Trying to print a function local static variable of a const-qualified
method still doesn't work after the previous fixes:
(gdb) p 'S::method() const'::static_var
$1 = {i1 = 1, i2 = 2, i3 = 3}
(gdb) p S::method() const::static_var
No symbol "static_var" in specified context.
The reason is that the expression parser/evaluator loses the "const",
and the above unquoted case is just like trying to print a variable of
the non-const overload, if it exists, even. As if the above unquoted
case had been written as:
(gdb) p S::method()::static_var
No symbol "static_var" in specified context.
We can see the problem without static vars in the picture. With:
struct S
{
void method ();
void method () const;
};
Compare:
(gdb) print 'S::method(void) const'
$1 = {void (const S * const)} 0x400606 <S::method() const>
(gdb) print S::method(void) const
$2 = {void (S * const)} 0x4005d8 <S::method()> # wrong method!
That's what we need to fix. If we fix that, the function local static
case starts working.
The grammar production for function/method types is this one:
exp: exp '(' parameter_typelist ')' const_or_volatile
This results in a TYPE_INSTANCE expression evaluator operator. For
the example above, we get something like this ("set debug expression 1"):
...
0 TYPE_INSTANCE 1 TypeInstance: Type @0x560fda958be0 (void)
5 OP_SCOPE Type @0x560fdaa544d8 (S) Field name: `method'
...
While evaluating TYPE_INSTANCE, we end up in
value_struct_elt_for_reference, trying to find the method named
"method" that has the prototype recorded in TYPE_INSTANCE. In this
case, TYPE_INSTANCE says that we're looking for a method that has
"(void)" as parameters (that's what "1 TypeInstance: Type
@0x560fda958be0 (void)" above means. The trouble is that nowhere in
this mechanism do we communicate to value_struct_elt_for_reference
that we're looking for the _const_ overload.
value_struct_elt_for_reference only compared parameters, and the
non-const "method()" overload has matching parameters, so it's
considered the right match...
Conveniently, the "const_or_volatile" production in the grammar
already records "const" and "volatile" info in the type stack. The
type stack is not used in this code path, but we can borrow the
information. The patch converts the info in the type stack to an
"instance flags" enum, and adds that as another element in
TYPE_INSTANCE operators. This type instance flags is then applied to
the temporary type that is passed to value_struct_elt_for_reference
for matching.
The other side of the problem is that methods in the debug info aren't
marked const/volatile, so with that in place, the matching never finds
const/volatile-qualified methods.
The problem is that in the DWARF, there's no indication at all whether
a method is const/volatile qualified... For example (c++filt applied
to the linkage name for convenience):
<2><d3>: Abbrev Number: 6 (DW_TAG_subprogram)
<d4> DW_AT_external : 1
<d4> DW_AT_name : (indirect string, offset: 0x3df): method
<d8> DW_AT_decl_file : 1
<d9> DW_AT_decl_line : 58
<da> DW_AT_linkage_name: (indirect string, offset: 0x5b2): S::method() const
<de> DW_AT_declaration : 1
<de> DW_AT_object_pointer: <0xe6>
<e2> DW_AT_sibling : <0xec>
I see the same with both GCC and Clang. The patch works around this
by extracting the cv qualification from the "const" and "volatile" in
the demangled name. This will need further tweaking for "&" and
"const &" overloads, but we don't support them in the parser yet,
anyway.
The TYPE_CONST changes were necessary otherwise the comparisons in valops.c:
if (TYPE_CONST (intype) != TYPE_FN_FIELD_CONST (f, j))
continue;
would fail, because when both TYPE_CONST() TYPE_FN_FIELD_CONST() were
true, their values were different.
BTW, I'm recording the const/volatile-ness of methods in the
TYPE_FN_FIELD info because #1 - I'm not sure it's kosher to change the
method's type directly (vs having to call make_cv_type to create a new
type), and #2 it's what stabsread.c does:
...
case 'A': /* Normal functions. */
new_sublist->fn_field.is_const = 0;
new_sublist->fn_field.is_volatile = 0;
(*pp)++;
break;
case 'B': /* `const' member functions. */
new_sublist->fn_field.is_const = 1;
new_sublist->fn_field.is_volatile = 0;
...
After all this, this finally all works:
print S::method(void) const
$1 = {void (const S * const)} 0x400606 <S::method() const>
(gdb) p S::method() const::static_var
$2 = {i1 = 1, i2 = 2, i3 = 3}
gdb/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* c-exp.y (function_method, function_method_void): Add current
instance flags to TYPE_INSTANCE.
* dwarf2read.c (check_modifier): New.
(compute_delayed_physnames): Assert that only C++ adds delayed
physnames. Mark fn_fields as const/volatile depending on
physname.
* eval.c (make_params): New type_instance_flags parameter. Use
it as the new type's instance flags.
(evaluate_subexp_standard) <TYPE_INSTANCE>: Extract the instance
flags element and pass it to make_params.
* expprint.c (print_subexp_standard) <TYPE_INSTANCE>: Handle
instance flags element.
(dump_subexp_body_standard) <TYPE_INSTANCE>: Likewise.
* gdbtypes.h: Include "enum-flags.h".
(type_instance_flags): New enum-flags type.
(TYPE_CONST, TYPE_VOLATILE, TYPE_RESTRICT, TYPE_ATOMIC)
(TYPE_CODE_SPACE, TYPE_DATA_SPACE): Return boolean.
* parse.c (operator_length_standard) <TYPE_INSTANCE>: Adjust.
(follow_type_instance_flags): New function.
(operator_check_standard) <TYPE_INSTANCE>: Adjust.
* parser-defs.h (follow_type_instance_flags): Declare.
* valops.c (value_struct_elt_for_reference): const/volatile must
match too.
gdb/testsuite/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* gdb.base/func-static.c (S::method const, S::method volatile)
(S::method volatile const): New methods.
(c_s, v_s, cv_s): New instances.
(main): Call method() on them.
* gdb.base/func-static.exp (syntax_re, cannot_resolve_re): New variables.
(cannot_resolve): New procedure.
(cxx_scopes_list): Test cv methods. Add print-scope-quote and
print-quote-unquoted columns.
(do_test): Test printing each scope too.
While the previous commit made "p method()::static_var" (no
single-quotes) Just Work, if users (or frontends) try wrapping the
expression with quotes, they'll get:
(gdb) p 'S::method()::static_var'
'S::method()::static_var' has unknown type; cast it to its declared type
even if we _do_ have debug info for that variable. That's better than
the bogus/confusing value what GDB would print before the
stop-assuming-int patch:
(gdb) p 'S::method()::static_var'
$1 = 1
but I think it'd still be nice to make this case Just Work too.
In this case, due to the quoting, the C/C++ parser (c-exp.y)
interprets the whole expression/string as a single symbol name, and we
end up calling lookup_symbol on that name. There's no debug symbol
with that fully-qualified name, but since the compiler gives the
static variable a mangled linkage name exactly like the above, it
appears in the mininal symbols:
$ nm -A local-static | c++filt | grep static_var
local-static:0000000000601040 d S::method()::static_var
... and that's what GDB happens to find/print. This only happens in
C++, note, since for C the compiler uses different linkage names:
local-static-c:0000000000601040 d static_var.1848
So while (in C++, not C) function local static variables are given a
mangled name that demangles to the same syntax that GDB
documents/expects as the way to access function local statics, there's
no global symbol in the debug info with that name at all. The debug
info for a static local variable for a non-inline function looks like
this:
<1><2a1>: Abbrev Number: 19 (DW_TAG_subprogram)
...
<2><2f7>: Abbrev Number: 20 (DW_TAG_variable)
<2f8> DW_AT_name : (indirect string, offset: 0x4e9): static_var
<2fc> DW_AT_decl_file : 1
<2fd> DW_AT_decl_line : 64
<2fe> DW_AT_type : <0x25>
<302> DW_AT_location : 9 byte block: 3 40 10 60 0 0 0 0 0 (DW_OP_addr: 601040)
and for an inline function, it looks like this (linkage name run
through c++filt for convenience):
<2><21b>: Abbrev Number: 16 (DW_TAG_variable)
<21c> DW_AT_name : (indirect string, offset: 0x21a): static_var
<220> DW_AT_decl_file : 1
<221> DW_AT_decl_line : 48
<222> DW_AT_linkage_name: (indirect string, offset: 0x200): S::inline_method()::static_var
<226> DW_AT_type : <0x25>
<22a> DW_AT_external : 1
<22a> DW_AT_location : 9 byte block: 3 a0 10 60 0 0 0 0 0 (DW_OP_addr: 6010a0)
(The inline case makes the variable external so that the linker can
merge the different inlined copies. It seems like GCC never outputs
the linkage name for non-extern globals.)
When we read the DWARF, we record the static_var variable as a regular
variable of the containing function's block. This makes stopping in
the function and printing the variable as usual. The variable just so
happens to have a memory address as location.
So one way to make "p 'S::method()::static_var'" work would be to
record _two_ copies of the symbols for these variables. One in the
function's scope/block, with "static_var" as name, as we currently do,
and another in the static or global blocks (depending on whether the
symbol is external), with a fully-qualified name. I wrote a prototype
patch for that, and it works. For the non-inline case above, since
the debug info doesn't point to the linkage same, that patch built the
physname of the static local variable as the concat of the physname of
the containing function, plus "::", plus the variable's name. We
could make that approach work for C too, though it kind of feels
awkward to record fake symbol names like that in C.
The other approach I tried is to change the C++ symbol lookup routines
instead. This is the approach this commit takes. We can already
lookup up symbol in namespaces and classes, so this feels like a good
fit, and was easy enough. The advantage is that this doesn't require
recording extra symbols.
The test in gdb.cp/m-static.exp that exposed the need for this is
removed, since the same functionality is now covered by
gdb.cp/local-static.exp.
gdb/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* cp-namespace.c (cp_search_static_and_baseclasses): Handle
function/method scopes; lookup the nested name as a function local
static variable.
gdb/testsuite/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* gdb.base/local-static.exp: Also test with
class::method::variable wholly quoted.
* gdb.cp/m-static.exp (class::method::variable): Remove test.
This commit makes "print S::method()::static_var" actually find the
debug symbol for static_var. Currently, you get:
(gdb) print S::method()::static_var
A syntax error in expression, near `'.
Quoting the whole string would seemingly work before the previous
patch that made GDB stop assuming int for no-debug-info variables:
(gdb) p 'S::method()::static_var'
$1 = 1
... except that's incorrect output, because:
(gdb) ptype 'S::method()::static_var'
type = <data variable, no debug info>
The way to make it work correctly currently is by quoting the
function/method part, like this:
(gdb) print 'S::method()'::static_var
$1 = {i1 = 1, i2 = 2, i3 = 3}
(gdb) ptype 'S::method()'::static_var
type = struct aggregate {
int i1;
int i2;
int i3;
}
At least after the "stop assuming int" patch, this is what we
now get:
(gdb) p 'S::method()::static_var'
'S::method()::static_var' has unknown type; cast it to its declared type
(gdb) p (struct aggregate) 'S::method()::static_var'
$1 = {i1 = 1, i2 = 2, i3 = 3}
However, IMO, users shouldn't really have to care about any of this.
GDB should Just Work, without quoting, IMO.
So here's a patch that implements support for that in the C++ parser.
With this patch, you now get:
(gdb) p S::method()::S_M_s_var_aggregate
$1 = {i1 = 1, i2 = 2, i3 = 3}
(gdb) ptype S::method()::S_M_s_var_aggregate
type = struct aggregate {
int i1;
int i2;
int i3;
}
gdb/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
(%type <voidval>): Add function_method.
* c-exp.y (exp): New production for calls with no arguments.
(function_method, function_method_void_or_typelist): New
productions.
(exp): New production for "method()::static_var".
* eval.c (evaluate_subexp_standard): Handle OP_FUNC_STATIC_VAR.
* expprint.c (print_subexp_standard, dump_subexp_body_standard):
Handle OP_FUNC_STATIC_VAR.
* parse.c (operator_length_standard):
Handle OP_FUNC_STATIC_VAR.
* std-operator.def (OP_FUNC_STATIC_VAR): New.
gdb/testsuite/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* gdb.base/local-static.c: New.
