Uros Bizjak [Fri, 14 Dec 2007 15:33:34 +0000 (16:33 +0100)]
sse.md (sse4_2_pcmpestr): Use reg_not_xmm0_operand constraint for operand2.
* config/i386/sse.md (sse4_2_pcmpestr): Use reg_not_xmm0_operand
constraint for operand2. Use nonimm_not_xmm0_operand constraint
for operand 4. Update arguments in the call to
gen_sse4_2_pcmpestr_cconly.
(sse_4_2_pcmpestr_cconly): Renumber insn operands and update insn
template accordingly.
(sse4_2_pcmpistr): Use reg_not_xmm0_operand constraint for operand2.
Use nonimm_not_xmm0_operand constraint for operand 3. Update
arguments in the call to gen_sse4_2_pcmpistr_cconly.
(sse_4_2_pcmpistr_cconly): Renumber insn operands and update insn
testsuite/ChangeLog:
* gcc.target/i386/sse-12.c (dg-options): Use -msse4
instead of -msse4.1.
* gcc.target/i386/sse-12.c (dg-options): Ditto.
* g++.dg/other/i386-2.C (dg-options): Ditto.
Richard Guenther [Fri, 14 Dec 2007 14:21:41 +0000 (14:21 +0000)]
re PR middle-end/34462 (tree check: expected ssa_name, have struct_field_tag in vuses_compare, at tree-vn.c:118)
2007-12-14 Richard Guenther <rguenther@suse.de>
PR middle-end/34462
* tree-ssa-operands.h (create_ssa_artificial_load_stmt): Add
parameter to say whether to unlink immediate uses.
* tree-ssa-operands.c (create_ssa_artificial_load_stmt): Do not
mark the artificial stmt as modified. Unlink immediate uses
only if requested.
* tree-ssa-dom.c (record_equivalences_from_stmt): Update caller.
* tree-ssa-pre.c (insert_fake_stores): Likewise.
* gcc.c-torture/compile/20071214-1.c: New testcase.
Uros Bizjak [Fri, 14 Dec 2007 12:43:57 +0000 (13:43 +0100)]
sse-14.c (test_1, [...]): New macros to test macroized SSE intrinsics.
* gcc.target/i386/sse-14.c (test_1, test_2, test_2x, test_4): New
macros to test macroized SSE intrinsics. Use new macros to test
macroized SSE intrinsics from ammintrin.h, smmintrin.h, tmmintrin.h,
emmintrin.h, xmmintrin.h and bmmintrin.h
* gcc.target/i386/sse-13.c (__builtin_ia32_pcmp?str*128): Redefine
to test with immediate operand.
Thomas Koenig [Thu, 13 Dec 2007 19:35:09 +0000 (19:35 +0000)]
re PR libfortran/34370 (file positioning after nonadvancing i/o)
2007-12-13 Thomas Koenig <tkoenig@gcc.gnu.org>
PR libfortran/34370
PR libfortran/34323
PR libfortran/34405
* io/io.h: Add previous_nonadvancing_write to gfc_unit.
Add prototype for finish_last_advance_record.
* io/file_pos.c (st_backspace): Generate error if backspace is
attempted for direct access or unformatted stream.
If there are bytes left from a previous ADVANCE="no", write
them out before performing the backspace.
(st_endfile): Generate error if endfile is attempted for
direct access.
If there are bytes left from a previous ADVANCE="no", write
them out before performing the endfile.
(st_rewind): Generate error if rewind is attempted for
direct access.
* unit.c (close_unit_1): Move functionality to write
previously written bytes to...
(finish_last_advance_record): ... here.
* transfer.c (data_transfer_init): If reading, reset
previous_nonadvancing_write.
(finalize_transfer): Set the previous_noadvancing_write
flag if we are writing and ADVANCE="no" was specified.
Only call next_record() if advance="no" wasn't specified.
2007-12-13 Thomas Koenig <tkoenig@gcc.gnu.org>
PR libfortran/34370
PR libfortran/34323
PR libfortran/34405
* gfortran.dg/advance_6.f90: New test case.
* gfortran.dg/direct_io_7.f90: New test case.
* gfortran.dg/streamio_13.f90: New test case.
Jakub Jelinek [Thu, 13 Dec 2007 19:05:37 +0000 (20:05 +0100)]
re PR rtl-optimization/32636 (25_algorithms/search_n/iterator.cc: miscompiled on hppa2.0w-hp-hpux11.11)
PR rtl-optimization/32636
* df-scan.c (df_get_entry_block_def_set): Set struct_value_rtx
regno in entry_block_defs even if HAVE_prologue && epilogue_completed.
Richard Guenther [Thu, 13 Dec 2007 14:22:23 +0000 (14:22 +0000)]
re PR middle-end/34450 (compile takes up 1.8 GB RAM at -O1)
2007-12-13 Richard Guenther <rguenther@suse.de>
PR tree-optimization/34450
* params.def (PARAM_SCCVN_MAX_SCC_SIZE): New param.
* invoke.texi (sccvn-max-scc-size): Document.
* Makefile.in (tree-ssa-sccvn.o): Add $(PARAMS_H) dependency.
