* libguile/guardians.c: Reimplement in terms of chained finalizers and
weak tables.
* test-suite/tests/guardians.test: Add test for weak mappings added
after registering an object with a guardian.
* libguile/weak-table.c: Rewrite to use finalizers instead of
disappearing links. This will allow weak table associations to
persist beyond resuscitation of an object by a guardian. It also
removes the need to periodically vacuum the table.
* test-suite/tests/guardians.test: Add test for weak maps persisting
across guardian resuscitations.
* libguile/weak-set.c:
* libguile/weak-table.c: Attempt to avoid bouncing between sizes when
growing the vector causes elements to be removed from the set/table.
* libguile/modules.c: In my current image, there are 1790 bindings in
the root module, which tips over to the next hash vector size, so
declare that to prevent rehashing.
* libguile/srcprop.c (scm_init_srcprop): Don't preallocate a big
source_whash table, as we might not need it (if everything is
compiled, for example).
* module/ice-9/boot-9.scm (make-module): Don't preall-cate big hash
tables for imported bindings. Instead trust that resizing works
correctly.
* libguile/fports.c (scm_revealed_count, scm_port_revealed)
(scm_set_port_revealed_x, scm_adjust_port_revealed_x): Move these APIs
here, and only operate on fports. To keep revealed ports alive, now
we will just keep them in a data structure that the GC knows about --
a static list.
* libguile/fports.h: Add revealed count to scm_t_fport, and move decls
of revealed-count functions here.
* libguile/ports.h:
* libguile/ports.c: Adapt to change. Remove SCM_REVEALED and
SCM_SETREVEALED; since they only apply to fports now, keeping them
around would be inviting type errors.
(finalize_port): We don't need to worry about resuscitating ports
here.
* libguile/init.c: Use the scm_set_port_revealed_x function to set the
revealed counts on stream ports.
* libguile/bytevectors.c (STRING_TO_UTF, scm_string_to_utf8)
(UTF_TO_STRING):
* libguile/ports.c (open_iconv_descriptors, close_iconv_descriptors):
* libguile/strings.c (scm_from_stringn, scm_to_stringn): Wrap operations
that acquire and destroy iconv contexts with a mutex. While iconv is
threadsafe, internally it uses a lock, and we need to make sure when
we fork() that no one has that lock -- so we surround it with another
one. Gross.
* libguile/async.c:
* libguile/deprecation.c:
* libguile/fluids.c:
* libguile/gc.c:
* libguile/instructions.c:
* libguile/ports.c:
* libguile/posix.c:
* libguile/strings.c:
* libguile/threads.c: Use the SCM_PTHREAD_ATFORK_LOCK_STATIC_MUTEX
mechanism to lock a number of static mutexen.
* libguile/null-threads.h:
* libguile/pthread-threads.h (SCM_DEFINE_ATFORK_HANDLERS_FOR_MUTEX)
(scm_i_pthread_atfork): Define a pthread_atfork shim, and a helper to
declare procedures to lock and unlock a mutex.
* libguile/threads.h (SCM_PTHREAD_ATFORK_LOCK_STATIC_MUTEX): New macro,
causes the snarfer to emit a pthread_atfork call to lock a mutex, if
threads are enabled.
* libguile/weak-set.c (make_weak_set):
* libguile/weak-table.c (make_weak_table): If we have a threaded Guile,
* keep a weak set (table) of weak sets (tables). Use this and the
pthread_atfork mechanism to lock and unlock weak sets and weak tables
during a fork().
* libguile/weak-set.h (scm_weak_set_prehistory): New internal API.
* libguile/init.c: Add call to scm_weak_set_prehistory().
* libguile/read.c (scm_read_number): Set source properties on
non-immediate numbers if the 'positions' reader option is set.
* doc/ref/api-debug.texi (Source Properties): Update manual.
* libguile/srcprop.c (scm_source_properties, scm_source_property,
scm_i_has_source_properties): Relax validation to allow _any_ object
to be queried for source properties.