* gdb.base/local-static.cc: New.
* gdb.base/local-static.exp: New.
An earlier commit made GDB no longer assume no-debug-info functions
return int. This commit gives the same treatment to variables.
Currently, you can end misled by GDB over output like this:
(gdb) p var
$1 = -1
(gdb) p /x var
$2 = 0xffffffff
until you realize that GDB is assuming that the variable is an "int",
because:
(gdb) ptype var
type = <data variable, no debug info>
You may try to fix it by casting, but that doesn't really help:
(gdb) p /x (unsigned long long) var
$3 = 0xffffffffffffffff # incorrect
^^
That's incorrect output, because the variable was defined like this:
uint64_t var = 0x7fffffffffffffff;
^^
What happened is that with the cast, GDB did an int -> 'unsigned long
long' conversion instead of reinterpreting the variable as the cast-to
type. To get at the variable properly you have to reinterpret the
variable's address manually instead, with either:
(gdb) p /x *(unsigned long long *) &var
$4 = 0x7fffffffffffffff
(gdb) p /x {unsigned long long} &var
$5 = 0x7fffffffffffffff
After this commit GDB does it for you. This is what you'll get
instead:
(gdb) p var
'var' has unknown type; cast it to its declared type
(gdb) p /x (unsigned long long) var
$1 = 0x7fffffffffffffff
As in the functions patch, the "compile" machinery doesn't currently
have the cast-to type handy, so it continues assuming no-debug
variables have int type, though now at least it warns.
The change to gdb.cp/m-static.exp deserves an explanation:
- gdb_test "print 'gnu_obj_1::method()::sintvar'" "\\$\[0-9\]+ = 4" \
+ gdb_test "print (int) 'gnu_obj_1::method()::sintvar'" "\\$\[0-9\]+ = 4" \
That's printing the "sintvar" function local static of the
"gnu_obj_1::method()" method.
The problem with that test is that that "'S::method()::static_var'"
syntax doesn't really work in C++ as you'd expect. The way to make it
work correctly currently is to quote the method part, not the whole
expression, like:
(gdb) print 'gnu_obj_1::method()'::sintvar
If you wrap the whole expression in quotes, like in m-static.exp, what
really happens is that the parser considers the whole string as a
symbol name, but there's no debug symbol with that name. However,
local statics have linkage and are given a mangled name that demangles
to the same string as the full expression, so that's what GDB prints.
After this commit, and without the cast, the print in m-static.exp
would error out saying that the variable has unknown type:
(gdb) p 'gnu_obj_1::method()::sintvar'
'gnu_obj_1::method()::sintvar' has unknown type; cast it to its declared type
TBC, if currently (even before this series) you try to print any
function local static variable of type other than int, you'll get
bogus results. You can see that with m-static.cc as is, even.
Printing the "svar" local, which is a boolean (1 byte) still prints as
"int" (4 bytes):
(gdb) p 'gnu_obj_1::method()::svar'
$1 = 1
(gdb) ptype 'gnu_obj_1::method()::svar'
type = <data variable, no debug info>
This probably prints some random bogus value on big endian machines.
If 'svar' was of some aggregate type (etc.) we'd still print it as
int, so the problem would have been more obvious... After this
commit, you'll get instead:
(gdb) p 'gnu_obj_1::method()::svar'
'gnu_obj_1::method()::svar' has unknown type; cast it to its declared type
... so at least GDB is no longer misleading. Making GDB find the real
local static debug symbol is the subject of the following patches. In
the end, it'll all "Just Work".
gdb/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* ax-gdb.c: Include "typeprint.h".
(gen_expr_for_cast): New function.
(gen_expr) <OP_CAST, OP_CAST_TYPE>: Use it.
<OP_VAR_VALUE, OP_MSYM_VAR_VALUE>: Error out if the variable's
type is unknown.
* dwarf2read.c (new_symbol_full): Fallback to int instead of
nodebug_data_symbol.
* eval.c: Include "typeprint.h".
(evaluate_subexp_standard) <OP_VAR_VALUE, OP_VAR_MSYM_VALUE>:
Error out if symbol has unknown type.
<UNOP_CAST, UNOP_CAST_TYPE>: Common bits factored out to
evaluate_subexp_for_cast.
(evaluate_subexp_for_address, evaluate_subexp_for_sizeof): Handle
OP_VAR_MSYM_VALUE.
(evaluate_subexp_for_cast): New function.
* gdbtypes.c (init_nodebug_var_type): New function.
(objfile_type): Use it to initialize types of variables with no
debug info.
* typeprint.c (error_unknown_type): New.
* typeprint.h (error_unknown_type): New declaration.
* compile/compile-c-types.c (convert_type_basic): Handle
TYPE_CODE_ERROR; warn and fallback to int for variables with
unknown type.
gdb/testsuite/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* gdb.asm/asm-source.exp: Add casts to int.
* gdb.base/nodebug.c (dataglobal8, dataglobal32_1, dataglobal32_2)
(dataglobal64_1, dataglobal64_2): New globals.
* gdb.base/nodebug.exp: Test different expressions involving the
new globals, with print, whatis and ptype. Add casts to int.
* gdb.base/solib-display.exp: Add casts to int.
* gdb.compile/compile-ifunc.exp: Expect warning. Add cast to int.
* gdb.cp/m-static.exp: Add cast to int.
* gdb.dwarf2/dw2-skip-prologue.exp: Add cast to int.
* gdb.threads/tls-nodebug.exp: Check that gdb errors out printing
tls variable with no debug info without a cast. Test with a cast
to int too.
* gdb.trace/entry-values.exp: Add casts.
The patch to make GDB stop assuming functions return int left GDB with
an inconsistency. While with normal expression evaluation the
"unknown return type" error shows the name of the function that misses
debug info:
(gdb) p getenv ("PATH")
'getenv' has unknown return type; cast the call to its declared return type
^^^^^^
which is handy in more complicated expressions, "ptype" does not:
(gdb) ptype getenv ("PATH")
function has unknown return type; cast the call to its declared return type
^^^^^^^^
This commit builds on the new OP_VAR_MSYM_VALUE to fix it, by making
OP_FUNCALL extract the function name from the symbol stored in
OP_VAR_VALUE/OP_VAR_MSYM_VALUE. We now get the same error in "print"
vs "ptype":
(gdb) ptype getenv()
'getenv' has unknown return type; cast the call to its declared return type
(gdb) p getenv()
'getenv' has unknown return type; cast the call to its declared return type
gdb/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* eval.c (evaluate_subexp_standard): <OP_FUNCALL>: Extract
function name from symbol/minsym and pass it to
error_call_unknown_return_type.
gdb/testsuite/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* gdb.base/nodebug.exp: Test that ptype's error about functions
with unknown return type includes the function name too.
The fact that GDB defaults to assuming that functions return int, when
it has no debug info for the function has been a recurring source of
user confusion. Recently this came up on the errno pretty printer
discussions. Shortly after, it came up again on IRC, with someone
wondering why does getenv() in GDB return a negative int:
(gdb) p getenv("PATH")
$1 = -6185
This question (with s/getenv/random-other-C-runtime-function) is a FAQ
on IRC.
The reason for the above is:
(gdb) p getenv
$2 = {<text variable, no debug info>} 0x7ffff7751d80 <getenv>
(gdb) ptype getenv
type = int ()
... which means that GDB truncated the 64-bit pointer that is actually
returned from getent to 32-bit, and then sign-extended it:
(gdb) p /x -6185
$6 = 0xffffe7d7
The workaround is to cast the function to the right type, like:
(gdb) p ((char *(*) (const char *)) getenv) ("PATH")
$3 = 0x7fffffffe7d7 "/usr/local/bin:/"...
IMO, we should do better than this.
I see the "assume-int" issue the same way I see printing bogus values
for optimized-out variables instead of "<optimized out>" -- I'd much
rather that the debugger tells me "I don't know" and tells me how to
fix it than showing me bogus misleading results, making me go around
tilting at windmills.
If GDB prints a signed integer when you're expecting a pointer or
aggregate, you at least have some sense that something is off, but
consider the case of the function actually returning a 64-bit integer.
For example, compile this without debug info:
unsigned long long
function ()
{
return 0x7fffffffffffffff;
}
Currently, with pristine GDB, you get:
(gdb) p function ()
$1 = -1 # incorrect
(gdb) p /x function ()
$2 = 0xffffffff # incorrect
maybe after spending a few hours debugging you suspect something is
wrong with that -1, and do:
(gdb) ptype function
type = int ()
and maybe, just maybe, you realize that the function actually returns
unsigned long long. And you try to fix it with:
(gdb) p /x (unsigned long long) function ()
$3 = 0xffffffffffffffff # incorrect
... which still produces the wrong result, because GDB simply applied
int to unsigned long long conversion. Meaning, it sign-extended the
integer that it extracted from the return of the function, to 64-bits.
and then maybe, after asking around on IRC, you realize you have to
cast the function to a pointer of the right type, and call that. It
won't be easy, but after a few missteps, you'll get to it:
..... (gdb) p /x ((unsigned long long(*) ()) function) ()
$666 = 0x7fffffffffffffff # finally! :-)
So to improve on the user experience, this patch does the following
(interrelated) things:
- makes no-debug-info functions no longer default to "int" as return
type. Instead, they're left with NULL/"<unknown return type>"
return type.
(gdb) ptype getenv
type = <unknown return type> ()
- makes calling a function with unknown return type an error.
(gdb) p getenv ("PATH")
'getenv' has unknown return type; cast the call to its declared return type
- and then to make it easier to call the function, makes it possible
to _only_ cast the return of the function to the right type,
instead of having to cast the function to a function pointer:
(gdb) p (char *) getenv ("PATH") # now Just Works
$3 = 0x7fffffffe7d7 "/usr/local/bin:/"...
(gdb) p ((char *(*) (const char *)) getenv) ("PATH") # continues working
$4 = 0x7fffffffe7d7 "/usr/local/bin:/"...
I.e., it makes GDB default the function's return type to the type
of the cast, and the function's parameters to the type of the
arguments passed down.
After this patch, here's what you'll get for the "unsigned long long"
example above:
(gdb) p function ()
'function' has unknown return type; cast the call to its declared return type
(gdb) p /x (unsigned long long) function ()
$4 = 0x7fffffffffffffff # correct!
Note that while with "print" GDB shows the name of the function that
has the problem:
(gdb) p getenv ("PATH")
'getenv' has unknown return type; cast the call to its declared return type
which can by handy in more complicated expressions, "ptype" does not:
(gdb) ptype getenv ("PATH")
function has unknown return type; cast the call to its declared return type
This will be fixed in the next patch.
gdb/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* ada-lang.c (ada_evaluate_subexp) <TYPE_CODE_FUNC>: Don't handle
TYPE_GNU_IFUNC specially here. Throw error if return type is
unknown.
* ada-typeprint.c (print_func_type): Handle functions with unknown
return type.
* c-typeprint.c (c_type_print_base): Handle functions and methods
with unknown return type.
* compile/compile-c-symbols.c (convert_symbol_bmsym)
<mst_text_gnu_ifunc>: Use nodebug_text_gnu_ifunc_symbol.
* compile/compile-c-types.c: Include "objfiles.h".
(convert_func): For functions with unknown return type, warn and
default to int.
* compile/compile-object-run.c (compile_object_run): Adjust call
to call_function_by_hand_dummy.
* elfread.c (elf_gnu_ifunc_resolve_addr): Adjust call to
call_function_by_hand.
* eval.c (evaluate_subexp_standard): Adjust calls to
call_function_by_hand. Handle functions and methods with unknown
return type. Pass expect_type to call_function_by_hand.
* f-typeprint.c (f_type_print_base): Handle functions with unknown
return type.
* gcore.c (call_target_sbrk): Adjust call to
call_function_by_hand.
* gdbtypes.c (objfile_type): Leave nodebug text symbol with NULL
return type instead of int. Make nodebug_text_gnu_ifunc_symbol be
an integer address type instead of nodebug.
* guile/scm-value.c (gdbscm_value_call): Adjust call to
call_function_by_hand.
* infcall.c (error_call_unknown_return_type): New function.
(call_function_by_hand): New "default_return_type" parameter.