* tree-ssa-sccvn.h (run_scc_vn): Return true on success, false
on error.
* tree-ssa-sccvn.c (params.h): Include.
(DFS): Return true if all went well, return false as soon as
a SCC exceeds the size of PARAM_SCCVN_MAX_SCC_SIZE.
(run_scc_vn): Return true if all went well, return false if
we aborted during DFS.
* tree-ssa-pre.c (execute_pre): Check if SCCVN finished
successfully, otherwise bail out.
ipa-struct-reorg.c (is_candidate): Print information to dump file when the type is initialized.
2007-12-13 Olga Golovanevsky <olga@il.ibm.com>
* ipa-struct-reorg.c (is_candidate): Print information to dump
file when the type is initialized.
(get_stmt_accesses): Likewise when a structure has bitfields
or field access is too complicate.
(safe_cond_expr_check): Likewise for unsafe condition expressions.
(exclude_cold_structs): Likewise for cold structures.
(collect_structures): Suppress redundant print when there are
no structures to transform.
(do_reorg): Print to dump file a number of structures that
will be transformed.
(dump_new_types): Print a number of new types to replace
an original structure type.
Bob Duff [Thu, 13 Dec 2007 11:33:49 +0000 (11:33 +0000)]
(Attribute_to_gnu): Check for violations of the No_Implicit_Dynamic_Code restriction.
(Attribute_to_gnu): Check for violations of the No_Implicit_Dynamic_Code
restriction. This checking used to be done in the front end, but is
moved here so we can make it more accurate.
Javier Miranda [Thu, 13 Dec 2007 10:48:09 +0000 (11:48 +0100)]
sem_ch3.ads, [...] (Check_Abstract_Overriding): Avoid generation of spurious error if parent is an interface type...
2007-12-06 Javier Miranda <miranda@adacore.com>
Ed Schonberg <schonberg@adacore.com>
* sem_ch3.ads, sem_ch3.adb (Check_Abstract_Overriding): Avoid
generation of spurious error if parent is an interface type; caused
because predefined primitive bodies will be generated later by
Freeze_Record_Type.
(Process_Subtype): The subtype inherits the Known_To_Have_Preelab_Init
flag.
(Derive_Subprograms): Handle derivations of predefined primitives
after all the user-defined primitives to ensure that they are
found in proper order in instantiations.
(Add_Interface_Tag_Components, Inherit_Components): Update occurrences
of Related_Interface to Related_Type.
(Record_Type_Declaration): Minor reordering of calls to decorate the
Tag component because the entity must have set its Ekind attribute
before setting its Is_Tag attribute.
(Analyze_Subtype_Declaration): In the case of subtypes with
Private_Kind, inherit Known_To_Have_Preelab_Init from the parent.
Gary Dismukes [Thu, 13 Dec 2007 10:45:44 +0000 (11:45 +0100)]
make.adb (Scan_Make_Arg): Add test for -aamp_target switch...
2007-12-06 Gary Dismukes <dismukes@adacore.com>
Arnaud Charlet <charlet@adacore.com>
* make.adb (Scan_Make_Arg): Add test for -aamp_target switch, passing
it to the front end and setting the aamp_target environment variable
to the switch's argument to ensure that gnaampbind and gnaamplink will
take the specified library into account.
(Make): Only set Check_Object_Consistency to False for JVM, not for CIL
target, since the CIL compiler supports an "object" file (.il files).
Bob Duff [Thu, 13 Dec 2007 10:45:14 +0000 (11:45 +0100)]
itypes.ads, itypes.adb (Create_Itype): For access-to-subprogram types...
2007-12-06 Bob Duff <duff@adacore.com>
* itypes.ads, itypes.adb (Create_Itype): For access-to-subprogram
types, set Can_Use_Internal_Rep appropriately, based on
Always_Compatible_Rep_On_Target.
Robert Dewar [Thu, 13 Dec 2007 10:42:14 +0000 (11:42 +0100)]
gnat_ugn.texi: Dcoument new rules for style check comment alignment Document that suffixes may...
2007-12-06 Robert Dewar <dewar@adacore.com>
Bob Duff <duff@adacore.com>
* gnat_ugn.texi: Dcoument new rules for style check comment alignment
Document that suffixes may be terminations of each others
Add doc for -gnatw.a and -gnatw.A
Document gnatbind -Wx switch
Document BOM recognition
Document pragma Implemented_By_Entry.
Document new units.
* gnat_rm.texi: (Favor_Top_Level): Document new pragma.
Add doc for pragma Unreferenced suppressing no entities referenced msg
Add documentation of GNAT.Directory_Operations.Iteration
Add documentation of GNAT.Random_Numbers
Add documentation for pragma Shared.
Correct documentation for Bit_Order
Add documentation for the Pool_Address attribute.
Fix and improve documentation of pragma machine_attribute.
New pragma Fast_Math
Document BOM recognition
* vms_data.ads: Add entries for -gnatw.a -gnatw.A
Add /Wide_Character_Encoding for binder
Add qualifier for the new gnatpp option --no-separate-loop-then
* a-ngcoar.adb: Fix typo.
