Обсуждение: Avoid unused value (src/fe_utils/print.c)

Hello Ranier,

03.06.2023 13:14, Ranier Vilela wrote:
> Hi,
>
> This is for Postgres 17 (head).
>
> Per Coverity.
> At function print_unaligned_text, variable "need_recordsep", is
> unnecessarily set to true and false.

Clang' scan-build detects 58 errors "Dead assignment", including that one.
Maybe it would be more sensible to eliminate all errors of this class?

Hi Michael,

04.06.2023 01:42, Michael Paquier wrote:
> On Sat, Jun 03, 2023 at 03:00:01PM +0300, Alexander Lakhin wrote:
>> Clang' scan-build detects 58 errors "Dead assignment", including that one.
>> Maybe it would be more sensible to eliminate all errors of this class?
> Depends on if this makes any code changed a bit easier to understand I
> guess, so that would be a case-by-case analysis.  Saying that, the
> proposed patch seems right while it makes slightly easier to
> understand the footer print part.

It also aligns the code with print_unaligned_vertical(), but I can't see why
need_recordsep = true;
is a no-op here (scan-build dislikes only need_recordsep = false;).
I suspect that removing that line will change the behaviour in cases when
need_recordsep = false after the loop "print cells" and the loop
"for (footers)" is executed.

Best regards,
Alexander

Hi,

Alexander wrote:

> It also aligns the code with print_unaligned_vertical(), but I can't see why
> need_recordsep = true;
> is a no-op here (scan-build dislikes only need_recordsep = false;).
> I suspect that removing that line will change the behaviour in cases when
> need_recordsep = false after the loop "print cells" and the loop
> "for (footers)" is executed.

As I understand cont->cells is supoused to have all cont->ncolumns * cont->nrows
entries filled so the loop "print cells" always assigns need_recordsep = true,
except when there are no cells at all or cancel_pressed == true.
If cancel_pressed == true then footers are not printed. So to have
need_recordsep == false before the loop "for (footers)" table should be empty,
and need_recordsep should be false before the loop "print cells". It can only
be false there when cont->opt->start_table == true and opt_tuples_only == true
so that headers are not printed. But when opt_tuples_only == true footers are
not printed either.

So technically removing "need_recordsep = true;" won't change the outcome.

But
it's not obvious, so I also have doubts about removing this line. If someday
print options are changed, for example to support printing footers and not
printing headers, or anything else changes in this function, the output might
be unexpected with this line removed.

Hello Karina,

30.06.2023 17:25, Karina Litskevich wrote:
> Hi,
>
> Alexander wrote:
>
>> It also aligns the code with print_unaligned_vertical(), but I can't see why
>> need_recordsep = true;
>> is a no-op here (scan-build dislikes only need_recordsep = false;).
>> I suspect that removing that line will change the behaviour in cases when
>> need_recordsep = false after the loop "print cells" and the loop
>> "for (footers)" is executed.
> As I understand cont->cells is supoused to have all cont->ncolumns * cont->nrows
> entries filled so the loop "print cells" always assigns need_recordsep = true,
> except when there are no cells at all or cancel_pressed == true.
> If cancel_pressed == true then footers are not printed. So to have
> need_recordsep == false before the loop "for (footers)" table should be empty,
> and need_recordsep should be false before the loop "print cells". It can only
> be false there when cont->opt->start_table == true and opt_tuples_only == true
> so that headers are not printed. But when opt_tuples_only == true footers are
> not printed either.
>
> So technically removing "need_recordsep = true;" won't change the outcome. But
> it's not obvious, so I also have doubts about removing this line. If someday
> print options are changed, for example to support printing footers and not
> printing headers, or anything else changes in this function, the output might
> be unexpected with this line removed.

I think that the question that should be answered before moving forward
here is: what this discussion is expected to result in?

If the goal is to avoid an unused value to make Coverity/clang`s scan-build
a little happier, then maybe just don't touch other line, that is not
recognized as dead (at least by scan-build;

I wonder what Coverity says
about that line).

Otherwise, if the goal is to do review the code and make it cleaner, then
why not get rid of "if (need_recordsep)" there?

 

But
it's not obvious, so I also have doubts about removing this line. If someday
print options are changed, for example to support printing footers and not
printing headers, or anything else changes in this function, the output might
be unexpected with this line removed.

 

That part I didn't understand.
How are we going to make this function less readable by removing the complicating part.

On Thu, Jul 6, 2023 at 5:37 PM Karina Litskevich
<litskevichkarina@gmail.com> wrote:
> My point is, technically right now you won't see any difference in output
> if you remove the line. Because if we get to that line the need_recordsep
> is already true. However, understanding why it is true is complicated. That's
> why if you remove the line people who read the code will wonder why we don't
> need a separator after "fputs"ing a footer. So keeping that line will make
> the code more readable.
> Moreover, removing the line will possibly complicate the future maintenance.
> As I wrote in the part you just quoted, if the function changes in the way
> that need_recordsep is not true right before printing footers any more, then
> output will be unexpected.

I agree with Karina here.  Either this patch should keep the
"need_recordsep = true;" line, thus removing the no-op assignment to
false and making the code slightly less unreadable; or the entire
function should be refactored for readability.

 As there is consensus to keep the no-op assignment,

I will go ahead and reject the patch.

On Wed, Jul 12, 2023 at 1:46 AM Ranier Vilela <ranier.vf@gmail.com> wrote:

 As there is consensus to keep the no-op assignment,

I will go ahead and reject the patch.

In your patch you suggest removing two assignments, and we only have
consensus to keep one of them. The other one is an obvious no-op.

I attached a patch that removes only one assignment. Could you please try
it and check whether Coverity is still complaining about need_recordsep

variable?

Checked today, Coverity does not complain about need_recordsep.