* libguile/fports.c (close_the_fd, fport_close): Arrange to always close
the fd, even if the flush procedure throws an exception. Perhaps the
port machinery should do this for us, though. Don't wrap the close
call in SCM_SYSCALL, EINTR leaves the fd in an unspecified state.
Don't bother freeing buffers, the collector will handle that; simply
drop references via scm_port_non_buffer.
* libguile/ports.c (do_free, finalize_port): Catch exceptions caused by
the free procedure. Don't bother setting the stream to 0 at all.
(scm_close_port): Ensure that exceptions thrown by the "close"
procedure don't prevent the port from being marked as closed.
* libguile/strings.c (scm_from_utf8_stringn): Embarassingly, my
scm_from_utf8_stringn implementation was buggy for non-ascii
characters, since October (41d1d984). Fixed. Will be tested with the
next patch.
* libguile/ports.h (scm_t_port_type_flags, scm_t_ptob_descriptor): Add
flags to ptob descriptors.
* libguile/ports.c (scm_set_port_flush): Set the SCM_PORT_TYPE_HAS_FLUSH
flag here.
(scm_c_make_port_with_encoding): Only add ports to the table if
SCM_PORT_TYPE_HAS_FLUSH is set. Only add finalizers to ports if there
is a free function.
(scm_close_port): Inline scm_i_remove_port here. Only remove from the
weak set if SCM_PORT_TYPE_HAS_FLUSH is set.
(scm_set_port_revealed_x): Add a comment.
* libguile/ports.h (struct scm_t_port): Add a flag for the port encoding
mode: UTF8 or iconv. The iconv descriptors are now in a separate
structure so that we can avoid attaching finalizers to the ports
themselves, in some cases.
* libguile/ports.c (scm_c_make_port_with_encoding): Init the encoding
mode.
(scm_i_remove_port): Adapt to call close_iconv_descriptors.
(finalize_iconv_descriptors, open_iconv_descriptors):
(close_iconv_descriptors): New infrastructure to manage iconv
descriptors.
(scm_i_port_iconv_descriptors): New internal helper.
(scm_i_set_port_encoding_x): Use open_iconv_descriptors, if needed.
(get_iconv_codepoint): Use pt->iconv_descriptors.
(get_codepoint): Check the port encoding mode flags.
* libguile/print.c (display_string_using_iconv): Use
scm_i_port_iconv_descriptors.
(display_string): Use pt->encoding_mode flag.
* libguile/strports.c (st_fill_input): Rename from stfill_buffer, and
remove an unneeded scm_return_first_int.
(st_resize_port): Minor variable renaming.
(st_write): Keep read_pos updated to be the same as write_pos. If we
need to resize, do so only once.
(st_seek): No more need to flush.
(st_truncate): Update read_pos here too.
(scm_mkstrport): No need to flush here.
(scm_strport_to_string): Just call scm_from_stringn; rely on it to
detect the latin1 case.
(scm_make_stptob): No more flush function.
* libguile/weak-set.c (resize_set):
* libguile/weak-table.c (resize_table): Drop the set/table lock while
allocating the new vector. Fixes a bug in which a finalizer could
recursively try to grab the port table lock.
* libguile/read.c (scm_read_array): New internal helper that
calls scm_i_read_array and sets its source property if the
'positions' reader option is set.
(scm_read_string): Set source properties on strings if the 'positions'
reader option is set.
(scm_read_vector, scm_read_srfi4_vector, scm_read_bytevector,
scm_read_guile_bitvector, scm_read_sharp): Add new arguments for the
'line' and 'column' of the first character of the datum being read.
Set source properties if the 'positions' reader option is set.
(scm_read_expression): Pass 'line' and 'column' to scm_read_sharp.
* doc/ref/api-debug.texi (Source Properties): Update manual.
* libguile/read.c (scm_read_string): Return a freshly allocated string
every time, even for empty strings. The motivation is to allow source
properties to be added to all strings. Previously, the shared global
'scm_nullstr' was returned for empty strings. Note that empty strings
still share a common global 'null_stringbuf'.
