Re: [PATCH 2/5] ctf-front-end: Fix size and name for underlying types

["Guillermo E. Martinez" <guillermo.e.martinez@oracle.com>]

On 29/11/22 5:53, Dodji Seketeli wrote:
> Hello Guillermo,
> 
> "Guillermo E. Martinez via Libabigail" <libabigail@sourceware.org> a
> écrit:
> 
>> It fixed an incorrect representation in size and name for underlying
>> enum and struct types when it has a bitfield members:
>>
>>      struct foo
>>      {
>>        unsigned bar : 2;
>>        unsigned baz : 1;
>>      };
>>
>>    <type-decl name='' is-anonymous='yes' size-in-bits='2' id='type-id-1'/>
>>
>> So, `name' empty property, no showing the right enumerator's type and size
>> in bits:
>>
>>    <type-decl name='unsigned char' size-in-bits='8' id='type-id-4'/>
> 
> Hmmh, I think here, the name should be "unsigned int", because "unsigned
> bar" implicitly means 'unsigned int bar".  And the size-in-bits should
> be set to 32.
> 

Oh. Sorry, that's right.

> I have thus changed the introductory paragraphs of this commit log to
> the following:
> 
>      This patch fixes an incorrect representation in size and name of the
>      underlying type of enums as well as underlying types of bitfield data
>      members types.
> 
>      For instance, consider this struct.
> 
>          struct foo
>          {
>            unsigned bar : 2;
>            unsigned baz : 1;
>          };
> 
>      The data members bar and baz have an underlying type that is "unsigned
>      int".  Yet, the CTF front-end represents the underlying type of these
>      data members as:
> 
>        <type-decl name='' is-anonymous='yes' size-in-bits='2' id='type-id-1'/>
> 
>      The name property is empty, and it should be "unsigned int".
> 
>      The size in bit is '2', but it should be the size of the underlying
>      "unsigned int", in bits, which is 32.
> 
>      In other words, the underlying type of bar and baz should be:
> 
>        <type-decl name='unsigned int' size-in-bits='32' id='type-id-4'/>
> 
>      Note that today, libabigail doesn't represent the bitfield properties
>      of the data member.  Those bitfield properties are properties of the
>      data member, not of their type.  This is a known "Not Yet Implemented"
>      feature request that has been filed upstream at
>      https://sourceware.org/bugzilla/show_bug.cgi?id=27334.
> 
>      Similarly, the underlying type of enums is not properly represented by
>      the CTF front-end.
> 
>      Fixed thus.
> 
> It's a little bit more verbose, but hopefully, that should give a little
> bit more context for future code maintenance.
>

Totally agree. Thanks Dodji.

>>[...]

Kind regards,
guillermo