[lttng-dev] [RFC PATCH CTF 1/3] Clarify that unlisted enum values are implementation-defined

[Mathieu Desnoyers]

----- On Apr 23, 2020, at 6:51 PM, Jeremie Galarneau jeremie.galarneau at efficios.com wrote:

> On Thu, 23 Apr 2020 at 16:52, Mathieu Desnoyers
> <mathieu.desnoyers at efficios.com> wrote:
>>
>> From: Geneviève Bastien <gbastien+lttng at versatic.net>
>>
>> Signed-off-by: Geneviève Bastien <gbastien+lttng at versatic.net>
>> Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers at efficios.com>
>> ---
>>  common-trace-format-specification.md | 3 +++
>>  1 file changed, 3 insertions(+)
>>
>> diff --git a/common-trace-format-specification.md
>> b/common-trace-format-specification.md
>> index fd49e59..f5fea51 100644
>> --- a/common-trace-format-specification.md
>> +++ b/common-trace-format-specification.md
>> @@ -464,6 +464,9 @@ enum {
>>  }
>>  ~~~
>>
>> +The mappings in the enumeration type do not have to be exhaustive.
>> +Unlisted values are implementation defined.
>> +
> 
> This is too vague to be useful knowing that the main rationale for this
> change is to allow enums to express some type of bitfield of flags
> in the tracer and readers [1].
> 
> What is the meaning of an unmapped value? This section should at
> least describe the correct interpretation of unmapped values as flags
> and when it is appropriate to do so.

Considering that this is a patchlevel update to CTF, I would not expect
that we introduce new features in the specification. Only clarifications
to parts of the specification that were unclear.

Specifying a new behavior related to unmapped values would fall IMO into the
realm of "new feature", and would belong to a CTF 1.9 or CTF 2.0. As we
all know, there is a CTF 2.0 in the making, but it does not solve the
immediate problem of LTTng 2.12 which produces those unmapped enum values
within traces identified as CTF 1.8.

Moreover, to add to the problem, Babeltrace 2 has a strict match on CTF 1.8
in the CTF source plugin, and won't accept a CTF 1.9 trace (unlike Babeltrace 1
which would emit a warning about possibly unsupported features, but would accept
a 1.9 CTF trace nevertheless).

So this is why I am proposing this minimal clarification to the CTF 1.8
specification: that unmapped enum values are implementation defined (rather
than saying nothing about them). Considering that tracers can generate this
kind of trace data anyway, it's really a consideration that should have been
explicitly expressed in the specification from the start and was an
involuntary omission.

So perhaps we need to state something more than just "implementation defined",
but it's unclear what without ending up adding features into a patchlevel
update.

Thanks,

Mathieu


> Thanks,
> Jérémie
> 
> [1] https://review.lttng.org/c/babeltrace/+/3045
> 
>>  ### 4.2 Compound types
>>
>>  Compound are aggregation of type declarations. Compound types include
>> --
>> 2.11.0
>>
> 
> 
> --
> Jérémie Galarneau
> EfficiOS Inc.
> http://www.efficios.com

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com