changelog.0

gcc-2.95.3 Linux下最常用的C编译器

	New functions that are "safe" variants of convert(),
	to catch errors that ffecom_expr_intrinsic_() now
	no longer catches.
	(ffecom_arglist_expr_): Ensure arguments are not
	converted to narrower types.
	(ffecom_call_): Ensure return value is not converted
	to a wider type.
	(ffecom_char_args_): Use new ffeintrin_gfrt_direct()
	routine.
	(ffecom_expr_intrinsic_): Simplify how run-time
	routine is selected (via `gfrt' only now; lose the
	redundant `ix' variable).
	Eliminate the `library' label; any code that doesn't
	return directly just `break's out now with `gfrt'
	set appropriately.
	Set `gfrt' to default choice initially, either a
	fast direct form or, if not available, a slower
	indirect-callable form.
	(ffecom_make_gfrt_): No longer need to do special
	check for complex; it's built into the new return-type
	regime.
	(ffecom_ptr_to_expr): Use new ffeintrin_gfrt_indirect()
	routine.
	* intrin.c, intrin.h: `gfrt' field replaced with three fields,
	so it is easier to provide faster direct-callable and
	GNU-convention indirect-callable routines in the future.
	DEFIMP macro adjusted accordingly, along with all its uses.
	(ffeintrin_gfrt_direct): New function.
	(ffeintrin_gfrt_indirect): Ditto.
	(ffeintrin_is_actualarg): If `-fno-f2c' is in effect,
	require a GNU-callable version of intrinsic instead of
	an f2c-callable version, so indirect calling is still checked.
	* intrin.def: Replace one GFRT field with the three new fields,
	as appropriate for each DEFIMP intrinsic.

	* com.c (ffecom_stabilize_aggregate_,
	ffecom_convert_to_complex_): Make these `static'.

Thu Aug  7 11:24:34 1997  Craig Burley  <[e-mail]>

	Provide means for front end to determine actual
	"standard" return type for an intrinsic if it is
	passed as an actual argument:
	* com.h, com.c (ffecom_gfrt_basictype,
	ffecom_gfrt_kindtype): New functions.
	(ffecom_gfrt_kind_type_): Replaced with new function.
	All callers updated.
	(ffecom_make_gfrt_): No longer need do anything
	with kind type.

	* intrin.c (ffeintrin_basictype, ffeintrin_kindtype):
	Now returns correct type info for specific intrinsic
	(based on type of run-time-library implementation).

Wed Aug  6 23:08:46 1997  Craig Burley  <[e-mail]>

	* global.c (ffeglobal_ref_progunit_): Don't reset
	number of arguments just due to new type info,
	so useful warnings can be issued.

1997-08-06  Dave Love  <[e-mail]>

	* intrin.def: Fix IDATE_vxt argument order.
	* intdoc.h: Likewise.

Thu Jul 31 22:22:03 1997  Craig Burley  <[e-mail]>

	* global.c (ffeglobal_proc_ref_arg): If REF/DESCR
	disagreement, DESCR is CHARACTER, and types disagree,
	pretend the argsummary agrees so the message ends up
	being about type disagreement.
	(ffeglobal_proc_def_arg): Ditto.

	* expr.c (ffeexpr_token_first_rhs_3_): Set info for LABTOK
	to NONE of everything, to avoid misdiagnosing filewide
	usage of alternate returns.

Sun Jul 20 23:07:47 1997  Craig Burley  <[e-mail]>

	* com.c (ffecom_sym_transform_): If type gets set
	to error_mark_node, just return that for transformed symbol.
	(ffecom_member_phase2_): If type gets set to error_mark_node,
	just return.
	(ffecom_check_size_overflow_): Add `dummy' argument to
	flag that type is for a dummy, update all callers.

Sun Jul 13 17:40:53 1997  Craig Burley  <[e-mail]>

	Fix 970712-1.f:
	* where.c (ffewhere_set_from_track): If start point
	is too large, just use initial start point.  0.6 should
	fix all this properly.

