[PATCH][gdb/testsuite] Fix shlib compilation with target board unix/-pie/-fPIE

[Tom de Vries]

On 12/14/20 6:21 PM, Simon Marchi wrote:
> On 2020-12-13 11:41 a.m., Tom de Vries wrote:
>> Hi,
>>
>> When running test-case gdb.base/info-shared.exp with target board
>> unix/-pie/-fPIE, we run into:
>> ...
>> spawn -ignore SIGHUP gcc -fno-stack-protector \
>>   outputs/gdb.base/info-shared/info-shared-solib1.c.o \
>>   -fdiagnostics-color=never -fPIC -shared -Wl,-soname,info-shared-solib1.so \
>>   -lm -fPIE -pie -o outputs/gdb.base/info-shared/info-shared-solib1.so^M
>> ld: Scrt1.o: in function `_start':^M
>> start.S:104: undefined reference to `main'^M
>> collect2: error: ld returned 1 exit status^M
>> compiler exited with status 1
>> ...
>>
>> The intention of the -pie/-fPIE flags is to build and test PIE executables on
>> platforms where that is not the default.  However, the flags clash with the
>> flags required to build shared libraries.
>>
>> Fix this by filtering out PIE-related flags out of the multilib_flags settings
>> in compile_shared_lib.
>>
>> Tested on x86_64-linux.
>>
>> Any comments?
>>
>> Thanks,
>> - Tom
>>
>> [gdb/testsuite] Fix shlib compilation with target board unix/-pie/-fPIE
>>
>> ---
>>  gdb/testsuite/lib/gdb.exp | 39 ++++++++++++++++++++++++++++++++++++++-
>>  1 file changed, 38 insertions(+), 1 deletion(-)
>>
>> diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
>> index e35d236018..071b5afa99 100644
>> --- a/gdb/testsuite/lib/gdb.exp
>> +++ b/gdb/testsuite/lib/gdb.exp
>> @@ -4283,7 +4283,7 @@ proc gdb_compile_pthreads {source dest type options} {
>>
>>  # Build a shared library from SOURCES.
>>
>> -proc gdb_compile_shlib {sources dest options} {
>> +proc gdb_compile_shlib_1 {sources dest options} {
>>      set obj_options $options
>>
>>      set ada 0
>> @@ -4416,6 +4416,43 @@ proc gdb_compile_shlib {sources dest options} {
>>      return ""
>>  }
>>
>> +# Build a shared library from SOURCES.  Ignore target boards PIE-related
>> +# multilib_flags.
>> +
>> +proc gdb_compile_shlib {sources dest options} {
>> +    global board
>> +
>> +    # Save multilib_flags.
>> +    set board [target_info name]
>> +    set save_multilib_flag [board_info $board multilib_flags]
>> +
>> +    # Ignore PIE-related setting in multilib_flags.
>> +    set multilib_flag ""
>> +    foreach op $save_multilib_flag {
>> +	if { $op == "-pie" || $op == "-no-pie" \
>> +		 || $op == "-fPIE" || $op == "-fno-PIE"} {
>> +	} else {
>> +	    append multilib_flag " $op"
>> +	}
>> +    }
>> +    unset_board_info "multilib_flags"
>> +    set_board_info multilib_flags "$multilib_flag"
>> +    set code [catch {gdb_compile_shlib_1 $sources $dest $options} result]
>> +
>> +    # Restore multilib_flags.
>> +    unset_board_info "multilib_flags"
>> +    set_board_info multilib_flags $save_multilib_flag
>> +
>> +    if {$code == 1} {
>> +	global errorInfo errorCode
>> +	return -code error -errorinfo $errorInfo -errorcode $errorCode $result
>> +    } elseif {$code > 1} {
>> +	return -code $code $result
>> +    }
>> +
>> +    return $result
> 
> That sounds reasonable.  I wonder if the lines above could be simplified
> to:
> 
>   if {$code == 1} {
>       ...
>   }
> 
>   return -code $code $result
> 
> IOW, if it's fine to return an explicit -code even if $code is 0 / OK.
> 

Thanks for the review.

I copied this code from gdb_do_cache_wrap, so I left it as is for the
commit.

This sounds like a good idea for a more broad refactoring.

Thanks,
- Tom

> Otherwise, LGTM.
> 
> Simon
>