[DTrace-devel] [PATCH 04/20] Increase the instruction counter to 4 digits for BPF programs
Eugene Loh
eugene.loh at oracle.com
Wed Jun 2 12:19:21 PDT 2021
On 6/2/21 2:37 PM, Kris Van Hees wrote:
> On Wed, Jun 02, 2021 at 02:11:07PM -0400, Eugene Loh wrote:
>> Reviewed-by: Eugene Loh <eugene.loh at oracle.com>
>>
>> though:
>>
>> *) We should show offsets in the variable and relocation listings in
>> the same format as in the instruction listings. That is, if offsets in
>> the instruction listings look like 00024 and so on, then the offsets in
>> the variable and relocation listings should as well. Adding the leading
>> zeroes will also align the offsets in the variable and relocation listings.
> Why? They are different things. If anything, if we want to make such a
> change I would be more in favour of dropping the leading zeros in the
> disassembly listing. That would be a deviation from the legacy version, as
> far as I can see, but I'd be OK with that. That was not the purpose of this
> patch though. But if we agree, we can make changes like that.
>
>> *) And instruction numbers need to have the same format in branch/jump
>> targets -- that is, in dt_dis_branch* and dt_dis_jump.
> I can see why we might want that to be consistent, though, again, then I think
> I'd rather make a change to drop leading zeros. Aligning at the right is the
> way to go then as well though I think.
>
> Thoughts? Opinions?
For me, the safe decision is consistency. Use the same format for a
piece of information regardless of where in the output the info is
appearing.
Regarding leading zeros, I'm less sure. They're certainly a nuisance
sometimes, but they do provide visual reminders of what one is looking
at. Is it an instruction count or an offset? Well, 0024 and 00392
include a visual hint (the field width); 28... not so much. Jump
target? Does "-> 24" mean pc+24 or instruction 0024? These are not
rock-clad arguments. I can well imagine people disagreeing on these
subjective questions. And such "visual hints" are not necessary,
especially in my jump-target example. But, again, for me, the leading
zeros -- and different widths for instruction counts and offsets -- have
advantages.
That's my $0.02.
>> On 6/2/21 1:47 AM, Kris Van Hees wrote:
>>> BPF programs are becoming longer as more features are implemented, and
>>> we are now routinely encountering progrmas that are 1000 or more
>>> instructions. The disassembler listings were limited to 3 digits for
>>> the instruction counter and 4 for the offset. This is now increased
>>> to 4 digits for the counter and 5 for the offset.
More information about the DTrace-devel
mailing list