	Fix 970712-2.f:
	* com.c (ffecom_sym_transform_): Preserve error_mark_node for type.
	(ffecom_type_localvar_): Ditto.
	(ffecom_sym_transform_): If type is error_mark_node,
	don't error-check decl size, because back end responds by
	setting that to an integer 0 instead of error_mark_node.
	(ffecom_transform_common_): Same as earlier fix to _transform_
	in that size is checked by dividing BITS_PER_UNIT instead of
	multiplying.
	(ffecom_transform_equiv_): Ditto.

	Fix 970712-3.f:
	* stb.c (ffestb_R10014_): Fix flaky fall-through in error
	test for FFELEX_typeCONCAT by just replicating the code,
	and do FFELEX_typeCOLONCOLON while at it.

1997-07-07  Dave Love  <[e-mail]>

	* intdoc.h: Add various missing pieces; correct GMTIME, LTIME
	result ordering.

	* intrin.def, com-rt.def: Add alarm.

	* com.c (ffecom_expr_intrinsic_): Add case for alarm.

Thu Jun 26 04:19:40 1997  Craig Burley  <[e-mail]>

	Fix 970302-3.f:
	* com.c (ffecom_sym_transform_): For sanity-check compare
	of gbe size of local variable to g77 expectation,
	use varasm.c/assemble_variable technique of dividing
	BITS_PER_UNIT out of gbe info instead of multiplying
	g77 info up, to avoid crash when size in bytes is very
	large, and overflows an `int' or similar when multiplied.

	Fix 970626-2.f:
	* com.c (ffecom_finish_symbol_transform_): Don't bother
	transforming a dummy argument, to avoid a crash.
	* ste.c (ffeste_R1227): Don't return a value if the
	result decl, or its type, is error_mark_node.

	Fix 970626-4.f:
	* lex.c (ffelex_splice_tokens): `-fdollar-ok' is
	irrelevant to whether a DOLLAR token should be made
	from an initial character of `$'.

	Fix 970626-6.f:
	* stb.c (ffestb_do3_): DO iteration variable is an
	lhs, not rhs, expression.

	Fix 970626-7.f and 970626-8.f:
	* expr.c (ffeexpr_cb_comma_i_1_): Set IMPDO expression
	to have clean info, because undefined rank, for example,
	caused crash on mangled source on UltraSPARC but not
	on Alpha for a series of weird reasons.
	(ffeexpr_cb_close_paren_): If not CLOSE_PAREN, push
	opANY expression onto stack instead of attempting
	to mimic what program might have wanted.
	(ffeexpr_cb_close_paren_): Don't wrap opPAREN around
	opIMPDO, just warn that it's gratuitous.
	* bad.def (FFEBAD_IMPDO_PAREN): New warning.

	Fix 970626-9.f:
	* expr.c (ffeexpr_declare_parenthesized_): Must shut down
	parsing in kindANY case, otherwise the parsing engine might
	decide there's an ambiguity.
	(ffeexpr_token_name_rhs_): Eliminate parentypeSUBROUTINE_
	case, so we crash right away if it comes through.
	* st.c, st.h, sta.c, sta.h (ffest_shutdown, ffesta_shutdown):
	New functions.

Tue Jun 24 19:47:29 1997  Craig Burley  <[e-mail]>

	* com.c (ffecom_check_size_overflow_): New function
	catches some cases of the size of a type getting
	too large.  varasm.c must catch the rest.
	(ffecom_sym_transform_): Use new function.
	(ffecom_type_localvar_): Ditto.

Mon Jun 23 01:09:28 1997  Craig Burley  <[e-mail]>

	* global.c (ffeglobal_proc_def_arg): Fix comparison
	of argno to #args.
	(ffeglobal_proc_ref_arg): Ditto.

	* lang-options.h, top.c: Rename `-fdebug' to `-fxyzzy',
	since it's an unsupported internals option and some
	poor user might guess that it does something.