* s-interr.adb: Minor reformatting
* env.c: Minor reformatting.
* g-bytswa.adb: Minor reformatting.
* g-rannum.ads: Minor documentation improvements
* s-tasinf-mingw.adb: Minor header fix
* a-clrefi.adb: Minor reformatting
* g-sttsne.ads: Minor documentation improvement
* g-sttsne-locking.ads: Minor documentation improvement
* g-soliop-solaris.ads: Minor documentation improvement
* g-soliop-mingw.ads: Minor documentation improvement
* g-soliop.ads: Minor documentation improvement
* exp_aggr.ads: Minor reformatting
* debug.adb: Add documentation for the gprbuild debug flags
* exp_ch2.adb: Use Nkind_In to simplify code throughout
* exp_pakd.adb: Minor reformatting
* g-altive.ads, g-alleve.adb: Remove assertions.
Add comment about minor differences between targets regarding
floating-point operations.
* g-thread.adb: Remove pragma unreferenced.
* lib.ads: Minor reformatting
* par-ch9.adb: Minor reformatting of error messages
* sem_case.adb: Minor reformatting
* s-fileio.adb: Minor reformattinng
* s-vmexta.ads: Minor typo
* vxaddr2line.adb:
Take into account 'Success' value as per new GNAT warning.
Olivier Hainque [Thu, 13 Dec 2007 10:36:42 +0000 (11:36 +0100)]
tb-alvms.c (struct tb_entry_t, [...]): Revert back to use of Procedure Value instead of Frame Pointer as the...
2007-12-06 Olivier Hainque <hainque@adacore.com>
* tb-alvms.c (struct tb_entry_t, __gnat_backtrace): Revert back to use
of Procedure Value instead of Frame Pointer as the invocation
identifier associated with the instruction pointer in each traceback
entry.
* g-trasym-vms-alpha.adb (Traceback_Entry, PV_For, FP_For,
TB_Entry_For): Revert back to use of Procedure Value instead of Frame
Pointer as the invocation identifier passed to tbk$symbolize.
* s-traent-vms.ads, s-traent-vms.adb
(Traceback_Entry, PV_For, FP_For, TB_Entry_For): Revert back to use of
Procedure Value instead of Frame Pointer as the invocation identifier
passed to tbk$symbolize.
Olivier Hainque [Thu, 13 Dec 2007 10:34:35 +0000 (11:34 +0100)]
s-stausa.ads (Stack_Analyzer): Remove First_Is_Topmost, redundant with Stack_Grows_Down in System.Parameters.
2007-12-06 Olivier Hainque <hainque@adacore.com>
* s-stausa.ads (Stack_Analyzer): Remove First_Is_Topmost, redundant
with Stack_Grows_Down in System.Parameters. Rename Array_Address into
Stack_Overlay_Address and document that we are using an internal
abstraction.
(Byte_Size, Unsigned_32_Size): Remove, now useless.
(Pattern_Type, Bytes_Per_Pattern): New subtype and constant, to be used
consistently throughout the various implementation pieces.
* s-stausa.adb (Stack_Slots): New type, abstraction for the stack
overlay we are using to fill the stack area with patterns.
(Top_Slot_Index_In, Bottom_Slot_Index_In): Operations on Stack_Slots.
(Push_Index_Step_For, Pop_Index_Step_For): Likewise.
(Fill_Stack, Compute_Result): Use the Stack_Slots abstraction.
Bob Duff [Thu, 13 Dec 2007 10:32:45 +0000 (11:32 +0100)]
sem_util.ads, [...] (Is_Concurrent_Interface): New routine.
2007-12-06 Bob Duff <duff@adacore.com>
Javier Miranda <miranda@adacore.com>
Robert Dewar <dewar@adacore.com>
* sem_util.ads, sem_util.adb (Is_Concurrent_Interface): New routine.
(Set_Convention): New procedure to set the Convention flag, and in
addition make sure the Favor_Top_Level flag is kept in sync (all
foreign-language conventions require Favor_Top_Level = True).
(Collect_Abstract_Interfaces): Update occurrences of Related_Interface
to Related_Type.
(Collect_Interfaces_Info): Minor update to handle the two secondary
dispatch tables. Update occurrence of Related_Interface to Related_Type.
(Generate_Parent_Ref): Add parameter to specify entity to check
(Is_Preelaborable_Expression): Allow the name of a discriminant to
initialize a component of a type with preelaborable initialization.
This includes the case of a discriminal used in such a context.
(Is_Dependent_Component_Of_Mutable_Object): Take into account the
latest Ada 2005 rules about renaming and 'Access of
discriminant-dependent components.
(Check_Nested_Access): Add handling when there are no enclosing
subprograms (e.g. case of a package body).
(Find_Parameter_Type): Factor routine from several other compiler files.
Remove routine from Find_Overridden_Synchronized_Primitive.
Ed Schonberg [Thu, 13 Dec 2007 10:32:11 +0000 (11:32 +0100)]
sem_disp.adb (Check_Dispatching_Call): If an actual in a call to an inherited operation is a defaulted...