* test-suite/tests/srfi-13.test (substring/shared): Fix tests to reflect
the fact that empty string literals are no longer guaranteed to be
'eq?' to each other.
* libguile/strings.c (scm_i_substring_copy): When asked to create an
empty substring, use 'scm_i_make_string' to make use of its
optimization for empty strings that reuses the global null_stringbuf.
* libguile/gc-malloc.c (do_realloc, do_calloc, do_gc_malloc)
(do_gc_malloc_atomic, do_gc_realloc, do_gc_free): Wrap the allocators
in helper methods. The non-GC allocators call
scm_gc_register_allocation, and the GC allocators ensure that
gc_malloc returns a nonzero value.
(scm_realloc, scm_calloc): Allow NULL return for malloc(0). It is
permitted by POSIX. Don't wrap in a SCM_SYSCALL; the only possible
errno is ENOMEM.
(scm_gc_malloc_pointerless, scm_gc_malloc, scm_gc_calloc):
(scm_gc_realloc, scm_gc_free, scm_gc_strndump): Use the wrappers.
Don't call the debug-malloc registrations and deregistrations, as it
is neither mandatory nor advised to call scm_gc_free. Fixes
compilation with debug-malloc.
(scm_must_free): Don't scm_malloc_unregister.
* libguile/control.c:
* libguile/continuations.c: Fix for the case in which SCM_ALIGNED is not
defined. Though I wonder, perhaps we should just error out in those
cases.
* configure.ac: Remove checks for <langinfo.h> and <nl_types.h>, and
`nl_langinfo'.
* libguile/i18n.c: Remove #ifdefs for HAVE_LANGINFO_H, HAVE_NL_TYPES_H,
HAVE_NL_ITEM, and HAVE_LANGINFO_CODESET.
(SCM_VALIDATE_OPTIONAL_LOCALE_COPY): Use `SCM_UNBNDP'.
* m4/gnulib-cache.m4 (gl_MODULES): Add `nl_langinfo'.
* module/ice-9/i18n.scm (define-vector-langinfo-mapping): Remove
`defaults' parameter; assume (provided? 'nl-langinfo) is always true.
Update users accordingly.
(define-simple-langinfo-mapping): Likewise.
(define-monetary-langinfo-mapping): Likewise, but do not assume
LOCAL-ITEM and INTL-ITEM are always defined.
This should help with regex portability, as reported in
<http://bugs.gnu.org/10684> for Darwin 8.11.
* m4/gnulib-cache.m4 (gl_MODULES): Add `regex'.
* configure.ac: Remove header checks for regex.h, rxposix.h, and
rx/rxposix.h. Remove check for the `regcomp' function. Remove
definition of `HAVE_REGCOMP'. Define `ENABLE_REGEX'.
* libguile/init.c: Check for `ENABLE_REGEX' instead of `HAVE_REGCOMP'.
* libguile/regex-posix.c: Always include <regex.h>. Remove #ifdefs for
rxposix.h and co.
* doc/ref/posix.texi (Network Databases): Update description of
`EAI_NODATA' to mention that Darwin provides it.
* libguile/net_db.c (scm_getaddrinfo): Likewise.
* test-suite/tests/net-db.test ("getaddrinfo")["wrong service name"]:
Accept `EAI_NODATA' too. Reported by David Fang <fang@csl.cornell.edu>,
see <http://bugs.gnu.org/10684>.
* libguile/load.c (search_path): When the provided 'filename' is an
absolute pathname, perform a 'stat' on that pathname to fill the
'stat_buf'. Previously, 'stat_buf' was left uninitialized in this
case, even though 'scm_primitive_load_path' assumes that 'stat_buf'
will be filled. Update the header comment to explicitly specify that
'stat_buf' will be filled. Also 'goto end' in a few failure cases
instead of replicating its code.
* libguile/foreign.h: Remove comment about the finalizer bit, as I don't
think that is the case any more.
* test-suite/tests/gc.test: Ignore flakiness in the gc-modules test.
