Cluster sizes are now independent of GC trigger values.
GUILE_GC_TRIGGER_n can now specify a relative trigger value:
A negative integer gives fraction of total heap size in percent.
(SCM_GC_TRIGGER_1, SCM_GC_TRIGGER_2): Default values set to -40.
Cluster sizes are now independent of GC trigger values.
GUILE_GC_TRIGGER_n can now specify a relative trigger value:
A negative integer gives fraction of total heap size in percent.
allocating heap. The basic idea is to trigger GC every Nth
allocated cell and grow heap when free list runs out. The scheme
has been extended so that GC isn't triggered until all remaining
cells are used. The implementation is also prepared for
development in the direction of POSIX threads.
* gc.c (SCM_EXPHEAP): In order to grow by a factor of 1.5,
SCM_EXPHEAP should return half of the heap size.
all of Guile basics fits into one segment and there suitable room
for work.
(SCM_EXPHEAP): Now takes an argument. Grow by a factor of 1.5
instead of 2.
(scm_freelist, scm_freelist2): Now of type scm_freelist_t.
Freelists now contains information about object span, cells
collected and amount of cells in heap segments belonging to the
list.
(scm_heap_size, scm_gc_cells_collected): Removed.
numbers.h, objects.c, ramap.c, random.c, unif.c, unif.h: Extensive
rewrite of handling of real and complex numbers.
(SCM_FLOATS, SCM_SINGLES): These #ifdef conditionals have been
removed along with the support for floats. (Float vectors are
still supported.)
* gc.c (scm_freelist2): multi-cell freelists.
(inner_map_free_list): map_free_list, parameterized on ncells.
"nn cells in segment mm" was misleading for ncells > 1; changed to
"objects". still print cells too, though.
(scm_map_free_list): rewritten using inner_map_free_list.
(scm_check_freelist): get freelist as parameter, since now we have
more than one.
(scm_debug_newcell2): multi-cell variants of
scm_debug_newcell.
(scm_gc_for_newcell): take ncells and freelist pointer as
parameters.
(scm_gc_mark): add case for tc7_pws (procedures with setters are
now double cells).
(scm_gc_sweep): don't free the float data, since it's not malloced
anymore.
(init_heap_seg): didn't understand what n_new_objects stood for,
so changed to n_new_cells.
(make_initial_segment): new function, makes an initial segment
according to given ncells.
(scm_init_storage): call make_initial_segment, for ncells={1,2,3}.
eliminating some warnings.
* unif.c, strports.c, print.c, options.c: Fix some warnings on
mis-use of SCM/long
* gc.c, gc.h: Added scm_return_first_int(), and added comment re:
what the scm_return_first* functions do.
name for a Scheme object (now a void*), and SCM as 32 bit word for
storing tags and immediates (now a long int). Introduced
SCM_ASWORD and SCM_ASSCM for conversion. Fixed various dubious
code in the process: arbiter.c (use macros), unif.c (scm_array_p),
displaying the expected type. Use SCM_LISTn in a couple places
instead of scm_cons-ing by hand.
* __scm.h: Added SCM_ASSERT_TYPE macro.
* validate.h, scm_validate.h: Added the former, as a renamed
version of the latter with SCM_ASSERT_TYPE used in
SCM_MAKE_VALIDATE (instead of just SCM_ASSERT)
* Makefile.am: Rename scm_validate.h to validate.h.
* *.c, *.h: Include validate.h, not scm_validate.h (old name's
prefix was superfluous).
regex-posix.c, scmsigs.c, stime.c, strings.c, variable.c, stime.c,
strings.c, variable.c: Added lots of documentation, cleaned up
some existing documentation. Occasionally changed formal params
to match docs. Also folded an #ifdef into the inners of a
primitive instead of having two copies of the primitive
(`get-internal-real-time', from stime.c)
removals from Dirk Hermann.