2007-12-06 Ed Schonberg <schonberg@adacore.com>
* sem_disp.adb (Check_Dispatching_Call): If an actual in a call to an
inherited operation is a defaulted tag-indeterminate call, and there is
a statically tagged actual, use the static tag as a controlling actual
for the defaulted actual.
sem_ch9.adb (Analyze_Requeue): Add a local flag to capture whether a requeue statement is dispatching.
2007-12-06 Hristian Kirtchev <kirtchev@adacore.com>
Gary Dismukes <dismukes@adacore.com>
* sem_ch9.adb (Analyze_Requeue): Add a local flag to capture whether a
requeue statement is dispatching. Do not emit an error when the name is
not an entry and the context is a dispatching select. Add code to
perform subtype conformance between the formals of the current entry
and those of the target interface primitive.
(Analyze_Asynchronous_Select, Analyze_Conditional_Entry_Call, Analyze_
Timed_Entry_Call): Analyze the triggering statement as the first step of
the processing. If this is a dispatching select, postpone the analysis
of all select statements until the Expander transforms the select. This
approach avoids generating duplicate identifiers after the Expander has
replicated some of the select statements. In case the Expander is
disabled, perform regular analysis.
(Check_Triggering_Statement): New routine.
(Analyze_Requeue): Exclude any interpretations that are not entries when
checking overloaded names in a requeue. Also test type conformance for
matching interpretations rather than requiring subtype conformance at
that point to conform with the RM's resolution rule for requeues.
Robert Dewar [Thu, 13 Dec 2007 10:30:55 +0000 (11:30 +0100)]
sem_ch8.adb (Analyze_Subprogram_Renaming): Special error message for renaming entry as subprogram using rename-as-body if...
2007-12-06 Robert Dewar <dewar@adacore.com>
Ed Schonberg <schonberg@adacore.com>
* sem_ch8.adb (Analyze_Subprogram_Renaming): Special error message for
renaming entry as subprogram using rename-as-body if subprogram spec
frozen.
(Use_One_Type): The clause is legal on an access type whose designated
type has a limited view.
(Find_Direct_Name): Use Namet.Sp.Is_Bad_Spelling_Of function
(Find_Expanded_Name): Use Namet.Sp.Is_Bad_Spelling_Of function
(Analyze_Renamed_Primitive_Operation): new procedure to determine the
operation denoted by a selected component.
(Analyze_Renamed_Entry): Resolve the prefix of the entry name, because
it can be an expression, possibly overloaded, that returns a task or
an access to one.
sem_ch4.adb (Analyze_Selected_Component): Include the requeue statement to the list of contexts where a selected...
2007-12-06 Hristian Kirtchev <kirtchev@adacore.com>
Ed Schonberg <schonberg@adacore.com>
* sem_ch4.adb (Analyze_Selected_Component): Include the requeue
statement to the list of contexts where a selected component with a
concurrent tagged type prefix should yield a primitive operation.
(Find_Primitive_Operation): Handle case of class-wide types.
(Analyze_Overloaded_Selected_Component): If type of prefix is
class-wide, use visible components of base type.
(Resolve_Selected_Component): Ditto.
(Try_Primitive_Operation, Collect_Generic_Type_Ops): If the type is a
formal of a generic subprogram. find candidate interpretations by
scanning the list of generic formal declarations.:
(Process_Implicit_Dereference_Prefix): If the prefix has an incomplete
type from a limited_with_clause, and the full view is available, use it
for subsequent semantic checks.
(Check_Misspelled_Selector): Use Namet.Sp.Is_Bad_Spelling_Of function
(Find_Primitive_Operation): New function.
(Analyze_Overloaded_Selected_Component): insert explicit dereference
only once if several interpretations of the prefix yield an access type.
(Try_Object_Operation): Code and comment cleanup.
(Analyze_Selected_Component): Reorder local variables. Minot comment and
code reformatting. When the type of the prefix is tagged concurrent, a
correct interpretation might be available in the primitive and
class-wide operations of the type.
Ed Schonberg [Thu, 13 Dec 2007 10:29:52 +0000 (11:29 +0100)]
sem_ch12.adb (Instantiate_Formal_Subprogram): In the subprogram renaming declaration...
2007-12-06 Ed Schonberg <schonberg@adacore.com>
* sem_ch12.adb (Instantiate_Formal_Subprogram): In the subprogram
renaming declaration, use the Slocs of the formal parameters from the
declaration of the formal subprogram when creating the formal parameter
entities in the renaming declaration.
(Analyze_Formal_Type_Declaration): Change the placement of the error
message concerning illegal known discriminants. It is now posted on the
type rather than on the first discriminant. This change ensures early
error report.
(Freeze_Subprogram_Body): If the generic subprogram is nested within
the package body that contains the instance, do not generate an
out-of-place freeze node for the enclosing package.
(Collect_Previous_Instantiations): Ignore internal instantiations
generated for formal packages.
(Validate_Derived_Type_Instance): Add a check that when a formal
derived type is Known_To_Have_Preelab_Init then the actual type must
have preelaborable initialization, and issue an error when this
condition is violated.