	* bad.def: Make a warning for each filewide diagnostic.
	Put all filewides together.
	* com.c (ffecom_sym_transform_): Don't substitute
	known global tree for global entities when `-fno-globals'.
	* global.c (ffeglobal_new_progunit_): Don't produce
	fatal diagnostics about globals when `-fno-globals'.
	Instead, produce equivalent warning when `-Wglobals'.
	(ffeglobal_proc_ref_arg): Ditto.
	(ffeglobal_proc_ref_nargs): Ditto.
	(ffeglobal_ref_progunit_): Ditto.
	* lang-options.h, top.c, top.h: New `-fno-globals' option.

Sat Jun 21 12:32:54 1997  Craig Burley  <[e-mail]>

	* expr.c (ffeexpr_fulfill_call_): Set array variable
	to avoid warning about uninitialized variable.

	* Make-lang.in: Get rid of any setting of HOST_* macros,
	since these will break gcc's build!
	* makefile: New file to make building derived files
	easier.

Thu Jun 19 18:19:28 1997  Craig Burley  <[e-mail]>

	* g77.c (main): Install Emilio Lopes' patch to support
	Ratfor, and to fix the printing of the version string
	to go to stderr, not stdout.
	* lang-specs.h: Install Emilio Lopes' patch to support
	Ratfor, and patch the result to support picking up
	`*f771' from the `specs' file.

Thu Jun 12 14:36:25 1997  Craig Burley  <[e-mail]>

	* storag.c (ffestorag_update_init, ffestorag_update_save):
	Also update parent, in case equivalence processing
	has already eliminated pointers to it via the
	local equivalence info.

Tue Jun 10 14:08:26 1997  Craig Burley  <[e-mail]>

	* intdoc.c: Add cross-reference to end of description
	of any generic intrinsic pointing to other intrinsics
	with the same name.

	Warn about explicit type declaration for intrinsic
	that disagrees with invocation:
	* expr.c (ffeexpr_paren_rhs_let_): Preserve type info
	for intrinsic functions.
	(ffeexpr_token_funsubstr_): Ditto.
	* intrin.c (ffeintrin_fulfill_generic): Warn if type
	info of fulfilled intrinsic invocation disagrees with
	explicit type info given symbol.
	(ffeintrin_fulfill_specific): Ditto.
	* stc.c (ffestc_R1208_item): Preserve type info
	for intrinsics.
	(ffestc_R501_item): Ditto.

Mon Jun  9 17:45:44 1997  Craig Burley  <[e-mail]>

	* com.c (ffecom_expr_intrinsic_): Fix several of the
	libU77/libF77-unix handlers to properly convert their
	arguments.

	* com-rt.def (FFECOM_gfrtFSTAT): Append missing "i" to
	arg string.

Fri Jun  6 14:37:30 1997  Craig Burley  <[e-mail]>

	* com.c (ffecom_expr_intrinsic_): Have a case statement
	for every intrinsic implementation, so missing ones
	are caught via gcc warnings.
	Don't call ffeintrin_codegen_imp anymore.
	* intrin.c (ffeintrin_fulfill_generic): Remove cg_imp
	stuff from here.
	(ffeintrin_codegen_imp): Delete this function.
	* intrin.def, intrin.h: Remove DEFIMQ stuff from here
	as well.

Thu Jun  5 13:03:07 1997  Craig Burley  <[e-mail]>

	* top.c (ffe_decode_option): New -fbadu77-intrinsics-*
	options.
	* top.h: Ditto.
	* intrin.h: New BADU77 family.
	* intrin.c (ffeintrin_state_family): Ditto.

	Implement new scheme to track intrinsic names vs. forms:
	* intrin.c (ffeintrin_fulfill_generic),
	(ffeintrin_fulfill_specific), (ffeintrin_is_intrinsic),
	intrin.def: The documented name is now either in the
	generic info or, if no generic, in the specific info.
	For a generic, the specific info contains merely the
	distinguishing form (usually "function" or "subroutine"),
	used for diagnostics about ambiguous references and
	in the documentation.

