Re: [PATCH v12 04/10] utf8: teach same_encoding() alternative UTF encoding names

[Eric Sunshine <sunshine@sunshineco.com>]

On Fri, Mar 16, 2018 at 1:50 PM, Junio C Hamano <gitster@pobox.com> wrote:
> Eric Sunshine <sunshine@sunshineco.com> writes:
>> However, I'm having a tough time imagining cases in which callers
>> would want same_encoding() to return true if both arguments are NULL,
>> but outright crash if only one is NULL (which is the behavior even
>> before this patch). In other words, same_encoding() takes advantage of
>> is_encoding_utf8() for its convenience, not for its NULL-handling.
>> Given that view, the two explicit is_encoding_utf8() calls in
>> same_encoding() seem redundant once the same_utf_encoding() call is
>> added.
>
> So... does that mean we'd want something like this, or do you have
> something else in mind?
>
>         int same_encoding(const char *src, const char *dst)
>         {
>                 static const char utf8[] = "UTF-8";
>
>                 if (!src)
>                         src = utf8;
>                 if (!dst)
>                         dst = utf8;
>                 if (same_utf_encoding(src, dst))
>                         return 1;
>                 return !strcasecmp(src, dst);
>         }

I am not proposing anything like that for this patch or patch series.
I'm merely asking why, after this patch, same_encoding() still
contains the (in my mind) now-unneeded conditional:

    if (is_encoding_utf8(src) && is_encoding_utf8(dst))
        return 1;

If I'm reading the current code correctly, same_encoding() will crash
when either (but not both) of its arguments is NULL and the non-NULL
argument is not a variation of "UTF-8". If both arguments are NULL,
then it won't crash, but I don't believe that it was intentional that
it should crash for some NULL cases but not others; rather, not
crashing when both are NULL is an _accidental_ side-effect of relying
upon is_encoding_utf8().

If I understood him correctly, Lars's justification for retaining the
conditional in question even after the patch is to maintain the
non-crashing behavior for both arguments being NULL, even though it
will continue to crash when only one is NULL. That justification
doesn't makes sense to me since I can't imagine clients relying on
accidental behavior of sometimes crashing, sometimes not, hence my
question to Lars.

As for your snippet above which ensures that 'src' and 'dst' are
non-NULL, that (or some variation) would be fine if we ever expect
callers of same_encoding() to pass NULL for either of those arguments,
but such a fix is outside the scope of this patch and even this patch
series (which does not require such a fix), IMHO.