* debug.c (scm_make_iloc): Prefer !SCM_<foo> over SCM_N<foo>.

[bpt/guile.git] / libguile / ChangeLog

2001-10-11  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* debug.c (scm_make_iloc):  Prefer !SCM_<foo> over SCM_N<foo>.

	(scm_memcons, scm_mem_to_proc):  When building lists, prefer
	scm_list_<n> over scm_cons[2]?.

	(scm_mem_to_proc):  Prefer SCM_CONSP over SCM_NIMP.

	(scm_procedure_name):  Use SCM_CADR instead of explicit form.

	(debugobj_print):  Coerce scm_intprint arg 1 to long, not int.
	Thanks to Rob Browning for the patch (see log entry 2001-09-21) -
	for some reason his patch didn't make it into the cvs.

2001-10-09  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* numbers.c (mem2decimal_from_point):  Cleaned up the parsing a
	little bit - should even be somewhat more accurate now.

2001-10-08  Rob Browning  <rlb@defaultvalue.org>

	* gc.c: support ia64 register backing store.
	(SCM_MARK_BACKING_STORE): new macro.

	* continuations.h: support ia64 register backing store.
	(struct scm_t_contregs): add ia64 register backing store.

	* continuations.c: support ia64 register backing store.
	(continuation_mark): mark ia64 register backing store.
	(continuation_free): free ia64 register backing store.
	(scm_make_continuation): capture ia64 register backing store.
	(copy_stack_and_call): copy ia64 register backing store.

2001-10-08  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* hashtab.c (scm_hash_fn_create_handle_x):  The result of assoc_fn
	is known to be #f if no entry is found.  Thus, use !SCM_FALSEP
	instead of SCM_NIMP to test for that case.

	* strings.h (SCM_SET_STRING_LENGTH):  Cast the length to
	scm_t_bits instead of long.

2001-10-06  Marius Vollmer  <mvo@zagadka.ping.de>

	* tags.h (SCM_T_BITS_MAX, SCM_T_SIGNED_BITS_MAX,
	SCM_T_SIGNED_BITS_MIN): New.
	* numbers.h (SCM_MOST_POSITIVE_FIXNUM, SCM_MOST_NEGATIVE_FIXNUM):
	Use them to make these macros computable by the preprocessor.

	* num2integral.i.c (INTEGRAL2NUM): Let the preprocessor test
	whether the integral type fits in a fixnum, not the compiler.
	This removes a spurious compiler warning.  Also, honor the
	NO_PREPRO_MAGIC flag to suppress any preprocessor tests.  This is
	needed for `long long's.

	* numbers.c: Define NO_PREPRO_MAGOC when including
	num2integral.c.i for `long long' and `signed long long'.

2001-10-06  Mikael Djurfeldt  <mdj@linnaeus>

	These changes fixes a race condition in the Guile coop - pthread
	compatibility code.
	
	* coop.c (mother_awake_p): New variable.
	(coop_create): Set mother_awake_p before creating or signalling
	mother; wait until mother is going to sleep before returning.
	(mother): Reset mother_awake_p before going to sleep.

2001-10-05  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* options.c (protected_objects, scm_init_options):  The content of
	protected_objects is now protected from garbage collection using
	scm_gc_register_root instead of scm_permanent_object.

	(get_option_setting):  New static function that	computes an option
	setting as it was formerly done in the function scm_options.

	(get_documented_option_setting):  New static function that
	returns option documentation as it was formerly done in the
	function scm_options.  Note that documentation C strings are no
	longer precomputed into SCM objects.  Instead, they are converted
	into SCM strings every time get_documented_option_setting is
	called.

	(change_option_setting):  New static functions that modifies the
	option setting as it was formerly done in the function
	scm_options.  The function is now exception safe, i. e. won't
	cause a memory leak when interrupted.  Further, only non-immediate
	option values are added to the protection list.

	(scm_options):  This function now has only the purpose to dispatch
	to to get_option_setting, get_documented_option_setting or
	change_option_setting, depending on the arguments given to
	scm_options.

	(scm_init_opts):  Don't convert documentation C strings into SCM
	strings.  Further, don't protect any object values:  They _must_
	be immediate values, otherwise there is no guarantee that they
	have not been collected before anyway.