Thomas Quinot [Thu, 13 Dec 2007 10:29:38 +0000 (11:29 +0100)]
sem_ch10.adb (Check_Private_Child_Unit): A non-private library level subprogram body that acts as its own spec may not...
2007-12-06 Thomas Quinot <quinot@adacore.com>
Ed Schonberg <schonberg@adacore.com>
* sem_ch10.adb (Check_Private_Child_Unit): A non-private library level
subprogram body that acts as its own spec may not have a non-private
WITH clause on a private sibling.
(Build_Unit_Name): If the parent unit in the name in a with_clause on a
child unit is a renaming, create an implicit with_clause on that
parent, and not on the unit it renames, to prevent visibility errors
in the current unit.
Arnaud Charlet [Thu, 13 Dec 2007 10:29:02 +0000 (11:29 +0100)]
rtsfind.adb (Check_CRT): Take into account RTE_Available_Call Fixes another case where...
* rtsfind.adb (Check_CRT): Take into account RTE_Available_Call
Fixes another case where RTE_Available_Call was ignored instead of being
taken into account.
(Load_Fail): Ditto.
Bob Duff [Thu, 13 Dec 2007 10:28:48 +0000 (11:28 +0100)]
restrict.ads, [...] (Check_Implicit_Dynamic_Code_Allowed): New procedure to be called from the back end to check the...
2007-12-06 Bob Duff <duff@adacore.com>
* restrict.ads, restrict.adb (Check_Implicit_Dynamic_Code_Allowed): New
procedure to be called from the back end to check the
No_Implicit_Dynamic_Code restriction.
Emmanuel Briot [Thu, 13 Dec 2007 10:28:39 +0000 (11:28 +0100)]
prj.ads, prj.adb (Is_A_Language): Now takes a Name_Id instead of a string
2007-12-06 Emmanuel Briot <briot@adacore.com>
Vincent Celier <celier@adacore.com>
* prj.ads, prj.adb (Is_A_Language): Now takes a Name_Id instead of a
string
(Must_Check_Configuration, Default_Language_Is_Ada): new flags in
prj.ads
(Hash): Move instantiation of System.HTable.Hash from spec to body
(prj-nmsc.adb): Optimize calls to Name_Find when on case sensitive
systems, since we do not need to recompute the Name_Id for the canonical
file name.
(Body_Suffix_Id_Of, Spec_Suffix_Id_Of): new version that takes a name_id
as a parameter. This parameter is in fact always "ada" in all calls, and
we were doing 160560 extra calls to Name_Find to convert it to Name_Ada
while loading a project with 40000 files
* prj-attr.adb: Fix name of attribute Dependency_Driver
Change the kind of indexing for attribute Root
* prj-dect.adb (Parse_Declarative_Items): Allow redeclarations of
variables already declared, in case constructions.
* prj-env.adb (Initialize): Reset Current_Source_Path_File and
Current_Object_Path_File to No_Path.
* prj-ext.adb (Initialize_Project_Path): In multi language mode, use
ADA_PROJECT_PATH if value of GPR_PROJECT_PATH is empty.
* prj-makr.adb: new parameter Current_Dir
* prj-nmsc.ads, prj-nmsc.adb (Find_Explicit_Sources): Do not look for
Ada sources when language is not Ada.
Change Opt.Follow_Links to Opt.Follow_Links_For_Files.
(Find_Excluded_Sources, Find_Explicit_Sources): new subprograms
(Must_Check_Configuration, Default_Language_Is_Ada): new flags.
(Locate_Directory): Always resolve links when computing Canonical_Path
(Look_For_Sources): Make sure that Name_Buffer contains the file name
in Source_Files before checking for the presence of a directory
separator.
Optimize calls to Name_Find when on case sensitive systems.
(Body_Suffix_Id_Of, Spec_Suffix_Id_Of): new version that takes a name_id
as a parameter.
(Prj.Nmsc.Check): new parameter Current_Dir
(Check_Ada_Naming_Schemes): Restrictions on suffixes are relaxed. They
cannot be empty and the spec suffix cannot be the same as the body or
separate suffix.
(Get_Unit): When a file name can be of several unit kinds (spec, body or
subunit), always consider the longest suffix.
(Check_Configuration): Do not issue an error if there is no compiler
for a language. Just issue a warning and ignore the sources for the
language.
(Check_Library_Attributes): Only check Library_Dir if Library_Name is
not empty.
(Check_Naming_Schemes.Maked_Unit): Only output message if high verbosity
(Unit_Exceptions): New hash table
(Check_Naming_Schemes): Check if a file that could be a unit because of
the naming scheme is not in fact a source because there is an exception
for the unit.
(Look_For_Sources): Put the unit exceptions in hash table
Unit_Exceptions
(Get_Unit_Exceptions): Give initial value No_Source to local variable
Other_Part to avoid exception when code is compiled with validity
checking.
(Get_Sources_From_File): Check that there is no directory information
in the file names.
(Look_For_Sources): Check that there is no directory information in the
list of file names in Source_Files.
