Re: [PATCH] Fix gdb.cp/no-dmgl-verbose.exp test

[Lancelot SIX via Gdb-patches <gdb-patches@sourceware.org>]

Hi,

Thanks for looking into it.

On Thu, Apr 28, 2022 at 06:28:31PM -0700, Carl Love via Gdb-patches wrote:
> 
> GDB maintainers:
> 
> The gdb.cp/no-dmgl-verbose.exp test does a gdb_test which is not
> written correctly.  The test fails on both PowerPC and Intel.  The
> following patch update the gdb_test statement to the correct format. 
> The test now works correctly on PowerPC and Intel.
> 
> Please let me know if the patch is acceptable for mainline.  Thanks.
> 
>                      Carl Love
> 
> 
> -----------------------------------------------------------------------
> Fix gdb.cp/no-dmgl-verbose.exp test
> 
> The test tries to check that setting a break point on f (std::string) fails
> since the function is not defined.  The test is not syntactically correct.
> The test fails on both PowerPC and Intel.
> 
> This patch fixes the test to set the breakpoint and verify that it fails
> since the function is not defined.
> 
> The test now runs correctly on Power 10 and Intel x86_64.
> ---
>  gdb/testsuite/gdb.cp/no-dmgl-verbose.exp | 5 +----
>  1 file changed, 1 insertion(+), 4 deletions(-)
> 
> diff --git a/gdb/testsuite/gdb.cp/no-dmgl-verbose.exp b/gdb/testsuite/gdb.cp/no-dmgl-verbose.exp
> index 14f11ddcf04..d4ec88f3b6d 100644
> --- a/gdb/testsuite/gdb.cp/no-dmgl-verbose.exp
> +++ b/gdb/testsuite/gdb.cp/no-dmgl-verbose.exp
> @@ -28,8 +28,5 @@ clean_restart ${testfile}.o
>  
>  gdb_test_no_output "set breakpoint pending off"
>  
> -gdb_breakpoint {'f(std::string)'}
> -
> -gdb_test {break 'f(std::basic_string<char, std::char_traits<char>, std::allocator<char> >)'} \
> -	 {Function ".*" not defined\.} \
> +gdb_test "break 'f(std::string)'" ".*Function.*not defined." \
                                                             ^

I guess here the last . should be escaped (\.) otherwise has a special
meaning for a regex.

>  	 "DMGL_VERBOSE-demangled f(std::string) is not defined"

Also looks like that the test was testing that we can break on
"f(std::string)", not "f($what_string_really_is)".  This is not GDB's
behaviour, as far as I can tell.  To me GDB does the opposite.  One can
break using the full name of std::string, not "std::string".

Also, on my system, std::string is in reality
"std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >",
not what is used in the test.

I can test such behaviour with something like:

    gdb_test "break 'f(std::string)'" ".*Function.*not defined\." \
        "f(std::string) is not defined"

    set realtype ""
    gdb_test_multiple "info types ^std::string$" "" {
        -re -wrap "typedef (\[^;\]*) std::string;" {
            set realtype $expect_out(1,string)
            pass $gdb_test_name
        }
    }

    if { $realtype != "" } {
        gdb_breakpoint "f($realtype)"
    }

That being said, it would be nice to be able to place a breakpoint using
"f(std::string)"…

Best,
Lancelot.