	* options.[ch] (scm_t_option):  Made type unsigned, name into a
	constant char* and val into a scm_t_bits type.

	(scm_options, scm_init_opts):  The number of options is guaranteed
	to be larger or equal to zero.  Thus, the type is changed to
	unsigned.

2001-10-05  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* num2integral.i.c (NUM2INTEGRAL):  Eliminated some warnings about
	testing an unsigned value for being >= 0.

2001-10-05  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* numbers.h:  Removed old comment about using SCM_CAR to access
	non-pair cells.

	(SCM_MOST_POSITIVE_FIXNUM, SCM_MOST_NEGATIVE_FIXNUM):  Make sure
	the return value is signed.  Thanks to Brian Crowder for the bug
	report.

	(SCM_SRS):  Avoid unnecessary casting and don't unpack input
	values.  With this patch, SCM_SRS can be safely used for other
	types than scm_t_signed_bits.  However, it should still better be
	an internal macro and thus be renamed to SCM_I_SRS.

	(SCM_MAKINUM, SCM_INUM):  Use proper casting.

2001-10-03  Gary Houston  <ghouston@arglist.com>

	* continuations.h, unif.h: in the descriptions of the bit patterns
	of the heap cells, make bit 0 the least significant.

2001-09-25  Thien-Thi Nguyen  <ttn@glug.org>

	* chars.h (SCM_MAKE_CHAR): Use `scm_t_bits' instead of `intptr_t'.
	Thanks to Golubev I. N.

2001-09-25  Gary Houston  <ghouston@arglist.com>

	* ports.c (scm_drain_input): extended the docstring.  thanks to
	Alex Schroeder and Thien-Thi Nguyen.

2001-09-23  Mikael Djurfeldt  <mdj@linnaeus>

	* validate.h (SCM_NUM2FLOAT, SCM_NUM2DOUBLE,
	SCM_VALIDATE_FLOAT_COPY, SCM_VALIDATE_DOUBLE_COPY): New
	macros. (The NUM names might soon change.)

	* numbers.h: Added missing declarations.

2001-09-22  Mikael Djurfeldt  <mdj@linnaeus>

	* Makefile.am: Distribute num2float.i.c.

	* num2float.i.c: New file, multiply included by numbers.c, used
	to "templatize" the float <-> num conversion routines.

	* numbers.c: New functions: scm_num2float, scm_float2num,
	scm_num2double, scm_double2num.

2001-09-21  Rob Browning  <rlb@defaultvalue.org>

	* .cvsignore: really add version.h

	* strings.h (SCM_SET_STRING_LENGTH): coerce "l" to a long.
	Otherwise it fails on the alpha.  However, we might rather choose
	this size conditionally.

	* numbers.c (scm_gcd): change "k" to a long from an int.
	Otherwise it fails on the alpha.  However, we might rather choose
	this size conditionally.

	* error.c (scm_wta): coerce char* to intptr_t before int
	assignment.

	* debug.c (debugobj_print): coerce scm_intprint arg 1 to long, not
	int.

	* chars.h (SCM_MAKE_CHAR): coerce value to intptr_t.

2001-09-20  Mikael Djurfeldt  <mdj@linnaeus>

	* numbers.c (scm_integer_expt): Accept inexact integer in second
	argument.  (Thanks to Bill Schottstaedt.)

2001-09-20  Rob Browning  <rlb@defaultvalue.org>

	* .cvsignore: add version.h

	* versiondat.h.in: removed (obsolete).

	* version.h.in: renamed from version.h.
	(SCM_GUILE_MAJOR_VERSION): new public macro.
	(SCM_GUILE_MINOR_VERSION): new public macro.
	(SCM_GUILE_MICRO_VERSION): new public macro.

	* version.h: renamed to version.h.in.

	* version.c
	(scm_major_version): support integer *_VERSION macros.
	(scm_minor_version): support integer *_VERSION macros.
	(scm_micro_version): support integer *_VERSION macros.
	(scm_version): support integer *_VERSION macros.

2001-09-20  Mikael Djurfeldt  <mdj@linnaeus>

	* error.c, error.h: Made error keys globally accessible.
	Applications might want to test for these or use them in a direct
	call to scm_error.