(Look_For_Sources): In multi-language mode, do not allow exception file
names that are excluded.
(Excluded_Sources_Htable): New hash table
(Search_Directories.Check_File): New procedure to simplify
Search_Directories.
(Search_Directories): Do not consider excluded sources
(Look_For_Sources): Populate Excluded_Sources_Htable before calling
Search_Directories.
(Get_Exceptions): Set component Lang_Kind of Source_Data
(Get_Unit_Exceptions): Ditto
(Search_Directories): Ditto
* prj-pars.adb: new parameter Current_Dir
* prj-part.ads, prj-part.adb:
Change Opt.Follow_Links to Opt.Follow_Links_For_Files.
(Opt.Follow_Links_For_Dirs): New flag
(Project_Path_Name_Of): Cache information returned by this routine as
Locate_Regular_File is a costly routine. The code to output a log
information and the effective call to Locate_Regular_File is now
factorized into a routine (code clean-up).
(Parse, Parse_Single_Project): new parameter Current_Dir
When main project file cannot be found, indicate in the error
message the project path that was used to do the search.
* prj-proc.ads, prj-proc.adb (Opt.Follow_Links_For_Dirs): New flag
(Prj.Proc.Process*): new parameter Current_Dir
* switch-m.adb: Change Opt.Follow_Links to Opt.Follow_Links_For_Files
Robert Dewar [Thu, 13 Dec 2007 10:28:24 +0000 (11:28 +0100)]
inline.adb (Back_End_Cannot_Inline): Use new flag Has_Pragma_Inline_Always instead of obsolete function Is_Always_Inlined
2007-12-06 Robert Dewar <dewar@adacore.com>
* inline.adb (Back_End_Cannot_Inline): Use new flag
Has_Pragma_Inline_Always instead of obsolete function Is_Always_Inlined
* sem_ch6.ads, sem_ch6.adb (Analyze_Subprogram_Body): Use new flag
Has_Pragma_Inline_Always instead.
of obsolete function Is_Always_Inlined
(Build_Body_To_Inline): Same change
(Cannot_Inline): Same change
Do not give warning on exception raise in No_Return function
* sem_ch13.adb (Analyze_Record_Representation_Clause): If an inherited
component has two inconsistent component clauses in the same record
representation clause, favor the message that complains about
duplication rather than inconsistency.
Update comments.
(Record_Representation_Clause): Do not warn on missing component
clauses for inherited components of a type extension.
(Rep_Item_Too_Late): Do not attempt to link pragma into rep chain for
an overloadable item if it is a pragma that can apply to multiple
overloadable entities (e.g. Inline) because a pragma cannot be on
more than one chain at a time.
(Validate_Unchecked_Conversion): Add code to warn on unchecked
conversion where one of the operands is Ada.Calendar.Time.
(Analyze_Attribute_Definition_Clause): Fix typo in error message.
For now, ignore Component_Size clause on VM targets, as done for
pragma Pack.
* s-wchcnv.ads, s-wchcnv.adb:
Minor code cleanup (make formal type consistent with spec)
* namet.adb: Update comments.
* par-endh.adb (Evaluate_End_Entry): Use new
Namet.Sp.Is_Bad_Spelling_Of function
* par-load.adb (Load): Use new Namet.Sp.Is_Bad_Spelling_Of function
* sem_aggr.adb (Resolve_Record_Aggregate): If a component of an
ancestor is an access type initialized with a box, set its type
explicitly, for use in subsequent expansion.
(Check_Misspelled_Component): Use new Namet.Sp.Is_Bad_Spelling_Of
function
Vasiliy Fofanov [Thu, 13 Dec 2007 10:27:42 +0000 (11:27 +0100)]
g-regist.ads, [...] (Set_Value): new parameter Expand...
2007-12-06 Vasiliy Fofanov <fofanov@adacore.com>
* g-regist.ads, g-regist.adb (Set_Value): new parameter Expand; when
set to True this procedure will create the value of type REG_EXPAND_SZ.
It was only possible to create REG_SZ values before.
Vincent Celier [Thu, 13 Dec 2007 10:27:32 +0000 (11:27 +0100)]
gnatcmd.adb (GNATCmd): Do not issue -d= switch to gnatmetric when object directory of main project...
2007-12-06 Vincent Celier <celier@adacore.com>
* gnatcmd.adb (GNATCmd): Do not issue -d= switch to gnatmetric when
object directory of main project does not exist.
On VMS, correctly set then environment variable for the source
directories.
Robert Dewar [Thu, 13 Dec 2007 10:27:21 +0000 (11:27 +0100)]
gnat1drv.adb (Gnat1drv): Properly set new flag Opt.Real_VMS_Target
2007-12-06 Robert Dewar <dewar@adacore.com>
Ed Schonberg <schonberg@adacore.com>
* gnat1drv.adb (Gnat1drv): Properly set new flag Opt.Real_VMS_Target
* layout.adb (Resolve_Attribute, case 'Access): If designated type of
context is a limited view, use non-limited view when available. If the
non-limited view is an unconstrained array, this enforces consistency
requirements in 3.10.2 (27).
