Re: [XEN PATCH v1 2/3] ocaml/libs: Fill build failure due to unused variable in ocaml macro

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 18/10/2024 9:50 am, Javi Merino wrote:
> On Thu, Oct 17, 2024 at 06:47:44PM +0100, Andrew Cooper wrote:
>> On 17/10/2024 5:20 pm, Javi Merino wrote:
>>> On Fedora 40, the build fails with:
>>>
>>>     gcc -m64 -DBUILD_ID -fno-strict-aliasing -std=gnu99 -Wall 
>>> -Wstrict-prototypes -Wno-unused-but-set-variable -Wno-unused-local-typedefs 
>>>   -Werror -O2 -fomit-frame-pointer 
>>> -D__XEN_INTERFACE_VERSION__=__XEN_LATEST_INTERFACE_VERSION__ -MMD -MP -MF 
>>> .domain_getinfo_stubs_v1.o.d -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE   
>>> -fPIC -I/usr/lib64/ocaml -I 
>>> /build/tools/ocaml/libs/xsd_glue/domain_getinfo_plugin_v1/../../../libs -I 
>>> /build/tools/ocaml/libs/xsd_glue/domain_getinfo_plugin_v1/../../../libs/xsd_glue
>>>  
>>> -I/build/tools/ocaml/libs/xsd_glue/domain_getinfo_plugin_v1/../../../../../tools/include
>>>  -D__XEN_TOOLS__ 
>>> -I/build/tools/ocaml/libs\/xsd_glue/domain_getinfo_plugin_v1/../../../../../tools/include
>>>   -c -o domain_getinfo_stubs_v1.o domain_getinfo_stubs_v1.c
>>>     In file included from domain_getinfo_stubs_v1.c:10:
>>>     domain_getinfo_stubs_v1.c: In function 'xsd_glue_failwith':
>>>     /usr/lib64/ocaml/caml/memory.h:275:29: error: unused variable 
>>> 'caml__frame' [-Werror=unused-variable]
>>>       275 |   struct caml__roots_block *caml__frame = *caml_local_roots_ptr
>>>           |                             ^~~~~~~~~~~
>>>     domain_getinfo_stubs_v1.c:48:9: note: in expansion of macro 'CAMLparam0'
>>>       48 |         CAMLparam0();
>>>          |         ^~~~~~~~~~
>>>     cc1: all warnings being treated as errors
>>>
>>> The CAMLparam0 macro is defined in /usr/lib64/ocaml/caml/memory.h:255 as:
>>>
>>>     #define CAMLparam0()                                                    
>>> \
>>>       struct caml__roots_block** caml_local_roots_ptr =                     
>>> \
>>>         (DO_CHECK_CAML_STATE ? Caml_check_caml_state() : (void)0,           
>>> \
>>>          &CAML_LOCAL_ROOTS);                                                
>>> \
>>>       struct caml__roots_block *caml__frame = *caml_local_roots_ptr
>>>
>>> We can't modify the macro.  Mark the xsd_glue_failwith() function with
>>> ignore "-Wunused-variable" to prevent gcc from failing the build due
>>> to the unused variable.
>>>
>>> Fixes: a6576011a4d2 ("ocaml/libs: Implement a dynamically-loaded plugin for 
>>> Xenctrl.domain_getinfo")
>>> Signed-off-by: Javi Merino <javi.merino@xxxxxxxxx>
>> That's horrible...
>>
>> I note the Ocaml manual even says:
>>
>> "Some C compilers give bogus warnings about unused variables
>> caml__dummy_xxx at each use of CAMLparam and CAMLlocal. You should
>> ignore them."  which a brave claim to make...
>>
>>
>> The problem with pragma gcc is that we support Clang too.
>>
>> Having had a play, this works too
>>
>> @@ -45,7 +45,9 @@ static struct custom_operations xsd_glue_xenctrl_ops = {
>>  static void Noreturn xsd_glue_failwith(
>>         xc_interface *xch, const char *func, unsigned int line)
>>  {
>> +#define caml__frame __attribute__((unused)) caml__frame
>>         CAMLparam0();
>> +#undef caml__frame
>>         CAMLlocal1(msg);
>>         const xc_error *error = xch ? xc_get_last_error(xch) : NULL;
>>         char *str = NULL;
>>
>> and is rather more selective.
>>
>>
>> However, looking through other bits of memory.h, there's this gem:
>>
>> #define CAMLnoreturn ((void) caml__frame)
>>
>> which has existed since db3745919 "suppression des warnings "unused
>> variable" de gcc" in 2004.
>>
>> So, I think this is a better fix:
>>
>> @@ -69,6 +69,7 @@ static void Noreturn xsd_glue_failwith(
>>         free(str);
>>  
>>         caml_raise_with_arg(*caml_named_value("xsg.error_v1"), msg);
>> +       CAMLnoreturn;
>>  }
>>  #define xsd_glue_failwith(xch) xsd_glue_failwith(xch, __func__, __LINE__)
>>  
> Right!  I actually tried put a (void)caml__frame to silence the
> warning, but I didn't think of looking for a macro that would do
> exactly that.

:)  It's Ocaml.  I had a hunch there'd be something.

> Do I have to resend the series with your patch, or can you commit it
> directly?

I'll just fold it on commit.  No point spending more time on this.

In the meantime, Ocaml upstream have already merged my docs PR about this.

~Andrew