diff mbox

[2/3] gdb: Improve output from "info types" commad

Message ID 20190719194331.GB23204@embecosm.com
State New
Headers show

Commit Message

Andrew Burgess July 19, 2019, 7:43 p.m. UTC
* Pedro Alves <palves@redhat.com> [2019-07-18 21:07:10 +0100]:

> On 7/12/19 12:37 PM, Andrew Burgess wrote:
> > This commit makes two changes to the "info types" command:
> > 
> > First, only use typedef_print for printing typedefs, and use
> > type_print for printing non-typedef scalar (non-struct) types.  The
> > result of this is the output for builtin types goes from this:
> > 
> >     typedef double;
> >     typedef float;
> >     typedef int;
> > 
> > to this:
> > 
> >     double;
> >     float;
> >     int;
> > 
> > which seems to make more sense.
> > 
> > Next GDB no longer matches msymbols as possible type names.  When
> > looking for function symbols it makes sense to report matching
> > msymbols from the text sections, and for variables msymbols from the
> > data/bss sections, but when reporting types GDB would match msymbols
> > of type absolute.  But I don't see why these are likely to indicate
> > type names.  As such I've updated the msymbol matching lists in
> > symtab.c:search_symbols so that when searching in the TYPES_DOMAIN, we
> > never match any msymbols.
> > 
> 
> That sounds fine to me.
> 
> > 	* symtab.c (search_symbols): Adjust msymbol matching type arrays
> > 	so that GDB doesn't match any msymbols when searching in the
> > 	TYPES_DOMAIN.
> > 	(print_symbol_info): Print using typedef_print or type_print based
> > 	on the type of the symbol.  Add updated FIXME comment mobed from...
> > 	(_initialize_symtab): ... move and update FIXME comment to above.
> > 
> 
> Typo: "mobed".
> 
> > gdb/testsuite/ChangeLog:
> > 
> > 	* gdb.base/info-types.c: New file.
> > 	* gdb.base/info-types.exp: New file.
> > 	* gdb.cp/info-types.cc: New file.
> > 	* gdb.cp/info-types.exp: New file.
> 
> I'm not immediately seeing what the C++ testcase is trying to
> test.  If you replace "class" with struct, and use "typedef struct AA"
> instead of "typedef AA", isn't that code basically C code as well?
> 
> Also, might it be a good idea to check the info-types.c stuff in
> C++ mode as well, to make sure we normalize C and C++ modes?
> I.e., make gdb.base/info-types.exp compile once as a C program, and
> once as a C++ program?

I've removed the gdb.cp/info-types.{cc,exp} files, and extended the
gdb.base test to include an example of 'class'.  Yes, we know
internally they are handled just like structs, but I'd like to cover
it "just in case".  The gdb.base case is now compiled as C and C++.

Thanks,
Andrew

---

    gdb: Improve output from "info types" commad
    
    This commit makes two changes to the "info types" command:
    
    First, only use typedef_print for printing typedefs, and use
    type_print for printing non-typedef scalar (non-struct) types.  The
    result of this is the output for builtin types goes from this:
    
        typedef double;
        typedef float;
        typedef int;
    
    to this:
    
        double;
        float;
        int;
    
    which seems to make more sense.
    
    Next GDB no longer matches msymbols as possible type names.  When
    looking for function symbols it makes sense to report matching
    msymbols from the text sections, and for variables msymbols from the
    data/bss sections, but when reporting types GDB would match msymbols
    of type absolute.  But I don't see why these are likely to indicate
    type names.  As such I've updated the msymbol matching lists in
    symtab.c:search_symbols so that when searching in the TYPES_DOMAIN, we
    never match any msymbols.
    
    gdb/ChangeLog:
    
            * symtab.c (search_symbols): Adjust msymbol matching type arrays
            so that GDB doesn't match any msymbols when searching in the
            TYPES_DOMAIN.
            (print_symbol_info): Print using typedef_print or type_print based
            on the type of the symbol.  Add updated FIXME comment moved from...
            (_initialize_symtab): ... move and update FIXME comment to above.
    
    gdb/testsuite/ChangeLog:
    
            * gdb.base/info-types.c: New file.
            * gdb.base/info-types.exp: New file.

Comments

Pedro Alves July 19, 2019, 7:54 p.m. UTC | #1
On 7/19/19 8:43 PM, Andrew Burgess wrote:
> * Pedro Alves <palves@redhat.com> [2019-07-18 21:07:10 +0100]:
> 
>> On 7/12/19 12:37 PM, Andrew Burgess wrote:

>> I'm not immediately seeing what the C++ testcase is trying to
>> test.  If you replace "class" with struct, and use "typedef struct AA"
>> instead of "typedef AA", isn't that code basically C code as well?
>>
>> Also, might it be a good idea to check the info-types.c stuff in
>> C++ mode as well, to make sure we normalize C and C++ modes?
>> I.e., make gdb.base/info-types.exp compile once as a C program, and
>> once as a C++ program?
> 
> I've removed the gdb.cp/info-types.{cc,exp} files, and extended the
> gdb.base test to include an example of 'class'.