(Layout_Type): For an access type whose designated type is a limited
view, examine its declaration to determine if it is an unconstrained
array, and size the access type accordingly.
(Layout_Type): Do not force 32-bits for convention c subprogram
pointers in -gnatdm mode, only if real vms target.
* sem_attr.adb (Analyze_Access_Attribute): Use new flag
Has_Pragma_Inline_Always instead of obsolete function Is_Always_Inlined
(Analyze_Access_Attribute,Attribute_Address): Remove checks for
violations of the No_Implicit_Dynamic_Code restriction.
(Resolve_Attribute, case 'Access): If designated type of context is a
limited view, use non-limited view when available. If the non-limited
view is an unconstrained array, this enforces consistency requirements
in 3.10.2 (27).
(Layout_Type): For an access type whose designated type is a limited
view, examine its declaration to determine if it is an unconstrained
array, and size the access type accordingly.
Robert Dewar [Thu, 13 Dec 2007 10:26:56 +0000 (11:26 +0100)]
exp_prag.adb (Expand_Pragma_Assert): Recognize new warning flag for assert fail
2007-12-06 Robert Dewar <dewar@adacore.com>
Ed Schonberg <schonberg@adacore.com>
* exp_prag.adb (Expand_Pragma_Assert): Recognize new warning flag for
assert fail
* ug_words: Add entries for -gnatw.a -gnatw.A
* sem_res.adb (Set_String_Literal_Subtype): If the context of the
literal is a subtype with non-static constraints, use the base type of
the context as the base of the string subtype, to prevent type
mismatches in gigi.
(Resolve_Actuals): If the actual is an entity name, generate a
reference before the actual is resolved and expanded, to prevent
spurious warnings on formals of enclosing protected operations.
(Analyze_Overloaded_Selected_Component): If type of prefix if
class-wide, use visible components of base type.
(Resolve_Selected_Component): Ditto.
(Resolve_Short_Circuit): Detect case of pragma Assert argument
evaluating to False, and issue warning message.
Javier Miranda [Thu, 13 Dec 2007 10:26:41 +0000 (11:26 +0100)]
exp_intr.adb (Expand_Dispatching_Constructor_Call): Add missing support for generic dispatching constructor calls in which we need...
2007-12-06 Javier Miranda <miranda@adacore.com>
* exp_intr.adb (Expand_Dispatching_Constructor_Call): Add missing
support for generic dispatching constructor calls in which we need to
locate the tag of a secondary dispatch table associated with an
interface type to properly dispatch the call.
(Expand_N_Attribute_Reference [case Address],
Expand_Dispatching_Constructor_Call, Expand_Unc_Deallocation): Fix
handling of VM targets.
2007-12-06 Hristian Kirtchev <kirtchev@adacore.com>
Javier Miranda <miranda@adacore.com>
* exp_disp.ads, exp_disp.adb (Default_Prim_Op_Position): Primitive
_Disp_Requeue occupies dispatch table slot number 15. Move
_Disp_Timed_Select to slot 16.
(Make_Disp_Requeue_Body, Make_Disp_Requeue_Spec): New routines which
generate the spec and body of _Disp_Reqeueue.
(Make_DT): Build and initialize the second dispatch table.
Handle initialization of RC_Offset when the parent
is a private type with variable size components.
(Make_Secondary_DT): Complete documentation. Add support to
initialize the second dispatch table.
(Make_Tags): Generate the tag of the second dispatch table.
(Register_Primitive): Add support to register primitives in the
second dispatch table.
exp_ch9.adb (Expand_N_Asynchronous_Select, [...]): Code and comment reformatting.
2007-12-06 Hristian Kirtchev <kirtchev@adacore.com>
Ed Schonberg <schonberg@adacore.com>
* exp_ch9.adb (Expand_N_Asynchronous_Select,
Expand_N_Conditional_Entry_Call, Expand_N_Timed_Entry_Call): Code and
comment reformatting.
(Set_Privals): Inherit aliased flag from formal. From code reading.
(Build_Simple_Entry_Call): Out parameters of an access type are passed
by copy and initialized from the actual. This includes entry parameters.
(Expand_N_Requeue_Statement): Reimplement in order to handle both Ada 95
and Ada 2005 models of requeue.
(Null_Statements): Still connsider do-end block null if it contains
Unreferenced and Warnings pragmas.
(Expand_N_Accept_Statement): Do not optimize away null do end if
dispatching policy is other than defaulted.
(Expand_N_Timed_Entry_Call): When the triggering statement is a
dispatching call, manually analyze the delay statement.
(Find_Parameter_Type): Move subprogram to Sem_Util.
Gary Dismukes [Thu, 13 Dec 2007 10:25:50 +0000 (11:25 +0100)]
exp_ch8.ads, [...] (Expand_N_Subprogram_Renaming_Declaration): In the case where the renamed subprogram is a dereference...
2007-12-06 Gary Dismukes <dismukes@adacore.com>
* exp_ch8.ads, exp_ch8.adb (Expand_N_Subprogram_Renaming_Declaration):
In the case where the renamed subprogram is a dereference, call
Force_Evaluation on the prefix.
