call it instead of 'ensure-gnutls'. Replace reference to
GNUTLS-MODULE by a call to 'resolve-interface'.
The current implementation of srfi-11s let-values allows later clauses
to access and modify variables bound in earlier clauses when the clause
is not a proper list.
* module/srfi/srfi-11.scm (let-values): Fix switched variable names.
* test-suite/tests/srfi-11.test (let-values): Add test checking that the
variable cannot be changed in later clauses.
Previously (fluid-ref (make-thread-local-fluid #t)) would return #f via
scm_fluid_ref because the internal scm_hashq_ref would return #f when
the fluid had not been set, and that was interpreted as an actual value
for the fluid.
Instead, just pass the fluid default as the default for the hash table
lookups so that we don't need a second step to determine if the fluid
was set.
Thanks to Andrew Gierth for tracking down the problem.