Pass it down.
(call_function_by_hand_dummy): New "default_return_type"
parameter. Use it instead of defaulting to int. If there's no
default and the return type is unknown, throw an error. If
there's a default return type, and the called function has no
debug info, then assume the function is prototyped.
* infcall.h (call_function_by_hand, call_function_by_hand_dummy):
New "default_return_type" parameter.
(error_call_unknown_return_type): New declaration.
* linux-fork.c (call_lseek): Cast return type of lseek.
(inferior_call_waitpid, checkpoint_command): Adjust calls to
call_function_by_hand.
* linux-tdep.c (linux_infcall_mmap, linux_infcall_munmap): Adjust
calls to call_function_by_hand.
* m2-typeprint.c (m2_procedure): Handle functions with unknown
return type.
* objc-lang.c (lookup_objc_class, lookup_child_selector)
(value_nsstring, print_object_command): Adjust calls to
call_function_by_hand.
* p-typeprint.c (pascal_type_print_varspec_prefix): Handle
functions with unknown return type.
(pascal_type_print_func_varspec_suffix): New function.
(pascal_type_print_varspec_suffix) <TYPE_CODE_FUNC,
TYPE_CODE_METHOD>: Use it.
* python/py-value.c (valpy_call): Adjust call to
call_function_by_hand.
* rust-lang.c (rust_evaluate_funcall): Adjust call to
call_function_by_hand.
* valarith.c (value_x_binop, value_x_unop): Adjust calls to
call_function_by_hand.
* valops.c (value_allocate_space_in_inferior): Adjust call to
call_function_by_hand.
* typeprint.c (type_print_unknown_return_type): New function.
* typeprint.h (type_print_unknown_return_type): New declaration.
gdb/testsuite/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* gdb.base/break-main-file-remove-fail.exp (test_remove_bp): Cast
return type of munmap in infcall.
* gdb.base/break-probes.exp: Cast return type of foo in infcall.
* gdb.base/checkpoint.exp: Simplify using for loop. Cast return
type of ftell in infcall.
* gdb.base/dprintf-detach.exp (dprintf_detach_test): Cast return
type of getpid in infcall.
* gdb.base/infcall-exec.exp: Cast return type of execlp in
infcall.
* gdb.base/info-os.exp: Cast return type of getpid in infcall.
Bail on failure to extract the pid.
* gdb.base/nodebug.c: #include <stdint.h>.
(multf, multf_noproto, mult, mult_noproto, add8, add8_noproto):
New functions.
* gdb.base/nodebug.exp (test_call_promotion): New procedure.
Change expected output of print/whatis/ptype with functions with
no debug info. Test all supported languages. Call
test_call_promotion.
* gdb.compile/compile.exp: Adjust expected output to expect
warning.
* gdb.threads/siginfo-threads.exp: Likewise.
Calling a prototyped function via a function pointer with the right
prototype doesn't work correctly, if the called function requires
argument coercion... Like, e.g., with:
float mult (float f1, float f2) { return f1 * f2; }
(gdb) p mult (2, 3.5)
$1 = 7
(gdb) p ((float (*) (float, float)) mult) (2, 3.5)
$2 = 0
both calls should have returned the same, of course. The problem is
that GDB misses marking the type of the function pointer target as
prototyped...
Without the fix, the new test fails like this:
(gdb) p ((int (*) (float, float)) t_float_values2)(3.14159,float_val2)
$30 = 0
(gdb) FAIL: gdb.base/callfuncs.exp: p ((int (*) (float, float)) t_float_values2)(3.14159,float_val2)
gdb/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* gdbtypes.c (lookup_function_type_with_arguments): Mark function
types with more than one parameter as prototyped.
gdb/testsuite/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* gdb.base/callfuncs.exp (do_function_calls): New parameter
"prototypes". Test calling float functions via prototyped and
unprototyped function pointers.
(perform_all_tests): New parameter "prototypes". Pass it down.
(top level): Pass down "prototypes" parameter to
perform_all_tests.
This patch improves the loop_break and loop_continue tests to verify
that they work as expected when multiple loops are nested (they affect
the inner loop).
gdb/testsuite/ChangeLog:
* gdb.base/commands.exp (loop_break_test, loop_continue_test):
Test with nested loops.
I grepped the testsuite for loop_break and loop_continue and didn't find
anything, so I wrote some simple tests for those.
gdb/testsuite/ChangeLog:
* gdb.base/commands.exp: Call the new procedures.
(loop_break_test, loop_continue_test): New procedures.
When using "if" (or while) without args directly on gdb's command line,
you get this:
(gdb) if
if/while commands require arguments
When doing the same when entering a command list, you only get an error
when the command is executed, when parse_exp_in_context_1 fails to
evaluate the expression.
(gdb) define foo
Type commands for definition of "foo".
End with a line saying just "end".
>if
>end
>end
(gdb) foo
Argument required (expression to compute).
I think it would make more sense to error out when inputting the command
list directly:
(gdb) define foo
Type commands for definition of "foo".
End with a line saying just "end".
>if
if/while commands require arguments.
The only required change is to check whether args is an empty string in
build_command_line.
gdb/ChangeLog:
* cli/cli-script.c (build_command_line): For if/while commands,
check whether args is empty.
gdb/testsuite/ChangeLog:
* gdb.base/commands.exp: Call new procedure.
(define_if_without_arg_test): New procedure.
Currently, with "list LINESPEC1,LINESPEC2", if one of the linespecs is
ambiguous, i.e., if it expands to multiple locations, you get this
seemingly odd output:
(gdb) list foo,bar
file: "file0.c", line number: 26
file: "file1.c", line number: 29
Since "foo" above expands to multiple locations, the specified range
is indeterminate, and GDB is trying to be helpful by showing you what
was ambiguous. It looks confusing to me, though. I think it'd be
much more user friendly if GDB actually told you that, like this:
(gdb) list foo,bar
Specified first line 'foo' is ambiguous:
file: "file0.c", line number: 26
file: "file1.c", line number: 29
(gdb) list bar,foo
Specified last line 'foo' is ambiguous:
file: "file0.c", line number: 26
file: "file1.c", line number: 29
Note, I'm using "first" and "last" in the output because that's what
the manual uses:
~~~
list first,last
Print lines from first to last. [...]
~~~
Tested on x86-64 GNU/Linux.
gdb/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* cli/cli-cmds.c (edit_command): Pass message to
ambiguous_line_spec.
(list_command): Pass message to ambiguous_line_spec. Say
"first"/"last" instead of "start" and "end" to be consistent with
the manual.
(ambiguous_line_spec): Add 'format' and vararg parameters. Use
them to print formatted message.
gdb/testsuite/ChangeLog:
2017-09-04 Pedro Alves <palves@redhat.com>
* gdb.base/list-ambiguous.exp: New file.
* gdb.base/list-ambiguous0.c: New file.
* gdb.base/list-ambiguous1.c: New file.
* gdb.base/list.exp (test_list_range): Adjust expected output.
This patch implements the ability to set/unset environment variables
on the remote target, mimicking what GDB already offers to the user.
There are two features present here: user-set and user-unset
environment variables.
User-set environment variables are only the variables that are
explicitly set by the user, using the 'set environment' command. This
means that variables that were already present in the environment when
starting GDB/GDBserver are not transmitted/considered by this feature.
User-unset environment variables are variables that are explicitly
unset by the user, using the 'unset environment' command.
The idea behind this patch is to store user-set and user-unset
environment variables in two separate sets, both part of gdb_environ.
Then, when extended_remote_create_inferior is preparing to start the
inferior, it will iterate over the two sets and set/unset variables
accordingly. Three new packets are introduced:
- QEnvironmentHexEncoded, which is used to set environment variables,
and contains an hex-encoded string in the format "VAR=VALUE" (VALUE
can be empty if the user set a variable with a null value, by doing
'set environment VAR=').
- QEnvironmentUnset, which is used to unset environment variables, and
contains an hex-encoded string in the format "VAR".
- QEnvironmentReset, which is always the first packet to be
transmitted, and is used to reset the environment, i.e., discard any
changes made by the user on previous runs.
The QEnvironmentHexEncoded packet is inspired on LLDB's extensions to
the RSP. Details about it can be seen here:
<https://raw.githubusercontent.com/llvm-mirror/lldb/master/docs/lldb-gdb-remote.txt>
I decided not to implement the QEnvironment packet because it is
considered deprecated by LLDB. This packet, on LLDB, serves the same
purpose of QEnvironmentHexEncoded, but sends the information using a
plain text, non-hex-encoded string.
The other two packets are new.
This patch also includes updates to the documentation, testsuite, and
unit tests, without introducing regressions.
gdb/ChangeLog:
2017-08-31 Sergio Durigan Junior <sergiodj@redhat.com>
* NEWS (Changes since GDB 8.0): Add entry mentioning new support
for setting/unsetting environment variables on the remote target.
(New remote packets): Add entries for QEnvironmentHexEncoded,
QEnvironmentUnset and QEnvironmentReset.
* common/environ.c (gdb_environ::operator=): Extend method to
handle m_user_set_env_list and m_user_unset_env_list.
(gdb_environ::clear): Likewise.
(match_var_in_string): Change type of first parameter from 'char
*' to 'const char *'.
(gdb_environ::set): Extend method to handle
m_user_set_env_list and m_user_unset_env_list.
(gdb_environ::unset): Likewise.
(gdb_environ::clear_user_set_env): New method.
(gdb_environ::user_set_envp): Likewise.
(gdb_environ::user_unset_envp): Likewise.
* common/environ.h (gdb_environ): Handle m_user_set_env_list and
m_user_unset_env_list on move constructor/assignment.
(unset): Add new default parameter 'update_unset_list = true'.
(clear_user_set_env): New method.
(user_set_envp): Likewise.
(user_unset_envp): Likewise.
(m_user_set_env_list): New std::set.
(m_user_unset_env_list): Likewise.
* common/rsp-low.c (hex2str): New function.
(bin2hex): New overload for bin2hex function.
* common/rsp-low.c (hex2str): New prototype.
(str2hex): New overload prototype.
* remote.c: Include "environ.h". Add QEnvironmentHexEncoded,
QEnvironmentUnset and QEnvironmentReset.
(remote_protocol_features): Add QEnvironmentHexEncoded,
QEnvironmentUnset and QEnvironmentReset packets.
(send_environment_packet): New function.
(extended_remote_environment_support): Likewise.
(extended_remote_create_inferior): Call
extended_remote_environment_support.
(_initialize_remote): Add QEnvironmentHexEncoded,
QEnvironmentUnset and QEnvironmentReset packet configs.
* unittests/environ-selftests.c (gdb_selftest_env_var):
New variable.
(test_vector_initialization): New function.
(test_init_from_host_environ): Likewise.
(test_reinit_from_host_environ): Likewise.
(test_set_A_unset_B_unset_A_cannot_find_A_can_find_B):
Likewise.
(test_unset_set_empty_vector): Likewise.
(test_vector_clear): Likewise.
(test_std_move): Likewise.
(test_move_constructor):
(test_self_move): Likewise.
(test_set_unset_reset): Likewise.
(run_tests): Rewrite in terms of the functions above.
gdb/gdbserver/ChangeLog:
2017-08-31 Sergio Durigan Junior <sergiodj@redhat.com>
* server.c (handle_general_set): Handle QEnvironmentHexEncoded,
QEnvironmentUnset and QEnvironmentReset packets.
(handle_query): Inform remote that QEnvironmentHexEncoded,
QEnvironmentUnset and QEnvironmentReset are supported.
gdb/doc/ChangeLog:
2017-08-31 Sergio Durigan Junior <sergiodj@redhat.com>
* gdb.texinfo (set environment): Add @anchor. Explain that
environment variables set by the user are sent to GDBserver.
(unset environment): Likewise, but for unsetting variables.
(Connecting) <Remote Packet>: Add "environment-hex-encoded",
"QEnvironmentHexEncoded", "environment-unset", "QEnvironmentUnset",
"environment-reset" and "QEnvironmentReset" to the table.
(Remote Protocol) <QEnvironmentHexEncoded, QEnvironmentUnset,
QEnvironmentReset>: New item, explaining the packet.
gdb/testsuite/ChangeLog:
2017-08-31 Sergio Durigan Junior <sergiodj@redhat.com>
* gdb.base/share-env-with-gdbserver.c: New file.