* alist.c: Rename formals to match the parameter names in the
documentation, updates to documentation. Thanks Dirk Hermann!
* scm_validate.h: Added SCM_VALIDATE_THREAD.
* *.c: Remove SCM_NIMP(X) when it is an extraneous pre-test given
that SCM_FOOP macros all now include SCM_NIMP in their expansion.
This simplifies lots of code, making it far more readable.
GUILE_PROC1. Build guile-procedures.txt, and add that file to
pkgdata_DATA.
* load.c: Added `pkgdata-dir', `site-dir', `library-dir'
primitives.
* guile-doc-snarf.awk: Drop trailing space when no arguments:
e.g., "(foo )" is now "(foo)".
* *.c: moved all the documentation for primitives from
guile-doc/ref/{appendices,posix,scheme}.texi into the source code.
This leaves about half of the primitives undocumented. Also, all
the markup is currently still texinfo. I don't have a problem
with texinfo per se, but the markup is not very descriptive or
accurate.
prototypes.
* eval.c: Make scm_m_mody's 3rd argument be a const char *, not a
char *. ANSI prototypes caught this.
* strorder.c: Use GUILE_PROC1 for the couple SCM_PROC1 expansions
that I missed.
* scm_validate.h: Use SCM_BOOLP for validating bools. Do not
expand macros if SCM_DOCSTRING_SNARF.
function headers and argument checking. Switched SCM_PROC,
SCM_PROC1 macros to be GUILE_PROC, GUILE_PROC1 (may change names
later, but was useful to keep old versions around while migrate)
that has docstrings and argument lists embedded in the GUILE_PROC
macro invocations that expand into a function header. Use lots of
new SCM_VALIDATE_* macros to simplify error checking and reduce
tons of redundancy. This is very similar to what I did for Scwm.
Note that none of the extraction of the docstrings, nor software
engineering checks of Scwm is yet added to Guile. I'll work on
that tomorrow, I expect.
* Makefile.am: Added scm_validate.h to modinclude_HEADERS.
* chars.c: Added docstrings for the primitives defined in here.
* snarf.h: Added GUILE_PROC, GUILE_PROC1. Added
SCM_REGISTER_PROC to be like old SCM_PROC, though old SCM_PROC
still remains for now. Changed naming convention for the s_foo
string name of the primitive to be s_scm_foo for ease of use with
the macro.
* scm_validate.h: Lots of new SCM_VALIDATE macros to simplify
argument checking through guile. Maybe some of these should be
folded into the header file for the types they check, but for now
it was easiest to just stick them all in one place.
* configure.in: add --disable-arrays option, probably temporary.
* the following changes allow guile to be built with the array
"module" omitted. some of this stuff is just tc7 type support,
which wouldn't be needed if uniform array types were converted
to smobs.
* tag.c (scm_utag_bvect ... scm_utag_cvect): don't define unless
HAVE_ARRAYS.
(scm_tag): don't check array types unless HAVE_ARRAYS.
* sort.c (scm_restricted_vector_sort_x, scm_sorted_p):
remove the unused array types.
* (scm_stable_sort, scm_sort): don't support vectors if not
HAVE_ARRAYS. a bit excessive.
* random.c (vector_scale, vector_sum_squares,
scm_random_solid_sphere_x, scm_random_hollow_sphere_x,
scm_random_normal_vector_x): don't define unless HAVE_ARRAYS.
* gh_data.c (makvect, gh_chars2byvect, gh_shorts2svect,
gh_longs2ivect, gh_ulongs2uvect, gh_floats2fvect, gh_doubles2dvect,
gh_uniform_vector_length, gh_uniform_vector_ref):
don't define unless HAVE_ARRAYS.
(gh_scm2chars, gh_scm2shorts, gh_scm2longs, gh_scm2floats,
gh_scm2doubles):
don't check vector types if not HAVE_ARRAYS.