	* intrin.def: Clean up formatting of DEFNAME block.
	Convert many libU77 intrinsics into generics that
	support both subroutine and function forms.
	Put the function forms of side-effect routines into
	the new BADU77 family.
	Make MCLOCK and TIME return INTEGER*4 again, and add
	INTEGER*8 equivalents called MCLOCK8 and TIME8.
	Fix up more status return values to be written and
	insist on them being I1 as well.
	* com.c (ffecom_expr_intrinsic_): Lots of changes to
	support new libU77 intrinsic interfaces.

Mon Jun  2 00:37:53 1997  Craig Burley  <[e-mail]>

	* com.c (ffecom_init_0): Pointer type is now INTEGER(KIND=7),
	not INTEGER(KIND=0), since we want to reserve KIND=0 for
	future use.

Thu May 29 14:30:33 1997  Craig Burley  <[e-mail]>

	Fix bugs preventing CTIME(I*4) from working correctly:
	* com.c (ffecom_char_args_): For FUNCREF case, process
	args to intrinsic just as they would be in
	ffecom_expr_intrinsic_.
	* com-rt.def (FFECOM_gfrtCTIME, FFECOM_gfrtTTYNAM): Fix
	argument decls to specify `&'.

Wed May 28 22:19:49 1997  Craig Burley  <[e-mail]>

	Fix gratuitous warnings exposed by dophot aka 970528-1:
	* global.c (ffeglobal_proc_def_arg, ffeglobal_proc_ref_arg):
	Support distinct function/subroutine arguments instead of
	just procedures.
	* global.h: Ditto.
	* expr.c (ffeexpr_fulfill_call_): A SYMTER with kindNONE
	also is a procedure (either function or subroutine).

Mon May 26 20:25:31 1997  Craig Burley  <[e-mail]>

	* bad.def: Have several lexer diagnostics refer to
	documentation for people who need more info on what Fortran
	source code is supposed to look like.

	* expr.c (ffeexpr_reduced_bool1_), bad.def: New diagnostics
	specific to .NOT. now mention only one operand instead
	of two.

	* g77.c: Recognize -fsyntax-only, similar to -c etc.
	(lookup_option): Fix bug that prevented non-`--' options
	from being recognized.

Sun May 25 04:29:04 1997  Craig Burley  <[e-mail]>

	* intrin.def (FFEINTRIN_impCTIME): Accept `I*' expression
	for STime instead of requiring `I2'.

Tue May 20 16:14:40 1997  Craig Burley  <[e-mail]>

	* symbol.c (ffesymbol_reference): All references to
	standard intrinsics are considered explicit, so as
	to avoid generating basically useless warnings.
	* intrin.c, intrin.h (ffeintrin_is_standard): Returns TRUE
	if intrinsic is standard.

Sun May 18 21:14:59 1997  Craig Burley  <[e-mail]>

	* com-rt.def: Changed all external names of the
	form `"\([a-z0-9]*\)_' to `"G77_\1_0"' so as to
	allow any name valid as an intrinsic to be used
	as such and as a user-defined external procedure
	name or common block as well.

Thu May  8 13:07:10 1997  Craig Burley  <[e-mail]>

	* expr.c (ffeexpr_cb_end_notloc_): For %VAL, %REF, and
	%DESCR, copy arg info into new node.

Mon May  5 14:42:17 1997  Craig Burley  <[e-mail]>

	From Uwe F. Mayer <[e-mail]>:
	* Make-lang.in (g77-cross): Fix typo in g77.c path.

	From Brian McIlwrath <[e-mail]>:
	* lang-specs.h: Have g77 pick up options from a section
	labeled `*f771' of the `specs' file.

Sat May  3 02:46:08 1997  Craig Burley  <[e-mail]>

	* intrin.def (FFEINTRIN_defSIGNAL): Add optional `Status'
	argument that com.c already expects (per Dave Love).

	More changes to support better tracking of (filewide)
	globals, in particular, the arguments to procedures: