Re: [EXTERNAL] [PATCH v3 4/6] trace: support dumping binary inside a struct

[David Marchand <david.marchand@redhat.com>]

On Tue, Feb 11, 2025 at 9:53 AM Sunil Kumar Kori <skori@marvell.com> wrote:
>
> > diff --git a/lib/eal/common/eal_common_trace_ctf.c
> > b/lib/eal/common/eal_common_trace_ctf.c
> > index 6bc8bb9036..d9b307e076 100644
> > --- a/lib/eal/common/eal_common_trace_ctf.c
> > +++ b/lib/eal/common/eal_common_trace_ctf.c
> > @@ -378,6 +378,9 @@ char *trace_metadata_fixup_field(const char *field)
> >               "->",
> >               "*",
> >               " ",
> > +             "&",
> > +             "(",
> > +             ")",
> Adding brackets makes token names a bit complex. Same name is used in metadata
> file to dump the traces to the user. With this complex name, user might not
> understand the purpose of that information.
>
> For example, _conf_src_port_pcie_sizeof_uint64_t_ is created in metafile and same will be dumped. But with this
> User might not get that which information is provided.

In practice, as there is no other documentation for a trace point
arguments, a user needs to read the trace point definitions.
So it seems trivial to me to link a variable name in the trace point
emitter, and the metadata in the trace files.

>
> This is the reason; we followed the existing naming convention which is user friendly.

User friendly? I don't see how this is different with '.' and '->'.
What is missing is documentation, as nothing describes this
sanitisation in the first place (even before this series).


-- 
David Marchand