* eq.c (scm_equal_p), eval.c (SCM_CEVAL), print.c (scm_iprin1),
gc.c (scm_gc_mark, scm_gc_sweep), objects.c (scm_class_of):
don't support the array types unless HAVE_ARRAYS is defined.
* tags.h: make nine tc7 types conditional on HAVE_ARRAYS.
* read.c (scm_lreadr): don't check for #* unless HAVE_ARRAYS is
defined (this should use read-hash-extend).
* ramap.c, unif.c: don't check whether ARRAYS is defined.
* vectors.c (scm_vector_set_length_x): moved here from unif.c. call
scm_uniform_element_size if HAVE_ARRAYS.
vectors.h: prototype too.
* unif.c (scm_uniform_element_size): new procedure.
* init.c (scm_boot_guile_1): don't call scm_init_ramap or
scm_init_unif unless HAVE_ARRAYS is defined.
* __scm.h: don't define ARRAYS.
* Makefile.am (EXTRA_libguile_la_SOURCES): unif.c and ramap.c
moved here from libguile_la_SOURCES.
* Makefile.am (ice9_sources): add arrays.scm.
* boot-9.scm: load arrays.scm if 'array is provided.
* arrays.scm: new file with stuff from boot-9.scm.
* socket.c (scm_htons, scm_ntohs, scm_htonl, scm_ntohl): new
functions for network data conversion.
* numbers.c (scm_num2long, scm_num2longlong):
throw out-of-range instead of wrong-type-arg if appropriate.
(scm_iint2str): handle -2^31 correctly.
(scm_num2long): handle -2^31 bignum correctly.
(scm_num2long_long): rewrite the bigdig case: basically copied
from scm_num2long.
numbers.h: (SCM_BITSPERLONGLONG): deleted.
* unif.c (rapr1): use sprintf instead of intprint for unsigned
longs: intprint can't cope with large values.
* numbers.c (scm_num2ulong): check more consistently that the
input is not negative. if it is, throw out-of-range instead of
wrong-type-arg.
* ramap.c (scm_array_fill_int): don't limit fill to INUM for
uvect, ivect or llvect.
Check that fill doesn't overflow short uniform array.
* __scm.h: add another long to the definition of long_long and
ulong_long.
* unif.c (scm_raprin1): use 'l' instead of "long_long" in the
print representation of llvect. read can't handle more than
one character.
(scm_dimensions_to_uniform_array): make "fill" an optional argument
instead of a rest argument.
* tags.h (scm_tc7_llvect): wasn't defined anywhere, so use the free
tag 29 for now.
* __scm.h: don't mention LONGLONGS.
* unif.c, numbers.c, eq.c, gc.c, print.c, eval.c, ramap.c:
replace LONGLONGS with HAVE_LONG_LONGS as set by configure.
* net_db.c (scm_inet_aton): throw errors using the misc-error key
instead of system-error. inet_aton doesn't set errno.
system-error isn't right in gethost either, since it's throwing
the value of h_errno instead of errno. so:
(scm_host_not_found_key, scm_try_again_key,
scm_no_recovery_key, scm_no_data_key): new error keys.
(scm_resolv_error): new procedure, use the new keys.
(scm_gethost): call scm_resolv_error not scm_syserror_msg.
* error.c: (various): use scm_cons instead of scm_listify
to build short lists.
* boot-9.scm (read-hash-extend to set up arrays): add 'l' for
long_long uniform vectors.
* networking.scm (sethostent, setnetent, setprotoent, setservent):
take an optional argument STAYOPEN. default is #f.
* readline.c (scm_init_readline): set rl_readline_name to Guile,
to allow conditionals in .inputrc.
associated with subrs.
(SCM_SUBRNUM, SCM_SUBR_ENTRY, SCM_SUBR_GENERIC, SCM_SUBR_PROPS,
SCM_SUBR_DOC): New macros.
(scm_subr_table): New variable.
(scm_mark_subr_table): New function.
* init.c (scm_boot_guile_1): Call scm_init_subr_table.