* gdb.base/share-env-with-gdbserver.exp: Likewise.
Commit
Command names: make them case sensitive
3d7b173c29900879c9a5958dd6029fd36666e57c
made command name lookup case sensitive. However, define_command, used
when creating a user-defined command, converts the command name to
lowercase, assuming that the command name lookup works in a case
insensitive way. This causes user-defined commands with capital letters
in their name to only be callable with a lowercase version:
(gdb) define Foo
Type commands for definition of "Foo".
End with a line saying just "end".
>print 1
>end
(gdb) Foo
Undefined command: "Foo". Try "help".
(gdb) foo
$1 = 1
This patch removes that conversion to lowercase, so that the user can
call the command with the same name they provided.
gdb/ChangeLog:
* cli/cli-script.c (define_command): Don't convert command name
to lower case.
gdb/testsuite/ChangeLog:
* gdb.base/commands.exp (user_defined_command_case_sensitivity):
New proc, call it from toplevel.
There is an assertion that is triggering when we start GDB and
instruct it to debug a remote inferior, but don't provide a local
binary, like:
./gdb -nx -q --data-directory=data-directory -ex "tar ext :1234" \
-ex "set remote exec-file /bin/ls" -ex r
In this case, when calling exec_file_locate_attach to locate the
inferior, GDB is incorrectly resetting the breakpoints without a
thread/inferior even running, which causes an assertion to be
triggered:
binutils-gdb/gdb/thread.c:1609: internal-error: scoped_restore_current_thread::scoped_restore_current_thread(): Assertion `tp != NULL' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
Quit this debugging session? (y or n)
This happens because add_current_inferior_and_thread (on remote.c) is
breaking an invariant: making inferior_ptid point to a non-existing
thread and then calling common code, which in this case is
breakpoint_re_set. The fix is to make sure that inferior_ptid points
to null_ptid if there is no thread present.
A testcase is provided. Regtested on buildbot.
gdb/ChangeLog:
2017-08-23 Pedro Alves <palves@redhat.com>
PR remote/21852
* remote.c (add_current_inferior_and_thread): Set inferior_ptid
to null_ptid and switch to thread without reading the registers
after adding the inferior.
gdb/testsuite/ChangeLog:
2017-08-23 Sergio Durigan Junior <sergiodj@redhat.com>
PR remote/21852
* gdb.server/normal.c: New file, copied from gdb.base.
* gdb.server/run-without-local-binary.exp: New file.
The M7 processor supports an Application Data Integrity (ADI) feature
that detects invalid data accesses. When software allocates data, it
chooses a 4-bit version number, sets the version in the upper 4 bits
of the 64-bit pointer to that data, and stores the 4-bit version in
every cacheline of the object. Hardware saves the latter in spare
bits in the cache and memory hierarchy. On each load and store, the
processor compares the upper 4 VA (virtual address) bits to the
cacheline's version. If there is a mismatch, the processor generates a
version mismatch trap which can be either precise or disrupting. The
trap is an error condition which the kernel delivers to the process as
a SIGSEGV signal.
The upper 4 bits of the VA represent a version and are not part of the
true address. The processor clears these bits and sign extends bit 59
to generate the true address.
Note that 32-bit applications cannot use ADI.
This patch adds ADI support in gdb which allows the user to examine
current version tags and assign new version tags in the program. It
also catches and reports precise or disrupting memory corruption
traps.
gdb/ChangeLog:
2017-08-07 Weimin Pan <weimin.pan@oracle.com>
* sparc64-tdep.h: (adi_normalize_address): New export.
* sparc-nat.h: (open_adi_tag_fd): New export.
* sparc64-linux-nat.c: (open_adi_tag_fd): New function.
* sparc64-linux-tdep.c:
(SEGV_ACCADI, SEGV_ADIDERR, SEGV_ADIPERR) New defines.
(sparc64_linux_handle_segmentation_fault): New function.
(sparc64_linux_init_abi): Register
sparc64_linux_handle_segmentation_fault
* sparc64-tdep.c: Include cli-utils.h,gdbcmd.h,auxv.h.
(sparc64_addr_bits_remove): New function.
(sparc64_init_abi): Register sparc64_addr_bits_remove.
(MAX_PROC_NAME_SIZE): New macro.
(AT_ADI_BLKSZ, AT_ADI_NBITS, AT_ADI_UEONADI) New defines.
(sparc64adilist): New variable.
(adi_proc_list): New variable.
(find_adi_info): New function.
(add_adi_info): New function.
(get_adi_info_proc): New function.
(get_adi_info): New function.
(info_adi_command): New function.
(read_maps_entry): New function.
(adi_available): New function.
(adi_normalize_address): New function.
(adi_align_address): New function.
(adi_convert_byte_count): New function.
(adi_tag_fd): New function.
(adi_is_addr_mapped): New function.
(adi_read_versions): New function.
(adi_write_versions): New function.
(adi_print_versions): New function.
(do_examine): New function.
(do_assign): New function.
(adi_examine_command): New function.
(adi_assign_command): New function.
(_initialize_sparc64_adi_tdep): New function.
gdb/doc/ChangeLog:
2017-08-07 Weimin Pan <weimin.pan@oracle.com>
* gdb.texinfo (Architectures): Add new Sparc64 section to document
ADI support.
* NEWS: Add "adi examine" and "adi assign" commands.
gdb/testsuite/ChangeLog:
2017-08-07 Weimin Pan <weimin.pan@oracle.com>
* gdb.arch/sparc64-adi.exp: New file.
* gdb.arch/sparc64-adi.c: New file.
This adds a test for the "list" command change done in 0d999a6ef0f9
("List actual code around more than one location").
gdb/ChangeLog:
2017-08-22 Pedro Alves <palves@redhat.com>
* gdb.cp/overload.exp (line_range_pattern): New procedure.
(top level): Add "list all overloads" tests.
This changes gdb_realpath to return a unique_xmalloc_ptr and fixes up
the callers. This allows removing some cleanups. This change by
itself caused xfullpath.exp to fail; and attempting to fix that ran
into various problems (like .get() being optimized out); so this patch
also rewrites xfullpath.exp to be a C++ selftest instead.
ChangeLog
2017-08-22 Tom Tromey <tom@tromey.com>
* exec.c (exec_file_attach): Update.
* linux-thread-db.c (try_thread_db_load): Update.
* guile/scm-safe-call.c (gdbscm_safe_source_script): Update.
* utils.c (gdb_realpath): Change return type.
(gdb_realpath_keepfile): Update.
(gdb_realpath_check_trailer, gdb_realpath_tests): New functions.
(_initialize_utils): Register the new self test.
* source.c (openp): Update.
(find_and_open_source): Update.
* nto-tdep.c (nto_find_and_open_solib): Update.
* main.c (set_gdb_data_directory): Update.
(captured_main_1): Update.
* dwarf2read.c (dwarf2_get_dwz_file): Update
(dw2_map_symbol_filenames): Update.
* auto-load.c (auto_load_safe_path_vec_update): Update.
(filename_is_in_auto_load_safe_path_vec): Change type of
"filename_realp".
(auto_load_objfile_script): Update.
(file_is_auto_load_safe): Update. Use std::string.
* utils.h (gdb_realpath): Return a gdb::unique_xmalloc_ptr.
testsuite/ChangeLog
2017-08-22 Tom Tromey <tom@tromey.com>
* gdb.gdb/xfullpath.exp: Remove.
(Ref: https://sourceware.org/ml/gdb/2017-06/msg00048.html)
This patch improves GDB support for function aliases defined with
__attribute__ alias. For example, in the test added by this commit,
there is no reference to "func_alias" in the debug info at all, only
to "func"'s definition:
$ nm ./testsuite/outputs/gdb.base/symbol-alias/symbol-alias | grep " func"
00000000004005ae t func
00000000004005ae T func_alias
$ readelf -w ./testsuite/outputs/gdb.base/symbol-alias/symbol-alias | grep func -B 1 -A 8
<1><db>: Abbrev Number: 5 (DW_TAG_subprogram)
<dc> DW_AT_name : (indirect string, offset: 0x111): func
<e0> DW_AT_decl_file : 1
<e1> DW_AT_decl_line : 27
<e2> DW_AT_prototyped : 1
<e2> DW_AT_type : <0xf8>
<e6> DW_AT_low_pc : 0x4005ae
<ee> DW_AT_high_pc : 0xb
<f6> DW_AT_frame_base : 1 byte block: 9c (DW_OP_call_frame_cfa)
<f8> DW_AT_GNU_all_call_sites: 1
So all GDB knows about "func_alias" is from the minsym (elf symbol):
(gdb) p func_alias
$1 = {<text variable, no debug info>} 0x4005ae <func>
(gdb) ptype func_alias
type = int ()
(gdb) p func
$2 = {struct S *(void)} 0x4005ae <func>
(gdb) ptype func
type = struct S {
int field1;
int field2;
} *(void)
The result is that calling func_alias from the command line produces
incorrect results.
This is similar (though not exactly the same) to the glibc
errno/__errno_location/__GI___errno_location situation. On glibc,
errno is defined like this:
extern int *__errno_location (void);
#define errno (*__errno_location ())
with __GI___errno_location being an internal alias for
__errno_location. On my system's libc (F23), I do see debug info for
__errno_location, in the form of name vs linkage name:
<1><95a5>: Abbrev Number: 18 (DW_TAG_subprogram)
<95a6> DW_AT_external : 1
<95a6> DW_AT_name : (indirect string, offset: 0x2c26): __errno_location
<95aa> DW_AT_decl_file : 1
<95ab> DW_AT_decl_line : 24
<95ac> DW_AT_linkage_name: (indirect string, offset: 0x2c21): __GI___errno_location
<95b0> DW_AT_prototyped : 1
<95b0> DW_AT_type : <0x9206>
<95b4> DW_AT_low_pc : 0x20f40
<95bc> DW_AT_high_pc : 0x11
<95c4> DW_AT_frame_base : 1 byte block: 9c (DW_OP_call_frame_cfa)
<95c6> DW_AT_GNU_all_call_sites: 1
however that doesn't matter in practice, because GDB doesn't record
demangled names anyway, and so we end up with the exact same situation
covered by the testcase.
So the fix is to make the expression parser find a debug symbol for
the same address as the just-found minsym, when a lookup by name
didn't find a debug symbol by name. We now get:
(gdb) p func_alias
$1 = {struct S *(void)} 0x4005ae <func>
(gdb) p __errno_location
$2 = {int *(void)} 0x7ffff6e92830 <__errno_location>
I've made the test exercise variable aliases too, for completeness.
Those already work correctly, because unlike for function aliases, GCC
emits debug information for variable aliases.
Tested on GNU/Linux.
gdb/ChangeLog:
2017-08-21 Pedro Alves <palves@redhat.com>
PR gdb/19487
* c-exp.y (variable production): Handle function aliases.
* minsyms.c (msymbol_is_text): New function.
* minsyms.h (msymbol_is_text): Declare.
* symtab.c (find_function_alias_target): New function.
* symtab.h (find_function_alias_target): Declare.
gdb/testsuite/ChangeLog:
2017-08-21 Pedro Alves <palves@redhat.com>
PR gdb/19487
* gdb.base/symbol-alias.c: New.
* gdb.base/symbol-alias2.c: New.
* gdb.base/symbol-alias.exp: New.
(Ref: https://sourceware.org/ml/gdb/2017-06/msg00020.html)
Assuming int_t is a typedef to int:
typedef int int_t;
gdb currently loses this expression's typedef:
(gdb) p (int_t) 0
$1 = 0
(gdb) whatis $1
type = int
or:
(gdb) whatis (int_t) 0
type = int
or, to get "whatis" out of the way:
(gdb) maint print type (int_t) 0
...
name 'int'
code 0x8 (TYPE_CODE_INT)
...
This prevents a type printer for "int_t" kicking in, with e.g.:
(gdb) p (int_t) 0
From the manual, we can see that that "whatis (int_t) 0" command
invocation should have printed "type = int_t":
If @var{arg} is a variable or an expression, @code{whatis} prints its
literal type as it is used in the source code. If the type was
defined using a @code{typedef}, @code{whatis} will @emph{not} print
the data type underlying the @code{typedef}.