	* num2integral.i.c (NUM2INTEGRAL): Report an error when these
	routines are passed an inexact.  This change in behavior is
	motivated by concordance with R5RS: It is more common that a
	primitive doesn't want to accept an inexact for an exact.

2001-09-17  Dirk Herrmann  <D.Herrmann@tu-bs.de>

        The following patch partially undoes my patch from 2001-06-30,
	where I added the function scm_gc_mark_cell_conservatively.  The
	function is buggy, since it breaks guile during conservative
	marking if a pointer on the stack points directly into the list of
	free cells on the heap:  With conservative cell marking this will
	cause the whole free list to be scanned and marked - boom!

	* gc.c (allocated_mark, MARK, heap_segment,
	scm_gc_mark_cell_conservatively, scm_init_storage), gc.h
	(scm_gc_mark_cell_conservatively):  Remove function
	scm_gc_mark_cell_conservatively and update the corresponding
	comments and uses accordingly.  Thanks to Christopher Cramer for
	the patch.  (Minor corrections by me.)

2001-09-15  Gary Houston  <ghouston@arglist.com>

	* root.h (scm_root_state): removed the continuation_stack and
	continuation_stack_ptr members, which have no apparent purpose.
	(scm_continuation_stack, scm_continuation_stack_ptr): #defines
	removed.

	* root.c (root_mark), init.c (restart_stack, start_stack), gc
	(scm_igc): remove all references to contination_stack and
	continuation_stack_ptr, avoiding allocation of a vector and
	useless processing during gc.

2001-09-13  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* guardians.c (tconc_t, t_tconc):  Renamed tconc_t to t_tconc.

	(TCONC_IN):  Make sure that the cell word 0 is initialized last.

	(guardians_t, t_guardians):  Renamed guardians_t to t_guardians.

	(GUARDIAN, GUARDIAN_DATA):  Renamed GUARDIAN to GUARDIAN_DATA.

	(guardian_apply, scm_get_one_zombie, scm_make_guardian,
	mark_and_zombify):  Prefer !SCM_<foo> over SCM_N<foo>.

2001-09-13  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* guardians.c (mark_dependencies_in_tconc,
	whine_about_self_centered_zombies, scm_init_guardians):  Register
	the static global variable `self_centered_zombies' via
	scm_gc_register_root, to make some cdr-ing unnecessary.

2001-09-13  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* backtrace.c (display_backtrace_file,
	display_backtrace_file_and_line):  Use SCM_EQ_P when comparing SCM
	values, use SCM_FALSEP when comparing SCM values against #f.
	Thanks to Rob Browning for the bug report.

2001-09-12  Martin Baulig  <martin@home-of-linux.org>

	* strings.[ch] (scm_str2string): New function.

2001-09-06  Marius Vollmer  <mvo@zagadka.ping.de>

	* gc.c (scm_done_free): Always subtract size from scm_mallocated
	when computing nm, even if it's negative.
	(scm_must_malloc): Abort on overflow of scm_mtrigger.
	(scm_must_realloc): Likewise.

2001-09-01  Michael Livshin  <mlivshin@bigfoot.com>

	* numbers.c (scm_sys_check_number_conversions): new function,
	defined if Guile is compiled in debugging mode.  currently checks
	`scm_num2ulong', should check much much more.

	* num2integral.i.c (NUM2INTEGRAL): when converting a bignum to
	unsigned, ensure that it's positive.  thanks to Martin Baulig!

2001-08-31  Dirk Herrmann  <D.Herrmann@tu-bs.de>

	* __scm.h: Added new section about compile time selectable
	features.

	(long_long, ulong_long, scm_sizet, SCM_WNA, SCM_OUTOFRANGE,
	SCM_NALLOC, SCM_HUP_SIGNAL, SCM_INT_SIGNAL, SCM_FPE_SIGNAL,
	SCM_BUS_SIGNAL, SCM_SEGV_SIGNAL, SCM_ALRM_SIGNAL, SCM_GC_SIGNAL,
	SCM_TICK_SIGNAL, SCM_SIG_ORD, SCM_ORD_SIG, SCM_NUM_SIGS):
	Removed.

	* deprecation.c (scm_include_deprecated_features): Simplified.

	* eval.c (EVALCAR, unmemocopy), eval.h (SCM_XEVALCAR): Use