Re: [PATCH 2/3] Allow offset changes to be considered harmless.

[Giuliano Procida <gprocida@google.com>]

Hi.

On Wed, 13 May 2020 at 12:46, Dodji Seketeli <dodji@seketeli.org> wrote:

> Giuliano Procida <gprocida@google.com> a écrit:
>
> > This changes lets the user flip offset changes from harmful to
> > harmless. This is a simple way of shrinking very verbose diffs where
> > most changes within structs are offset changes caused by much rarer
> > member addition, removal and size changes.
>
> Hmmh.
>
> I think this is probably too simple.
>
> Offset changes can really signal a problem on their own.  So I am not
> for categorizing them as harmless.
>
> I understand that the reporting of offset changes that are a consequence
> of an addition, removal or change of a data member can be verbose, but
> I'd rather keep them as is for now, rather than risking some false
> negative because we want to go the easy route.
>
> In other words, if we really want to filter out those /consequential/
> offset changes, then we need to properly detect them and flag them as
> being filtered.
>

The tests do show them being filtered out. Did you mean in some other way?

+        1 data member changes (2 filtered):
+          type of 'int S::a[4]' changed:
+            type name changed from 'int[4]' to 'int[8]'
+            array type size changed from 128 to 256
+            array type subrange 1 changed length from 4 to 8


> I understand that that is a bigger undertaking, but I think that would
> the right way to do it.
>
> So I'd rather not apply this patch, have an enhancement request filed
> for that task and discuss/work on it properly instead.
>

I think the wider discussion is mostly covered by the other email thread.

Regards,
Giuliano.


> Cheers,
>
> --
>                 Dodji
>
> --
> To unsubscribe from this group and stop receiving emails from it, send an
> email to kernel-team+unsubscribe@android.com.
>
>