* gc.c (scm_gc_mark): Don't mark subr names here.
(scm_igc): Call scm_mark_subr_table.
* snarf.h (SCM_GPROC, SCM_GPROC1): New macros.
* procs.c, procs.h (scm_subr_p): New function (used internally).
* gsubr.c, gsubr.h (scm_make_gsubr_with_generic): New function.
* objects.c, objects.h (scm_primitive_generic): New class.
* objects.h (SCM_CMETHOD_CODE, SCM_CMETHOD_ENV): New macros.
* print.c (scm_iprin1): Print primitive-generics.
* __scm.h (SCM_WTA_DISPATCH_1, SCM_GASSERT1,
SCM_WTA_DISPATCH_2, SCM_GASSERT2): New macros.
* eval.c (SCM_CEVAL, SCM_APPLY): Replace scm_wta -->
SCM_WTA_DISPATCH_1 for scm_cxr's (unary floating point
primitives). NOTE: This means that it is now *required* to use
SCM_GPROC1 when creating float scm_cxr's (float scm_cxr's is an
obscured representation that will be removed in the future anyway,
so backward compatibility is no problem here).
* numbers.c: Converted most numeric primitives (all but bit
comparison operations and bit operations) to dispatch on generic
if args don't match.
* eval.c, eval.h (scm_eval_body): New function.
* objects.c (scm_call_generic_0, scm_call_generic_1,
scm_call_generic_2, scm_call_generic_3, scm_apply_generic): New
functions.
* eval.c (SCM_CEVAL): Apply the cmethod directly after having
called scm_memoize_method instead of doing a second lookup.
* objects.h (scm_memoize_method): Now returns the memoized cmethod.
* procs.c (scm_make_subr_opt): Use scm_sysintern0 instead of
scm_sysintern so that the binding connected with the subr name
isn't cleared when we give set = 0.
code, particularly an unnecessary test (len != size, where len ==
size). (Was this leftovers from debugging code, or have I missed
something profound?)
* weaks.c (scm_make_weak_vector): Add another extra slot before
vector contents, to be used only during garbage collection.
* weaks.h (SCM_WVECT_GC_CHAIN): New macro to access it.
* gc.c (scm_weak_vectors): Now a SCM instead of a SCM*, and now
static.
(scm_weak_size, scm_n_weak): Deleted.
(scm_igc): Use SCM_WVECT_GC_CHAIN to build up a chain of weak
vectors without allocating new storage during GC, using
scm_weak_vectors as the head of the chain.
(scm_mark_weak_vector_spines): Walk SCM_WVECT_GC_CHAIN chain
instead of stepping through an array.
(scm_gc_sweep): Update offset used to find start of weak vector to
free it.
(scm_init_storage): Set scm_weak_vectors to EOL.
Fix from Ken Raeburn <raeburn@raeburn.org>:
* gc.c (already_in_gc): New variable.
(scm_igc): Set and clear already_in_gc; abort if it's set at
entry.
scmsigs.c, strports.c, vports.c: Install the sources which
actually correspond to the changes described below. I got the
ChangeLog entries and the patch from two different places...
it shows up as terrible performance, as we GC constantly.
* gc.c (scm_init_storage): install an atexit proc to flush the
ports.
(cleanup): the new proc. it sets a global variable which can be
checked by the ptob flush procs to avoid trying to throw
exceptions during exit. not very pleasant but it seems more reliable.
* fports.c (local_fflush): check terminating variable and if set
don't throw exception.
* CHECKME: that the atexit proc is installed if unexec used.
* gc.c (scm_gc_mark): Mark struct setter slot.
* gc.c (scm_gc_sweep): Bugfix: Look for SCM_STRUCT_F_LIGHT flag at
scm_struct_i_flags instead of scm_vtable_index_layout!
* gc.c (scm_gc_mark, scm_gc_sweep): Added case labels for
scm_tc7_pws.