Ah, OK, it's "class" specifically that you were looking at test.
I wasn't sure.

> Yes, we know
> internally they are handled just like structs, but I'd like to cover
> it "just in case".  The gdb.base case is now compiled as C and C++.

Thanks.

Pedro Alves
Pedro Alves July 20, 2019, 1:47 p.m. UTC | #2
On 7/19/19 8:43 PM, Andrew Burgess wrote:

> I've removed the gdb.cp/info-types.{cc,exp} files, and extended the
> gdb.base test to include an example of 'class'.  Yes, we know
> internally they are handled just like structs, but I'd like to cover
> it "just in case".  The gdb.base case is now compiled as C and C++.
> 
LGTM.

Thanks,
Pedro Alves
diff mbox

Patch

diff --git a/gdb/symtab.c b/gdb/symtab.c
index 16861e2bc92..ba0d5eed6b3 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -4339,13 +4339,13 @@  search_symbols (const char *regexp, enum search_domain kind,
   struct symbol *sym;
   int found_misc = 0;
   static const enum minimal_symbol_type types[]
-    = {mst_data, mst_text, mst_abs};
+    = {mst_data, mst_text, mst_unknown};
   static const enum minimal_symbol_type types2[]
-    = {mst_bss, mst_file_text, mst_abs};
+    = {mst_bss, mst_file_text, mst_unknown};
   static const enum minimal_symbol_type types3[]
-    = {mst_file_data, mst_solib_trampoline, mst_abs};
+    = {mst_file_data, mst_solib_trampoline, mst_unknown};
   static const enum minimal_symbol_type types4[]
-    = {mst_file_bss, mst_text_gnu_ifunc, mst_abs};
+    = {mst_file_bss, mst_text_gnu_ifunc, mst_unknown};
   enum minimal_symbol_type ourtype;
   enum minimal_symbol_type ourtype2;
   enum minimal_symbol_type ourtype3;
@@ -4628,7 +4628,23 @@  print_symbol_info (enum search_domain kind,
   /* Typedef that is not a C++ class.  */
   if (kind == TYPES_DOMAIN
       && SYMBOL_DOMAIN (sym) != STRUCT_DOMAIN)
-    typedef_print (SYMBOL_TYPE (sym), sym, gdb_stdout);
+    {
+      /* FIXME: For C (and C++) we end up with a difference in output here
+	 between how a typedef is printed, and non-typedefs are printed.
+	 The TYPEDEF_PRINT code places a ";" at the end in an attempt to
+	 appear C-like, while TYPE_PRINT doesn't.
+
+	 For the struct printing case below, things are worse, we force
+	 printing of the ";" in this function, which is going to be wrong
+	 for languages that don't require a ";" between statements.  */
+      if (TYPE_CODE (SYMBOL_TYPE (sym)) == TYPE_CODE_TYPEDEF)
+	typedef_print (SYMBOL_TYPE (sym), sym, gdb_stdout);
+      else
+	{
+	  type_print (SYMBOL_TYPE (sym), "", gdb_stdout, -1);
+	  printf_filtered ("\n");
+	}
+    }
   /* variable, func, or typedef-that-is-c++-class.  */
   else if (kind < TYPES_DOMAIN
 	   || (kind == TYPES_DOMAIN
@@ -6080,14 +6096,6 @@  Prints the functions.\n"),
 				  _("functions")));
   set_cmd_completer_handle_brkchars (c, info_print_command_completer);
 