Javier Miranda [Thu, 13 Dec 2007 10:25:35 +0000 (11:25 +0100)]
exp_ch7.adb (Expand_N_Package_Body): Replace occurrence of attribute Is_Complation_Unit by Is_Library_Level_Entity...
2007-12-06 Javier Miranda <miranda@adacore.com>
Ed Schonberg <schonberg@adacore.com>
* exp_ch7.adb (Expand_N_Package_Body): Replace occurrence of attribute
Is_Complation_Unit by Is_Library_Level_Entity in the code
that decides if the static dispatch tables need to be built.
(Wrap_Transient_Declaration): Do not generate a finalization call if
this is a renaming declaration and the renamed object is a component
of a controlled type.
* exp_ch3.adb (Predefined_Primitive_Bodies): Generate the body of
predefined primitive _Disp_Requeue.
(Make_Predefined_Primitive_Specs): Create the spec for predefined
primitive _Disp_Requeue.
(Make_Predefined_Primitive_Specs/Predefined_Primitive_Bodies): Set the
type of formal Renamed_Eq to Entity_Id (instead of Node_Id).
(Make_Predefined_Primitive_Specs): Spec of "=" needed if the parent is
an interface type. In case of limited interfaces we now declare all the
predefined primitives associated with synchronized interfaces as
abstract.
(Predef_Spec_Or_Body): For interface types generate abstract subprogram
declarations.
(Predefined_Primitive_Bodies): Add body of "=" if the parent of the
tagged type is an interface type and there is no user-defined equality
function.
Add also bodies of predefined primitives associated with synchronized
interfaces.
(Freeze_Record_Type): Do not build bodies of predefined primitives of
interface types because they are now defined abstract.
Add missing documentation.
(Expand_Record_Controller): Update occurrence of Related_Interface
to Related_Type.
(Build_Offset_To_Top_Functions): Do nothing in case of VM.
(Expand_N_Object_Declaration): Take into account VM_Target when handling
class wide interface object declaration.
(Expand_Previous_Access_Type): Do not create a duplicate master entity
if the access type already has one.
(Expand_N_Object_Declaration): Defend against attempt to validity check
generic types. Noticed for -gnatVcf specified with previous errors.
Ed Schonberg [Thu, 13 Dec 2007 10:24:30 +0000 (11:24 +0100)]
exp_aggr.adb (Build_Record_Aggr_Code): If there is an aggregate for a limited ancestor part...
2007-12-06 Ed Schonberg <schonberg@adacore.com>
* exp_aggr.adb (Build_Record_Aggr_Code): If there is an aggregate for a
limited ancestor part, initialize controllers of enclosing record
before expanding ancestor aggregate.
(Gen_Assign): If a component of the aggregate is box-initialized, add
code to call Initialize if the component is controlled, and explicit
assignment of null if the component is an access type.
Handle properly aggregates for limited types that appear in object
declarations when the aggregate contains controlled values such as
protected types.
When expanding limited aggregates into individual components, do not
call Adjust on controlled components that are limited.
* einfo.ads, einfo.adb: Flag 232 is now Implemented_By_Entry.
(Implemented_By_Entry, Set_Implemented_By_Entry): New routines.
(Write_Entry_Flags): Add an entry for Implemented_By_Entry.
(Renamed_In_Spec): New flag
(Has_Pragma_Inline_Always): New flag
Add missing doc for pragma Obsolescent_Warning
Add missing doc for 17 additional unused flags (230-247)
(Is_Derived_Type): Remove condition "not Is_Generic_Type".
Alphabetize with clauses.
Separate Is_Thunk and Has_Thunks flags
(Write_Entity_Flags): Add forgotten entry for Has_Thunks
(Related_Interface): Renamed to Related_Type.
(Has_Thunks/Set_Has_Thunks): Subprograms of new attribute.
(Set_Is_Flag): Restrict the assertion.
Bob Duff [Thu, 13 Dec 2007 10:23:29 +0000 (11:23 +0100)]
clean.adb (Usage): Add line for -aP
2007-12-06 Bob Duff <duff@adacore.com>
* clean.adb (Usage): Add line for -aP
(Check_Version_And_Help): Change Check_Version_And_Help to be generic,
with a parameter "procedure Usage", instead of passing a pointer to a
procedure. This is to eliminate trampolines (since the Usage procedure
is often nested in a main procedure, and it would be inconvenient to
unnest it).
* g-comlin.adb (For_Each_Simple_Switch): Change For_Each_Simple_Switch
to be generic, with a parameter "procedure Callback (...)", instead of
passing a pointer to a procedure. This is to eliminate trampolines
(since the Callback procedure is usually nested).
* gnatfind.adb, switch.adb, switch.ads, gnatlink.adb, gnatls.adb,
gnatname.adb, gnatxref.adb, gnatchop.adb, gprep.adb, gnatbind.adb
(Check_Version_And_Help): Change Check_Version_And_Help to be generic.
* g-pehage.adb (Compute_Edges_And_Vertices, Build_Identical_Key_Sets):
Use the generic Heap_Sort_G instead of Heap_Sort_A.