(...)
If @var{arg} is a type name that was defined using @code{typedef},
@code{whatis} @dfn{unrolls} only one level of that @code{typedef}.
That one-level stripping is currently done here, in
gdb/eval.c:evaluate_subexp_standard, handling OP_TYPE:
...
else if (noside == EVAL_AVOID_SIDE_EFFECTS)
{
struct type *type = exp->elts[pc + 1].type;
/* If this is a typedef, then find its immediate target. We
use check_typedef to resolve stubs, but we ignore its
result because we do not want to dig past all
typedefs. */
check_typedef (type);
if (TYPE_CODE (type) == TYPE_CODE_TYPEDEF)
type = TYPE_TARGET_TYPE (type);
return allocate_value (type);
}
However, this stripping is reachable in both:
#1 - (gdb) whatis (int_t)0 # ARG is an expression with a cast to
# typedef type.
#2 - (gdb) whatis int_t # ARG is a type name.
while only case #2 should strip the typedef. Removing that code from
evaluate_subexp_standard is part of the fix. Instead, we make the
"whatis" command implementation itself strip one level of typedefs
when the command argument is a type name.
We then run into another problem, also fixed by this commit:
value_cast always drops any typedefs of the destination type.
With all that fixed, "whatis (int_t) 0" now works as expected:
(gdb) whatis int_t
type = int
(gdb) whatis (int_t)0
type = int_t
value_cast has many different exit/convertion paths, for handling many
different kinds of casts/conversions, and most of them had to be
tweaked to construct the value of the right "to" type. The new tests
try to exercise most of it, by trying castin of many different
combinations of types. With:
$ make check TESTS="*/whatis-ptype*.exp */gnu_vector.exp */dfp-test.exp"
... due to combinatorial explosion, the testsuite results for the
tests above alone grow like:
- # of expected passes 246
+ # of expected passes 3811
You'll note that the tests exposed one GCC buglet, filed here:
Missing DW_AT_type in DW_TAG_typedef of "typedef of typedef of void"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81267
gdb/ChangeLog:
2017-08-21 Pedro Alves <palves@redhat.com>
* eval.c (evaluate_subexp_standard) <OP_TYPE>: Don't dig past
typedefs.
* typeprint.c (whatis_exp): If handling "whatis", and expression
is OP_TYPE, strip one typedef level. Otherwise don't strip
typedefs here.
* valops.c (value_cast): Save "to" type before resolving
stubs/typedefs. Use that type as resulting value's type.
gdb/testsuite/ChangeLog:
2017-08-21 Pedro Alves <palves@redhat.com>
* gdb.base/dfp-test.c
(d32_t, d64_t, d128_t, d32_t2, d64_t2, d128_t2, v_d32_t, v_d64_t)
(v_d128_t, v_d32_t2, v_d64_t2, v_d128_t2): New.
* gdb.base/dfp-test.exp: Add whatis/ptype/cast tests.
* gdb.base/gnu_vector.exp: Add whatis/ptype/cast tests.
* gdb.base/whatis-ptype-typedefs.c: New.
* gdb.base/whatis-ptype-typedefs.exp: New.
* gdb.python/py-prettyprint.c (int_type, int_type2): New typedefs.
(an_int, an_int_type, an_int_type2): New globals.
* gdb.python/py-prettyprint.exp (run_lang_tests): Add tests
involving typedefs and cast expressions.
* gdb.python/py-prettyprint.py (class pp_int_typedef): New.
(lookup_typedefs_function): New.
(typedefs_pretty_printers_dict): New.
(top level): Register lookup_typedefs_function in
gdb.pretty_printers.
This patch uses GDB self test in GDBserver. The self tests are run if
GDBserver is started with option --selftest.
gdb:
2017-08-18 Yao Qi <yao.qi@linaro.org>
* NEWS: Mention GDBserver's new option "--selftest".
* Makefile.in (SFILES): Remove selftest.c, add common/selftest.c.
* selftest.c: Move it to common/selftest.c.
* selftest.h: Move it to common/selftest.h.
* selftest-arch.c (reset): New function.
(tests_with_arch): Call reset.
gdb/gdbserver:
2017-08-18 Yao Qi <yao.qi@linaro.org>
* Makefile.in (OBS): Add selftest.o.
* configure.ac: AC_DEFINE GDB_SELF_TEST if $development.
* configure, config.in: Re-generated.
* server.c: Include common/sefltest.h.
(captured_main): Handle option --selftest.
gdb/testsuite:
2017-08-18 Yao Qi <yao.qi@linaro.org>
* gdb.server/unittest.exp: New.
gdb/doc:
2017-08-18 Yao Qi <yao.qi@linaro.org>
* gdb.texinfo (Server): Document "--selftest".
When I made commit 9a6c7d9c021cfeb290d76584db7a01e57e7c3d4e, which
C++-fied gdb/common/environ.[ch], I mistakenly altered the behaviour
of the 'unset environment' command. This command, which should delete
all environment variables, is now resetting the list of variables to
the state they were when GDB was started.
This commit fixes this regression, and also adds a test on
gdb.base/environ.exp which really checks if 'unset environment'
worked.
gdb/ChangeLog:
2017-08-15 Sergio Durigan Junior <sergiodj@redhat.com>
PR gdb/21954
* infcmd.c (unset_environment_command): Use the 'clear' method on
the environment instead of resetting it.
gdb/testsuite/ChangeLog:
2017-08-15 Sergio Durigan Junior <sergiodj@redhat.com>
PR gdb/21954
* gdb.base/environ.exp: Add test to check if 'unset environment'
works.
Newer GCC versions yield colored diagnostic messages by default, which may
be useful when executing GDB interactively from a terminal. But when run
from a GDB test case, the compiler output is written into gdb.log, where
such escape sequences are usually more inhibiting than helpful to the
evaluation of test results. So this patch suppresses that.
gdb/testsuite/ChangeLog:
* lib/gdb.exp (universal_compile_options): New caching proc.
(gdb_compile): Suppress GCC's coloring of messages.
PR gdb/21675 points out a few regressions in scalar printing.
One type of regression is due to not carrying over the old handling of
floating point printing -- where a format like "/d" causes a floating
point number to first be cast to a signed integer. This patch restores
this behavior.
The other regression is a longstanding bug in print_octal_chars: one of
the constants was wrong. This patch fixes the constant and adds static
asserts to help catch this sort of error.
ChangeLog
2017-08-14 Tom Tromey <tom@tromey.com>
PR gdb/21675
* valprint.c (LOW_ZERO): Change value to 034.
(print_octal_chars): Add static_asserts for octal constants.
* printcmd.c (print_scalar_formatted): Add 'd' case.
testsuite/ChangeLog
2017-08-14 Tom Tromey <tom@tromey.com>
PR gdb/21675:
* gdb.base/printcmds.exp (test_radices): New function.
* gdb.dwarf2/var-access.exp: Use p/u, not p/d.
* gdb.base/sizeof.exp (check_valueof): Use p/d.
* lib/gdb.exp (get_integer_valueof): Use p/d.
The get_valueof procedure allows tests to conveniently make gdb evaluate
an expression an return the value as a string. However, it includes an
end-of-line character in its result. I stumbled on this when trying to
use that result as part of a regex further in a test.
You can see this for example by adding a puts in
gdb.dwarf2/implref-struct.exp:get_members:
set members [get_valueof "" ${var} ""]
puts "<$members>"
The output is
<{a = 0, b = 1, c = 2}
>
This is because the regex in get_valueof is too greedy, the captured
portion matches anything up to the gdb_prompt, including the end of line
characters. This patch changes it to capture everything but end of line
characters.
The output of the puts becomes:
<{a = 0, b = 1, c = 2}>
I tested this by running gdb.dwarf2/implref-array.exp and
gdb.dwarf2/implref-struct.exp, the two only current users of that
procedure.
gdb/testsuite/ChangeLog:
* lib/gdb.exp (get_valueof): Don't capture end-of-line
characters.
I ran the gdb.rust tests against Rust 1.20 (beta) and saw a few
failures. The failures all came because a particular item moved to a
different module. Since the particular choice of module name isn't
important here, I simply widened the allowable results.
Tested locally against rustc 1.19, 1.20, and 1.21.
testsuite/ChangeLog
2017-08-05 Tom Tromey <tom@tromey.com>
* gdb.rust/simple.exp: Allow String to appear in a different
namespace.
Now, GDB is able to dynamically create i386-linux target descriptions
from features, instead of using pre-generated target descriptions. These
pre-generated target descriptions are no longer used by GDB (note that
they are still used by GDBserver).
This patch add a new maint command "maint check xml-descriptions" to test
dynamically generated tdesc are identical to these generated from xml files.
gdb:
2017-07-26 Yao Qi <yao.qi@linaro.org>
* cli/cli-cmds.c (maintenancechecklist): New variable.
* gdbcmd.h (maintenancechecklist): Declare it.
* i386-linux-tdep.c (_initialize_i386_linux_tdep) [GDB_SELF_TEST]:
Call i386_linux_read_description with different masks.
* maint.c (maintenance_check_command): New function.
(_initialize_maint_cmds): Call add_prefix_cmd.
* target-descriptions.c (tdesc_reg): override operator != and ==.
(tdesc_type): Likewise.
(tdesc_feature): Likewise.
(target_desc): Likewise.
[GDB_SELF_TEST] (selftests::record_xml_tdesc): New function.
(maintenance_check_xml_descriptions): New function.
(_initialize_target_descriptions) Add command "xml-descriptions".
* target-descriptions.h (selftests::record_xml_tdesc): Declare.
gdb/testsuite:
2017-07-26 Yao Qi <yao.qi@linaro.org>
* gdb.gdb/unittest.exp: Invoke command
"maintenance check xml-descriptions".
gdb/doc:
2017-07-26 Yao Qi <yao.qi@linaro.org>
* gdb.texinfo (Maintenance Commands): Document command
"maint check xml-descriptions".
PR 21555 is caused by the exception during the prologue analysis when re-set
a breakpoint.
(gdb) bt
#0 memory_error_message (err=TARGET_XFER_E_IO, gdbarch=0x153db50, memaddr=93824992233232) at ../../binutils-gdb/gdb/corefile.c:192
#1 0x00000000005718ed in memory_error (err=TARGET_XFER_E_IO, memaddr=memaddr@entry=93824992233232) at ../../binutils-gdb/gdb/corefile.c:220
#2 0x00000000005719d6 in read_memory_object (object=object@entry=TARGET_OBJECT_CODE_MEMORY, memaddr=93824992233232, memaddr@entry=1, myaddr=myaddr@entry=0x7fffffffd0a0 "P\333S\001", len=len@entry=1) at ../../binutils-gdb/gdb/corefile.c:259
#3 0x0000000000571c6e in read_code (len=1, myaddr=0x7fffffffd0a0 "P\333S\001", memaddr=<optimized out>) at ../../binutils-gdb/gdb/corefile.c:287
#4 read_code_unsigned_integer (memaddr=memaddr@entry=93824992233232, len=len@entry=1, byte_order=byte_order@entry=BFD_ENDIAN_LITTLE) at ../../binutils-gdb/gdb/corefile.c:362
#5 0x000000000041d4a0 in amd64_analyze_prologue (gdbarch=gdbarch@entry=0x153db50, pc=pc@entry=93824992233232, current_pc=current_pc@entry=18446744073709551615, cache=cache@entry=0x7fffffffd1e0) at ../../binutils-gdb/gdb/amd64-tdep.c:2310
#6 0x000000000041e404 in amd64_skip_prologue (gdbarch=0x153db50, start_pc=93824992233232) at ../../binutils-gdb/gdb/amd64-tdep.c:2459
#7 0x000000000067bfb0 in skip_prologue_sal (sal=sal@entry=0x7fffffffd4e0) at ../../binutils-gdb/gdb/symtab.c:3628
#8 0x000000000067c4d8 in find_function_start_sal (sym=sym@entry=0x1549960, funfirstline=1) at ../../binutils-gdb/gdb/symtab.c:3501
#9 0x000000000060999d in symbol_to_sal (result=result@entry=0x7fffffffd5f0, funfirstline=<optimized out>, sym=sym@entry=0x1549960) at ../../binutils-gdb/gdb/linespec.c:3860
....
