[RFA] Typedef support for linespecs

[RFA] Typedef support for linespecs

[Keith Seitz]

[-- Attachment #1: Type: text/plain, Size: 1030 bytes --]

Hi,

CVS HEAD currently fails if a user does something like:

class foo
{
public:
   void a_function (void) { }
};
typedef MyFoo foo;
MyFoo instance;

(gdb) list MyFoo::a_function

This happens because when we lookup "MyFoo", lookup_prefix_sym will only 
search STRUCT_DOMAIN, and it won't find "MyFoo", since typedefs live in 
VAR_DOMAIN.

The attached patch attempts to address this issue by searching 
VAR_DOMAIN if the search through STRUCT_DOMAIN fails.

[In case I am not always explicit: This patch introduces no regressions 
or failures/errors of any kind with the test suite on x86 linux.]

Comments/Questions/Concerns?
Keith

ChangeLog
2009-11-11  Keith Seitz  <keiths@redhat.com>

	* linespec.c (lookup_prefix_sym): Lookup the symbol
	in both STRUCT_DOMAIN and VAR_DOMAIN.

testsuite/ChangeLog
2009-11-11  Keith Seitz  <keiths@redhat.com>

	* gdb.cp/classes.cc (ByAnyOtherName): Add typedef and
	use it instead of "Foo".
	* gdb.cp/classes.exp (do_tests): Add a test to access
	a method through a typedef'd class name.


[-- Attachment #2: linespec-typedef.patch --]
[-- Type: text/plain, Size: 2689 bytes --]

Index: linespec.c
===================================================================
RCS file: /cvs/src/src/gdb/linespec.c,v
retrieving revision 1.92
diff -u -p -r1.92 linespec.c
--- linespec.c	19 Oct 2009 09:51:41 -0000	1.92
+++ linespec.c	11 Nov 2009 22:37:47 -0000
@@ -1428,6 +1428,7 @@ lookup_prefix_sym (char **argptr, char *
 {
   char *p1;
   char *copy;
+  struct symbol *sym;
 
   /* Extract the class name.  */
   p1 = p;
@@ -1446,7 +1447,26 @@ lookup_prefix_sym (char **argptr, char *
   /* At this point p1->"::inA::fun", p->"inA::fun" copy->"AAA",
      argptr->"inA::fun" */
 
-  return lookup_symbol (copy, 0, STRUCT_DOMAIN, 0);
+  sym = lookup_symbol (copy, 0, STRUCT_DOMAIN, 0);
+  if (sym == NULL)
+    {
+      /* Typedefs are in VAR_DOMAIN so the above symbol lookup will
+	 fail when the user attempts to lookup a method of a class
+	 via a typedef'd name (NOT via the class's name, which is already
+	 handled in symbol_matches_domain).  So try the lookup again
+	 using VAR_DOMAIN (where typedefs live) and double-check that we
+	 found a struct/class type.  */
+      struct symbol *s = lookup_symbol (copy, 0, VAR_DOMAIN, 0);
+      if (s != NULL)
+	{
+	  struct type *t = SYMBOL_TYPE (s);
+	  CHECK_TYPEDEF (t);
+	  if (TYPE_CODE (t) == TYPE_CODE_STRUCT)
+	    return s;
+	}
+    }
+
+  return sym;
 }
 
 /* This finds the method COPY in the class whose type is T and whose
Index: testsuite/gdb.cp/classes.cc
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.cp/classes.cc,v
retrieving revision 1.7
diff -u -p -r1.7 classes.cc
--- testsuite/gdb.cp/classes.cc	21 Sep 2009 19:23:22 -0000	1.7
+++ testsuite/gdb.cp/classes.cc	11 Nov 2009 22:37:48 -0000
@@ -417,6 +417,8 @@ class Foo
   int times (int y);
 };
 
+typedef Foo ByAnyOtherName;
+
 class Bar : public Base1, public Foo {
  public:
   int z;
@@ -431,7 +433,7 @@ int Foo::st = 100;
 
 Foo::operator int() { return x; }
 
-Foo foo(10, 11);
+ByAnyOtherName foo(10, 11);
 Bar bar(20, 21, 22);
 
 class ClassWithEnum {
Index: testsuite/gdb.cp/classes.exp
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.cp/classes.exp,v
retrieving revision 1.22
diff -u -p -r1.22 classes.exp
--- testsuite/gdb.cp/classes.exp	21 Sep 2009 19:23:22 -0000	1.22
+++ testsuite/gdb.cp/classes.exp	11 Nov 2009 22:37:48 -0000
@@ -636,6 +636,8 @@ proc do_tests {} {
     gdb_test "print base1::Base1" "<.*Base1.*>" "print ctor of typedef class"
     gdb_test "print base1::~Base1" "<.*~Base1(\\(\\))?>" \
 	"print dtor of typedef class"
+
+    gdb_test "list ByAnyOtherName::times" ".*int Foo::times.*"
 }
 
 do_tests

Re: [RFA] Typedef support for linespecs

[Daniel Jacobowitz]

On Wed, Nov 11, 2009 at 02:49:13PM -0800, Keith Seitz wrote:
> ChangeLog
> 2009-11-11  Keith Seitz  <keiths@redhat.com>
> 
> 	* linespec.c (lookup_prefix_sym): Lookup the symbol
> 	in both STRUCT_DOMAIN and VAR_DOMAIN.
> 
> testsuite/ChangeLog
> 2009-11-11  Keith Seitz  <keiths@redhat.com>
> 
> 	* gdb.cp/classes.cc (ByAnyOtherName): Add typedef and
> 	use it instead of "Foo".
> 	* gdb.cp/classes.exp (do_tests): Add a test to access
> 	a method through a typedef'd class name.

OK.  I still hate linespecs, though.

(Seriously... if we figure out that it isn't a file:line, we should
use the normal expression parser for an expression portion of
it... yes, I do realize just how difficult this would be, but the
inconsistencies drive folks mad.)

-- 
Daniel Jacobowitz
CodeSourcery

Re: [RFA] Typedef support for linespecs

[Keith Seitz]

On 11/11/2009 03:01 PM, Daniel Jacobowitz wrote:

> OK.  I still hate linespecs, though.

Committed. Thank you for your review.

> (Seriously... if we figure out that it isn't a file:line, we should
> use the normal expression parser for an expression portion of
> it... yes, I do realize just how difficult this would be, but the
> inconsistencies drive folks mad.)

Yeah, I rather agree with you. Something like this may yet get onto an 
Archer agenda in the future, but for now, there are, unfortunately, 
"bigger fish to fry" -- at least for me.

Thanks,
Keith