-  /* FIXME:  This command has at least the following problems:
-     1.  It prints builtin types (in a very strange and confusing fashion).
-     2.  It doesn't print right, e.g. with
-     typedef struct foo *FOO
-     type_print prints "FOO" when we want to make it (in this situation)
-     print "struct foo *".
-     I also think "ptype" or "whatis" is more likely to be useful (but if
-     there is much disagreement "info types" can be fixed).  */
   c = add_info ("types", info_types_command, _("\
 All type names, or those matching REGEXP.\n\
 Usage: info types [-q] [REGEXP]\n\
diff --git a/gdb/testsuite/gdb.base/info-types.c b/gdb/testsuite/gdb.base/info-types.c
new file mode 100644
index 00000000000..6746238801d
--- /dev/null
+++ b/gdb/testsuite/gdb.base/info-types.c
@@ -0,0 +1,78 @@ 
+/* Copyright 2019 Free Software Foundation, Inc.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
+
+typedef int my_int_t;
+typedef float my_float_t;
+typedef my_int_t nested_int_t;
+typedef my_float_t nested_float_t;
+
+struct baz_t
+{
+  float f;
+  double d;
+};
+
+typedef struct baz_t baz_t;
+typedef struct baz_t baz;
+typedef baz_t nested_baz_t;
+typedef baz nested_baz;
+typedef struct baz_t *baz_ptr;
+
+enum enum_t
+{
+ AA, BB, CC
+};
+
+typedef enum enum_t my_enum_t;
+typedef my_enum_t nested_enum_t;
+
+volatile int var_a;
+volatile float var_b;
+volatile my_int_t var_c;
+volatile my_float_t var_d;
+volatile nested_int_t var_e;
+volatile nested_float_t var_f;
+volatile struct baz_t var_g;
+volatile baz_t var_h;
+volatile baz var_i;
+volatile nested_baz_t var_j;
+volatile nested_baz var_k;
+volatile baz_ptr var_l;
+volatile enum enum_t var_m;
+volatile my_enum_t var_n;
+volatile nested_enum_t var_o;
+
+#ifdef __cplusplus
+
+class CL
+{
+  int a;
+};
+
+typedef CL my_cl;
+typedef CL *my_ptr;
+
+volatile CL var_cpp_a;
+volatile my_cl var_cpp_b;
+volatile my_ptr var_cpp_c;
+
+#endif /* __cplusplus */
+
+int
+main ()
+{
+  asm ("" ::: "memory");
+  return 0;
+}
diff --git a/gdb/testsuite/gdb.base/info-types.exp b/gdb/testsuite/gdb.base/info-types.exp
new file mode 100644
index 00000000000..a5808425398
--- /dev/null
+++ b/gdb/testsuite/gdb.base/info-types.exp
@@ -0,0 +1,109 @@ 
+# Copyright 2019 Free Software Foundation, Inc.
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+# Check that 'info types' produces the expected output for an inferior
+# containing a number of different types.
+
+# Only test C++ if we are able.  Always use C.
+if { [skip_cplus_tests] || [get_compiler_info "c++"] } {
+    set lang {c}
+} else {
+    set lang {c c++}
+}
+
+foreach l $lang {
+    set dir "$l"
+    remote_exec host "rm -rf [standard_output_file ${dir}]"
+    remote_exec host "mkdir -p [standard_output_file ${dir}]"
+}
+
+# Run 'info types' test, compiling the test file for language LANG,
+# which should be either 'c' or 'c++'.
+proc run_test { lang } {
+    global testfile
+    global srcfile
+    global binfile
+    global subdir
+    global srcdir
+    global compile_flags
+
+    standard_testfile .c
+
+    if {[prepare_for_testing "failed to prepare" \
+	     "${lang}/${testfile}" $srcfile "debug $lang"]} {
+	return -1
+    }
+
+    if ![runto_main] then {
+	fail "can't run to main"
+	return 0
+    }
+
+    if { $lang == "c++" } {
+	set output_re \
+	    [multi_line \
+		 "All defined types:" \
+		 "" \
+		 "File .*:" \
+		 "59:\[\t \]+CL;" \
+		 "21:\[\t \]+baz_t;" \
+		 "33:\[\t \]+enum_t;" \
+		 "28:\[\t \]+typedef baz_t baz;" \
+		 "31:\[\t \]+typedef baz_t \\* baz_ptr;" \
+		 "27:\[\t \]+typedef baz_t baz_t;" \
+		 "\[\t \]+double" \
+		 "\[\t \]+float" \
+		 "\[\t \]+int" \
+		 "64:\[\t \]+typedef CL my_cl;" \
+		 "38:\[\t \]+typedef enum_t my_enum_t;" \
+		 "17:\[\t \]+typedef float my_float_t;" \
+		 "16:\[\t \]+typedef int my_int_t;" \
+		 "65:\[\t \]+typedef CL \\* my_ptr;" \
+		 "30:\[\t \]+typedef baz_t nested_baz;" \
+		 "29:\[\t \]+typedef baz_t nested_baz_t;" \
+		 "39:\[\t \]+typedef enum_t nested_enum_t;" \
+		 "19:\[\t \]+typedef float nested_float_t;" \
+		 "18:\[\t \]+typedef int nested_int_t;" \
+		 "\[\t \]+unsigned int"]
+    } else {
+	set output_re \
+	    [multi_line \
+		 "All defined types:" \
+		 "" \
+		 "File .*:" \
+		 "28:\[\t \]+typedef struct baz_t baz;" \
+		 "31:\[\t \]+typedef struct baz_t \\* baz_ptr;" \
+		 "21:\[\t \]+struct baz_t;" \
+		 "\[\t \]+double" \
+		 "33:\[\t \]+enum enum_t;" \
+		 "\[\t \]+float" \
+		 "\[\t \]+int" \
+		 "38:\[\t \]+typedef enum enum_t my_enum_t;" \
+		 "17:\[\t \]+typedef float my_float_t;" \
+		 "16:\[\t \]+typedef int my_int_t;" \
+		 "30:\[\t \]+typedef struct baz_t nested_baz;" \
+		 "29:\[\t \]+typedef struct baz_t nested_baz_t;" \
+		 "39:\[\t \]+typedef enum enum_t nested_enum_t;" \
+		 "19:\[\t \]+typedef float nested_float_t;" \
+		 "18:\[\t \]+typedef int nested_int_t;" \
+		 "\[\t \]+unsigned int" ]
+    }
+
+    gdb_test "info types" $output_re
+}
+
+foreach_with_prefix l $lang {
+    run_test $l
+}