#16 0x000000000054b733 in location_to_sals (b=b@entry=0x15792d0, location=0x157c230, search_pspace=search_pspace@entry=0x1148120, found=found@entry=0x7fffffffdc64) at ../../binutils-gdb/gdb/breakpoint.c:14211
#17 0x000000000054c1f5 in breakpoint_re_set_default (b=0x15792d0) at ../../binutils-gdb/gdb/breakpoint.c:14301
#18 0x00000000005412a9 in breakpoint_re_set_one (bint=bint@entry=0x15792d0) at ../../binutils-gdb/gdb/breakpoint.c:14412
This problem can be fixed by
- either each prologue analyzer doesn't throw exception,
- or catch the exception thrown from gdbarch_skip_prologue,
I choose the latter because the former needs to fix *every* prologue
analyzer to not throw exception.
This error can be reproduced by changing reread.exp. The test reread.exp
has already test that breakpoint can be reset correctly after the
executable is re-read. This patch extends this test by compiling test c
file with and without -fPIE.
(gdb) run ^M
The program being debugged has been started already.^M
Start it from the beginning? (y or n) y^M
x86_64/gdb/testsuite/outputs/gdb.base/reread/reread' has changed; re-reading symbols.
Error in re-setting breakpoint 1: Cannot access memory at address 0x555555554790^M
Error in re-setting breakpoint 2: Cannot access memory at address 0x555555554790^M
Starting program: /scratch/yao/gdb/build-git/x86_64/gdb/testsuite/outputs/gdb.base/reread/reread ^M
This is foo^M
[Inferior 1 (process 27720) exited normally]^M
(gdb) FAIL: gdb.base/reread.exp: opts= "-fPIE" "ldflags=-pie" : run to foo() second time (the program exited)
This patch doesn't re-indent the code, to keep the patch simple.
gdb:
2017-07-25 Yao Qi <yao.qi@linaro.org>
PR gdb/21555
* arch-utils.c (gdbarch_skip_prologue_noexcept): New function.
* arch-utils.h (gdbarch_skip_prologue_noexcept): Declare.
* infrun.c: Include arch-utils.h
(handle_step_into_function): Call gdbarch_skip_prologue_noexcept.
(handle_step_into_function_backward): Likewise.
* symtab.c (skip_prologue_sal): Likewise.
gdb/testsuite:
2017-07-25 Yao Qi <yao.qi@linaro.org>
PR gdb/21555
* gdb.base/reread.exp: Wrap the whole test with two kinds of
compilation flags, with -fPIE and without -fPIE.
s390-vregs.exp yields a Tcl error:
ERROR: can't read "i": no such variable
while executing
"expr $a_high * ($i + 1) * $a_high "
(procedure "hex128" line 2)
invoked from within
"hex128 $a_high $a_low $b_high $b_low"
...
This is a regression, caused by commit 30a254669b16b8 -- "Don't always
zero pad in print_*_chars". That patch introduced a new procedure
"hex128" for formatting a 128-bit value as hex, but it accidentally moved
the calculation of the 128-bit value into that new procedure as well
instead of leaving it in the original context. This is fixed.
gdb/testsuite/ChangeLog:
* gdb.arch/s390-vregs.exp: Calculate parameters to hex128 in the
calling context.
(hex128): Drop erroneous calculation of parameters.
[I made some typo fixes but forgot to amend my commit before sending the patch,
hence this v2.]
I see the following failure on Ubuntu 16.04's gcc 5.4.0:
Running /home/emaisin/src/binutils-gdb/gdb/testsuite/gdb.python/py-unwind.exp ...
FAIL: gdb.python/py-unwind.exp: continue to breakpoint: break backtrace-broken
FAIL: gdb.python/py-unwind.exp: Backtrace restored by unwinder (pattern 1)
The problem is that the test expects a very particular stack layout.
When stack protection is enabled, it adds a canary value which looks
like an additional local variable. This makes the test complain about
a bad stack layout and fail.
The simple solution is to disable stack protection for that test using
-fno-stack-protector. I checked older compilers (gcc 4.4, clang 3.5)
and they support that flag, so I don't think it's necessary to probe for
whether the compiler supports it.
Maybe a better solution would be to change the test to make it cope with
different stack layouts (perhaps it could save addresses of stuff in
some global variables which GDB/the unwinder would read). I'll go with
the simple solution for now though.
gdb/testsuite/ChangeLog:
* gdb.python/py-unwind.exp: Disable stack protection when
building test file.
Profiling GDB with the rest of series applied, I saw calls to
language_def showing up high in some runs. The problem is that
language_def is O(N) currently, since walk the languages vector each
time to find the matching language_defn.
IMO, the add_language mechanism is pointless, because "enum language"
implies the core of GDB needs to know about all languages anyway. So
simply make the languages vector array be an array where each
element's index is the corresponding enum language enumerator. Note
that "local_language_defn" is gone along the way. It's just a copy of
"auto", so the new code simply maps one to the other. One fewer place
to update when we need to change the language vector...
Also, a while ago the output of "set language" was made out of order
as side effect of some other change. While I was at it, I made them
sorted again.
gdb/ChangeLog:
2017-07-20 Pedro Alves <palves@redhat.com>
* ada-lang.c (ada_language_defn): Make extern.
(_initialize_ada_language): Remove add_language call.
* c-lang.c (c_language_defn, cplus_language_defn)
(asm_language_defn, minimal_language_defn): Make extern.
(_initialize_c_language): Delete.
* completer.c (compare_cstrings): Delete, moved to utils.h.
* d-lang.c (d_language_defn): Make extern.
(_initialize_d_language): Remove add_language calls.
* defs.h (enum language): Add comment.
* f-lang.c (f_language_defn): Make extern.
(_initialize_f_language): Remove add_language call.
* go-lang.c (go_language_defn): Make extern.
(_initialize_go_language): Remove add_language call.
* language.c: Include <algorithm>.
(languages): Redefine as const array.
(languages_size, languages_allocsize, DEFAULT_ALLOCSIZE): Delete.
(set_language_command): Handle "local". Use for-range loop.
(set_language): Remove loop.
(language_enum): Rewrite.
(language_def, language_str): Remove loops.
(add_language): Delete.
(add_set_language_command): New, based on add_languages.
(skip_language_trampoline): Adjust.
(local_language_defn): Delete.
(language_gdbarch_post_init): Adjust.
(_initialize_language): Remove add_language calls. Call
add_set_language_command.
* language.h (add_language): Delete.
(auto_language_defn)
(unknown_language_defn, minimal_language_defn, ada_language_defn)
(asm_language_defn, c_language_defn, cplus_language_defn)
(d_language_defn, f_language_defn, go_language_defn)
(m2_language_defn, objc_language_defn, opencl_language_defn)
(pascal_language_defn, rust_language_defn): Declare.
* m2-lang.c (m2_language_defn): Make extern.
(_initialize_m2_language): Remove add_language call.
* objc-lang.c (objc_language_defn): Make extern.
(_initialize_objc_language): Remove add_language call.
* opencl-lang.c (opencl_language_defn): Make extern.
(_initialize_opencl_language): Remove add_language call.
* p-lang.c (pascal_language_defn): Make extern.
(_initialize_pascal_language): Delete.
* rust-lang.c (rust_language_defn): Make extern.
(_initialize_rust_language): Delete.
* utils.h (compare_cstrings): New static inline function.
gdb/testsuite/ChangeLog:
2017-07-20 Pedro Alves <palves@redhat.com>
* gdb.base/default.exp (set language): Adjust expected output.
This test is using "set language ada" expecting that to cause GDB to
do Ada symbol name matching. That won't work when GDB uses the
symbol's language to decide which symbol matching algorithm to use,
because the test's symbols are C symbols.
So generalize the test a bit to not rely on Ada name matching rules.
Confirmed that by undoing the original fix the test was written for,
the test still fails.
gdb/testsuite/ChangeLog:
2017-07-20 Pedro Alves <palves@redhat.com>
* gdb.base/dmsym.c (pck__foo__bar__minsym): Rename to ...
(test_minsym): ... this, and make static.
(get_pck__foo__bar__minsym): Rename to ...
(get_test_minsym): ... this.
* gdb.base/dmsym.exp (): Remove "set language ada" call. Adjust
symbol names and comments.
* gdb.base/dmsym_main.c (get_pck__foo__bar__minsym): Rename to ...
(get_test_minsym): ... this.
(pck__foo__bar__minsym__2): Rename to ...
(test_minsym): ... this.
(main): Adjust.
Continuing the theme of the explicit locations patch, this patch gets
rid of the need for quoting function names in linespec TAB completion.
To recap, when you have overloads in your program, and you want to set
a breakpoint in one of them:
void function(int); // set breakpoint here.
void function(long);
(gdb) b function(i[TAB]
<all the symbols in the program that start with "i" are uselessly shown...>
This patch gets rid of the need for quoting by switching the linespec
completer to use the custom completion word point mechanism added in
the previous explicit location patch (extending it as needed), to
correctly determine the right completion word point. In the case
above, we want the completer to figure out that it's completing a
function name that starts with "function(i", and it now does.
We also want the completer to know when it's potentially completing a
source file name, for:
(gdb) break source.[TAB] -> source.c:
(gdb) break source.c: # Type line number or function name now
And we want it to know to complete label names, which it doesn't today:
(gdb) break function:lab[TAB]
etc., etc.
So what we want is for completion to grok the input string as closely
to how the linespec parser groks it.
With that in mind, the solution suggests itself - make the linespec
completer use the same parsing code as normal linespec parsing.
That's what the patch does. The old completer is replaced by one that
reuses the actual linespec parser as much as possible. This (ideally)
eliminate differences between what completion understands and actually
setting breakpoints understands by design.
The completer now offers sensible completion candidates depending on
which component of the linespec is being completed, source filename,
function, line number, expression, and (a new addition), labels. For
example, when completing the function part, we now show the full name
of the method as completion candidates, instead of showing whatever
comes after what readline considered the word break character:
(gdb) break klass::method[TAB]
klass:method1(int)
klass:method2()
If input is past the function, then we now offer keyword condidates:
(gdb) b function(int) [TAB]
if task thread
If input is past a keyword, we offer expression completion, which is
different from linespec completion:
(gdb) b main if 1 + glo[TAB]
global
(e.g., completes on types, struct data fields, etc.)
As mentioned, this teaches the linespec completer about completing
label symbols too:
(gdb) b source.c:function:lab[TAB]
A nice convenience is that when completion uniquely matches a source
name, gdb adds the ":" automatically for you:
(gdb) b filenam[TAB]
(gdb) b filename.c: # ':' auto-added, cursor right after it.
It's the little details. :-)
I worked on this patch in parallel with writing the (big) testcase
added closer to the end of the series, which exercises many many
tricky cases around quoting and whitespace insertion placement. In
general, I think it now all Just Works.
gdb/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* completer.c (complete_source_filenames): New function.
(complete_address_and_linespec_locations): New function.
(location_completer): Use complete_address_and_linespec_locations.
(completion_tracker::build_completion_result): Honor the tracker's
request to suppress append.
* completer.h (completion_tracker::suppress_append_ws)
(completion_tracker::set_suppress_append_ws): New methods.
(completion_tracker::m_suppress_append_ws): New field.
(complete_source_filenames): New declaration.
* linespec.c (linespec_complete_what): New.
(struct ls_parser) <complete_what, completion_word,
completion_quote_char, completion_quote_end, completion_tracker>:
New fields.
(string_find_incomplete_keyword_at_end): New.
(linespec_lexer_lex_string): Record quote char. If in completion
mode, don't throw.
(linespec_lexer_consume_token): Advance the completion word point.
(linespec_lexer_peek_token): Save/restore completion info.
(save_stream_and_consume_token): New.
(set_completion_after_number): New.
(linespec_parse_basic): Set what to complete next depending on
token. Handle function and label completions specially.
(parse_linespec): Disable objc shortcut in completion mode. Set
what to complete next depending on token type. Skip keyword if in
completion mode.
(complete_linespec_component, linespec_complete): New.
* linespec.h (linespec_complete): Declare.
gdb/testsuite/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* gdb.base/completion.exp: Adjust expected output.
* gdb.linespec/ls-errs.exp: Don't send tab characters, now that
the completer works.
One of the most annoying (to me) things about GDB's completion is when
you have overloads in your program, and you want to set a breakpoint
in one of them:
void function(int); // set breakpoint here.
void function(long);
(gdb) b -f func[TAB]
(gdb) b -f function( # ok, gdb completed as much as possible.
(gdb) b -f function([TAB] # show me the overloads, please.
<_all_ symbols in the program are shown...>
E.g., when debugging GDB, that'd be:
(gdb) b -f function([TAB]
(anonymous namespace)::get_global()::global pt_insn_get_offset@plt scm_new_port_table_entry
asprintf pt_pkt_alloc_decoder scm_new_port_table_entry@plt
asprintf@plt pt_pkt_alloc_decoder@plt scm_out_of_range
bt_ctf_get_char_array pt_pkt_sync_forward scm_out_of_range@plt
bt_ctf_get_char_array@plt pt_pkt_sync_forward@plt scm_putc
bt_ctf_get_uint64 pwrite scm_putc@plt
bt_ctf_get_uint64@plt pwrite@plt scm_reverse_x
bt_ctf_iter_read_event PyErr_Restore scm_reverse_x@plt
bt_ctf_iter_read_event@plt PyErr_Restore@plt scm_set_port_filename_x
<snip...>
Now that's a load of completely useless completions.
The reason GDB offers those is that the completer relies on readline
figuring out the completion word point in the input line based on the
language's word break characters, which include "(". So readline
tells the completer to complete on "", the string that is after '('.
Likewise, if you type "function(i[TAB]" to try to complete to "int",
you're out of luck. GDB shows you all the symbols in the program that
start with "i"... This makes sense for the expression completer, as
what you'd want to type is e.g., a global variable, say:
(gdb) print function(i[TAB]
but, it makes no sense when specifying a function name for a
breakpoint location.
To get around that limitation, users need to quote the function name,
like:
(gdb) b -f 'function([TAB]
function(int) function(long)
(gdb) b 'function(i[TAB]
(gdb) b 'function(int)' # now completes correctly!
Note that the quoting is only necessary for completion. Creating the
breakpoint does not require the quoting:
(gdb) b -f function(int) [RET]
Breakpoint 1 at ....
This patch removes this limitation.
(
Actually, it's a necessary patch, though not sufficient. That'll
start working correctly by the end of the series. With this patch, if try it,
you'll see:
(gdb) b -f function(i[TAB]
(gdb) b -f function
i.e., gdb strips everything after the "(". That's caused by some code
in symtab.c that'll be eliminated further down the series. These
patches are all unfortunately interrelated, which is also the reason
new tests only appear much later in the series.
But let's ignore that reality for the remainder of the description.
)
So... this patch gets rid of the need for quoting.
It does that by adding a way for a completer to control the exact
completion word point that readline should start the completion
request for, instead of letting readline try to figure it out using
the current language's word break chars array, and often failing.
In the case above, we want the completer to figure out that it's
completing a function name that starts with "function(i". It now
does.
It took me a while to figure out a way to ask readline to "use this
exact word point", and for a while I feared that it'd be impossible
with current readline (and having to rely on master readline for core
functionality is something I'd like to avoid very much). Eventually,
after several different attempts, I came up with what is described in
the comment above gdb_custom_word_point_brkchars in the patch.
With this patch, the handle_brkchars phase of the explicit location
completer advances the expected word point as it parses the input line
left to right, until it figures out exactly what we're completing,
instead of expecting readline to break the string using the word break
characters, and then having the completer heuristically fix up a bad
decision by parsing the input string backwards. This allows correctly
knowning that we're completing a symbol name after -function, complete
functions without quoting, etc.
Later, we'll make use of this same mechanims to implement a proper
linespec completer that avoids need for quoting too.
gdb/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* ada-lang.c (ada_collect_symbol_completion_matches): Add
complete_symbol_mode parameter.
* cli/cli-cmds.c (complete_command): Get the completion result out
of the handle_brkchars tracker if used a custom word point.
* completer.c: Include "linespec.h".
(enum explicit_location_match_type) <MATCH_LINE>: New enumerator.
(advance_to_expression_complete_word_point): New.
(completion_tracker::completes_to_completion_word): New.
(complete_files_symbols): Pass down
complete_symbol_mode::EXPRESSION.
(explicit_options, probe_options): New.
(collect_explicit_location_matches): Complete on the
explictit_loc->foo instead of word. Use
linespec_complete_function. Handle MATCH_LINE. Handle offering
keyword and options completions.
(backup_text_ptr): Delete.
(skip_keyword): New.
(complete_explicit_location): Remove 'word' parameter. Add
language, quoted_arg_start and quoted_arg_end parameters.
Rewrite, parsing left to right.
(location_completer): Rewrite.
(location_completer_handle_brkchars): New function.
(symbol_completer): Pass down complete_symbol_mode::EXPRESSION.
(enum complete_line_internal_reason): Adjust comments.
(completion_tracker::discard_completions): New.
(completer_handle_brkchars_func_for_completer): Handle
location_completer.
(gdb_custom_word_point_brkchars)
(gdb_org_rl_basic_quote_characters): New.
(gdb_completion_word_break_characters_throw)
(completion_find_completion_word): Handle trackers that use a
custom word point.
(completion_tracker::advance_custom_word_point_by): New.
(completion_tracker::build_completion_result): Don't rely on
readline appending the quote char.
(gdb_rl_attempted_completion_function_throw): Handle trackers that
use a custom word point.
(gdb_rl_attempted_completion_function): Restore
rl_basic_quote_characters.
* completer.h (class completion_tracker): Extend intro comment.
(completion_tracker::set_quote_char)
(completion_tracker::quote_char)
(completion_tracker::set_use_custom_word_point)
(completion_tracker::use_custom_word_point)
(completion_tracker::custom_word_point)
(completion_tracker::set_custom_word_point)
(completion_tracker::advance_custom_word_point_by)
(completion_tracker::completes_to_completion_word)
(completion_tracker::discard_completions): New methods.
(completion_tracker::m_quote_char)
(completion_tracker::m_use_custom_word_point)
(completion_tracker::m_custom_word_point): New fields.
(advance_to_expression_complete_word_point): Declare.
* f-lang.c (f_collect_symbol_completion_matches): Add
complete_symbol_mode parameter.
* language.h (struct language_defn)
<la_collect_symbol_completion_matches>: Add complete_symbol_mode
parameter.
* linespec.c (linespec_keywords): Add NULL terminator. Make extern.
(linespec_complete_function): New function.
(linespec_lexer_lex_keyword): Adjust.
* linespec.h (linespec_keywords, linespec_complete_function): New
declarations.
* location.c (find_end_quote): New function.
(explicit_location_lex_one): Add explicit_completion_info
parameter. Save quoting info. Don't throw if being called for
completion. Don't handle Ada operators here.
(is_cp_operator, skip_op_false_positives, first_of)
(explicit_location_lex_one_function): New function.
(string_to_explicit_location): Replace 'dont_throw' parameter with
an explicit_completion_info pointer parameter. Handle it. Don't
use explicit_location_lex_one to lex function names. Use
explicit_location_lex_one_function instead.
* location.h (struct explicit_completion_info): New.
(string_to_explicit_location): Replace 'dont_throw' parameter with
an explicit_completion_info pointer parameter.
* symtab.c (default_collect_symbol_completion_matches_break_on):
Add complete_symbol_mode parameter. Handle LINESPEC mode.
(default_collect_symbol_completion_matches)
(collect_symbol_completion_matches): Add complete_symbol_mode
parameter.
(collect_symbol_completion_matches_type): Pass down
complete_symbol_mode::EXPRESSION.
(collect_file_symbol_completion_matches): Add complete_symbol_mode
parameter. Handle LINESPEC mode.
* symtab.h (complete_symbol_mode): New.
(default_collect_symbol_completion_matches_break_on)
(default_collect_symbol_completion_matches)
(collect_symbol_completion_matches)
(collect_file_symbol_completion_matches): Add complete_symbol_mode
parameter.
gdb/testsuite/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* gdb.linespec/ls-errs.exp (do_test): Adjust expected output.
Make sure all of the tests have unique names in
gdb.mi/mi-vla-fortran.exp.
gdb/testsuite/ChangeLog:
* gdb.mi/mi-vla-fortran.exp: Make test names unique.
PR rust/21764 notes that "sizeof" does not work correctly for all types
in Rust. The bug turns out to be an error in the conversion of the AST
to gdb expressions. This patch fixes the bug and also avoids generating
incorrect expressions in another case.
Tested on the buildbot. I'm checking this in.
2017-07-14 Tom Tromey <tom@tromey.com>
PR rust/21764:
* rust-exp.y (convert_ast_to_expression): Add "want_type"
parameter.
<UNOP_SIZEOF>: Split into separate case.
<UNOP_VAR_VALUE>: Handle want_type. Add error case.
2017-07-14 Tom Tromey <tom@tromey.com>
PR rust/21764:
* gdb.rust/simple.exp: Add tests.
PR rust/21763 points out that gdb.lookup_typename does not work properly
for (some) Rust types. I tracked this down to a missing case in
symbol_matches_domain.
Tested by the buildbot.
2017-07-14 Tom Tromey <tom@tromey.com>
PR rust/21763:
* symtab.c (symbol_matches_domain): Add language_rust to special
case.
* rust-exp.y (convert_ast_to_expression) <OP_VAR_VALUE>: Don't
treat LOC_TYPEDEF symbols as variables.
2017-07-14 Tom Tromey <tom@tromey.com>
* gdb.rust/simple.exp: Add regression test for PR rust/21763.
This is the same patch as posted at
<https://sourceware.org/ml/gdb-patches/2017-02/msg00644.html>, with
the test at
<https://sourceware.org/ml/gdb-patches/2017-02/msg00687.html> squashed
in.
This patch fixes:
-FAIL: gdb.base/completion.exp: tab complete break break.c:ma (timeout)
-FAIL: gdb.base/completion.exp: complete break break.c:ma
+PASS: gdb.base/completion.exp: tab complete break break.c:ma
+PASS: gdb.base/completion.exp: delete breakpoint for tab complete break break.c:ma
+PASS: gdb.base/completion.exp: complete break break.c:ma
When run with --target_board=dwarf4-gdb-index.
The issue here is that make_file_symbol_completion_list_1, used when
completing a symbol restricted to a given source file, uses
lookup_symtab to look up the symtab with the given name, and search
for matching symbols inside. This assumes that there's only one
symtab for the given source file. This is an incorrect assumption
with (for example) -fdebug-types-section, where we'll have an extra
extra symtab containing the types. lookup_symtab finds that symtab,
and inside that symtab there are no functions...
gdb/ChangeLog:
2017-07-14 Pedro Alves <palves@redhat.com>
* symtab.c (make_file_symbol_completion_list_1): Iterate over
symtabs matching all symtabs with SRCFILE as file name instead of
only considering the first hit, with lookup_symtab.
gdb/testsuite/ChangeLog:
2017-07-14 Pedro Alves <palves@redhat.com>
* gdb.linespec/base/one/thefile.cc (z1): New function.
* gdb.linespec/base/two/thefile.cc (z2): New function.
* gdb.linespec/linespec.exp: Add tests.
In the test gdb.mi/mi-vla-fortran.exp the parameters passed to
mi_create_breakpoint are passed in the wrong order. By good luck the
tests still passes, however the wrong test name is used. All fixed in
this commit.
A previous commit fixed most of these, but I missed this last one.
gdb/testsuite/ChangeLog:
* gdb.mi/mi-vla-fortran.exp: Correct even more parameter passing
to mi_create_breakpoint.
In the test gdb.mi/mi-vla-fortran.exp the parameters passed to
mi_create_breakpoint are passed in the wrong order. By good luck the
tests still passes, however the wrong test name is used. All fixed in
this commit.
gdb/testsuite/ChangeLog:
* gdb.mi/mi-vla-fortran.exp: Correct parameter passing to
mi_create_breakpoint.
Rationale behind the change instead of adding a `.init$' postfix being
that "initializer for symbol" is much more informative when inspecting D
runtime type information in gdb, which is the only place where you would
encounter references to this compiler-generated symbol.
gdb/testsuite/ChangeLog:
* gdb.dlang/demangle.exp: Update for demangling changes.
This Rust bug report:
https://github.com/rust-lang/rust/issues/41970
noted an error from gdb. What is happening here (for me, the original
report had a different error) is that a pieced DWARF expression is not
writing to every byte in the resulting value. GDB errors in this
case. However, it seems to me that it is always valid to write fewer
bytes; the issue comes from writing too many -- that is, the test is
reversed. The test was also checking the sub-object, but this also
seems incorrect, as it's expected for the expression to write the
entirety of the enclosing object. So, this patch reverses the test
and applies it to the outer type, not the subobject type.
Regtested on the buildbot.
gdb/ChangeLog
2017-07-09 Tom Tromey <tom@tromey.com>
* dwarf2loc.c (dwarf2_evaluate_loc_desc_full): Reverse size
check and apply to outer type.
gdb/testsuite/ChangeLog
2017-07-09 Tom Tromey <tom@tromey.com>
* gdb.dwarf2/shortpiece.exp: New file.
In some cases a compiler may produce a single object file (& thus single
DWO file) representing multiple source files. The most common example of
this is in whole program optimization (such as LLVM's LTO). Fission may
still be a beneficial feature to use here - to avoid the need to
read/link the debug info with system libraries and the like.
This change adds basic support for multiple CUs in a single DWO file to
support LLVM's output in this situation.
There is still outstanding work to design and implement a solution for
cross-CU references (usually using DW_FORM_ref_addr) in this scenario.
For now LLVM works around this by duplicating DIEs rather than making
cross-CU references in DWO files. This degrades debugger
behavior/quality especially for file-local entities.
2017-07-06 David Blaikie <dblaikie@gmail.com>
* dwarf2read.c (struct dwo_file): Use a htab of dwo_unit* (rather than
a singular dwo_unit*) to support multiple CUs in the same way that
multiple TUs are supported.
(create_cus_hash_table): Replace create_dwo_cu with a function for
parsing multiple CUs from a DWO file.
(open_and_init_dwo_file): Use create_cus_hash_table rather than
create_dwo_cu.
(lookup_dwo_cutu): Lookup CU in the hash table in the dwo_file with
htab_find, rather than comparing the signature to a singleton CU in
the dwo_file.
2017-07-06 David Blaikie <dblaikie@gmail.com>
* gdb.dwarf2/fission-multi-cu.S: Test containing multiple CUs in a DWO,
built from fissiont-multi-cu{1,2}.c.
* gdb.dwarf2/fission-multi-cu.exp: Test similar to fission-base.exp,
except putting 'main' and 'func' in separate CUs in the same DWO file.
* gdb.dwarf2/fission-multi-cu1.c: First CU for the multi-CU-single-DWO
test.
* gdb.dwarf2/fission-multi-cu2.c: Second CU in the multi-CU-single-DWO
test.
The gdb.python/py-unwind.exp test is crashing GDB / leaving core dumps
in the test dir, even though it all passes cleanly. The crash is not
visible in gdb.sum/gdb.log because it happens as side effect of the
"quit" command, while flushing the frame cache.
The problem is simply a typo in a 'for' loop's condition, introduced
by a recent change [4fa847d78edd ("Remove MAX_REGISTER_SIZE from
py-unwind.c")], resulting in infinite loop / double-free.
The new test exposes the crash, like:
Running src/gdb/testsuite/gdb.python/py-unwind.exp ...
ERROR: Process no longer exists
gdb/ChangeLog:
2017-07-06 Pedro Alves <palves@redhat.com>
* python/py-unwind.c (pyuw_dealloc_cache): Fix for loop condition.
gdb/testsuite/ChangeLog:
2017-07-06 Pedro Alves <palves@redhat.com>
* gdb.python/py-unwind.exp: Test flushregs.
My last commit fixed a regression that happened when using
inline/multi-line commands for Python/Compile/Guile, but introduced
another regression: it is now not possible to use aliases for the
commands mentioned above. The fix is to almost revert the change I've
made and go back to using the 'struct cmd_list_element *', but at the
same time make sure that we advance the 'cmd_name' variable past all
the whitespace characters after the command name. If, after skipping
the whitespace, we encounter a '\0', it means that the command is not
inline. Otherwise, it is.
This patch also expands the testcase in order to check for aliases and
for trailing whitespace after the command name.
gdb/ChangeLog:
2017-06-30 Sergio Durigan Junior <sergiodj@redhat.com>
Pedro Alves <palves@redhat.com>
PR cli/21688
* cli/cli-script.c (command_name_equals_not_inline): Remove function.
(process_next_line): New variable 'inline_cmd'.
Adjust 'if' clauses for "python", "compile" and "guile" to use
'command_name_equals' and check for '!inline_cmd'.
gdb/testsuite/ChangeLog:
2017-06-30 Sergio Durigan Junior <sergiodj@redhat.com>
PR cli/21688
* gdb.python/py-cmd.exp (test_python_inline_or_multiline): Add new
tests for alias commands and trailing whitespace.
This bug is a regression caused by the following commit:
604c4576fdcfc4e7c28f569b3748a1b6b4e0dbd4 is the first bad commit
commit 604c4576fdcfc4e7c28f569b3748a1b6b4e0dbd4
Author: Jerome Guitton <guitton@adacore.com>
Date: Tue Jan 10 15:15:53 2017 +0100
The problem happens because, on cli/cli-script.c:process_next_line,
GDB is not using the command line string to identify which command to
run, but it instead using the 'struct cmd_list_element *' that is
obtained by using the mentioned string. The problem with that is that
the 'struct cmd_list_element *' doesn't have any information on
whether the command issued by the user is a multi-line or inline one.
A multi-line command is a command that will necessarily be composed of
more than 1 line. For example:
(gdb) if 1
>python
>print ('hello')
>end
>end
As can be seen in the example above, the 'python' command actually
"opens" a new command line (represented by the change in the
indentation) that will then be used to enter Python code. OTOH, an
inline command is a command that is "self-contained" in a single line,
for example:
(gdb) if 1
>python print ('hello')
>end
This Python command is a one-liner, and therefore there is no other
Python code that can be entered for this same block. There is also no
change in the indentation.
So, the fix is somewhat simple: we have to revert the change and use
the full command line string passed to process_next_line in order to
identify whether we're dealing with a multi-line or an inline command.
This commit does just that. As can be seen, this regression also
affects other languages, like guile or the compile framework. To make
things clearer, I decided to create a new helper function responsible
for identifying a non-inline command.
Testcase is attached.
gdb/ChangeLog:
2017-06-30 Sergio Durigan Junior <sergiodj@redhat.com>
PR cli/21688
* cli/cli-script.c (command_name_equals_not_inline): New function.
(process_next_line): Adjust 'if' clauses for "python", "compile"
and "guile" to use command_name_equals_not_inline.
gdb/testsuite/ChangeLog:
2017-06-30 Sergio Durigan Junior <sergiodj@redhat.com>
PR cli/21688
* gdb.python/py-cmd.exp (test_python_inline_or_multiline): New
procedure. Call it.
This commit fixes a mismatch between what "print" command completer
thinks the command understands, and what the command actually
understands.
The explicit location options are understood by commands that take
(linespecs and) explicit locations as argument. I.e, breakpoint
commands, and "list". For example:
(gdb) b -source file.c -function my_func
So for those commands, it makes sense that the completer
completes:
"b -sour[TAB]" -> "b -source "
"b -functi[TAB]" -> "b -function "
etc.
However, completion for commands that take expressions (not
linespecs/locations) as arguments, such as the "print" command, also
completes the explicit location options, even though those switches
aren't really understood by these commands. Instead, "-foo" is
understood as an expression applying unary minus on a symbol named
"foo" (think "print -1"):
(gdb) p -func[TAB]
(gdb) p -function [RET]
No symbol "function" in current context.
The patch fixes this by having the expression_completer function
bypass the function that completes explicit locations.
New regression tests included.
gdb/ChangeLog:
2017-06-29 Pedro Alves <palves@redhat.com>
* completer.c (expression_completer): Call
linespec_location_completer instead of location_completer.
gdb/testsuite/ChangeLog:
2017-06-29 Pedro Alves <palves@redhat.com>
* gdb.base/printcmds.exp: Add tests.
Fix issue exposed by commit 3e29f34.
The basic issue is that section data referenced through an objfile
pointer can also be referenced via the program-space data pointer,
although via a separate mapping mechanism, which is set up by
update_section_map. Thus once section data attached to an objfile
pointer is released, the section map associated with the program-space
data pointer must be marked dirty to ensure that update_section_map is
called to prevent stale data being referenced. For the matter at hand
this marking is being done via a call to objfiles_changed.
Before commit 3e29f34 objfiles_changed could be called after all of
the objfile pointers were processed in reread_symbols since section
data references via the program-space data pointer would not occur in
the calls of read_symbols performed by reread_symbols.
With commit 3e29f34 MIPS target specific calls to find_pc_section were
added to the code for DWARF information processing, which is called
via read_symbols. Thus in reread_symbols the call to objfiles_changed
needs to be called before calling read_symbols, otherwise stale
section data can be referenced.
Thanks to Luis Machado for providing text for the main comment
associated with the change.
gdb/
2017-06-28 Doug Gilmore <Doug.Gilmore@imgtec.com>
PR gdb/21337
* symfile.c (reread_symbols): Call objfiles_changed just before
read_symbols.
gdb/testsuite/
2017-06-28 Doug Gilmore <Doug.Gilmore@imgtec.com>
PR gdb/21337
* gdb.base/reread-readsym.exp: New file.
* gdb.base/reread-readsym.c: New file.
This is the portion of gdb_test which performs the match against
the RE (regular expression) passed to it:
return [gdb_test_multiple $command $message {
-re "\[\r\n\]*($pattern)\[\r\n\]+$gdb_prompt $" {
if ![string match "" $message] then {
pass "$message"
}
}
In a test that I've been working on recently, I wanted to use
a backreference - that's the \1 in the the RE below:
gdb_test "info threads" \
{.*[\r\n]+\* +([0-9]+) +Thread[^\r\n]* do_something \(n=\1\) at.*}
Put into English, I wanted to make sure that the value of n passed to
do_something() is the same as the thread number shown in the "info
threads" Id column. (I've structured the test case so that this
*should* be the case.)
It didn't work though. It turned out that ($pattern) in the RE
noted above is capturing the attempted backreference. So, in this
case, the backreference does not refer to ([0-9]+) as intended, but
instead refers to ($pattern). This is wrong because it's not what I
intended, but is also wrong because, if allowed, it could only match a
string of infinite length.
This problem can be fixed by using parens for a "noncapturing
subpattern". The way that this is done, syntactically, is to use
(?:$pattern) instead of ($pattern).
My research shows that this feature has been present since tcl8.1 which
was released in 1999.
The current tcl version is 8.6 - at least that's what I have on my
machine. It appears to me that mingw uses some subversion of tcl8.4
which will also have this feature (since 8.4 > 8.1).
So it seems to me that any platform upon which we might wish to test
GDB will have a version of tcl which has this feature. That being the
case, my hope is that there won't be any objections to its use.
When I looked at the implementation of gdb_test, I wondered whether
the parens were needed at all. I've concluded that they are. In the
event that $pattern is an RE which uses alternation at the top level,
e.g. a|b, we need to make $pattern a subpattern (via parens) to limit
the extend of the alternation. I.e, we don't want the alternation to
extend to the other portions of the RE which gdb_test uses to match
potential blank lines at the beginning of the pattern or the gdb
prompt at the end.
gdb/testsuite/ChangeLog:
* gdb.exp (gdb_test): Using noncapturing parens for the $pattern
subpattern.
When I added the new lnia extended mnemonic for addpcis, I updated the
assembler/disassembler test cases, but overlooked the GDB test cases.
This patch fixes that oversight and associated test case failure.
* gdb.arch/powerpc-power9.exp: Update test case for new lnia
extended mnemonic.
* gdb.arch/powerpc-power9.s: Likewise.
Pedro Alves