Re: [PATCH 30/32] ident.c: fix LGTM warning on the possible abuse of the '=' operator
- From
Johannes Schindelin <johannes.schindelin@gmx.de>
- Date
- Nov 6, 2019, 11:32 UTC
- Message-ID
- <nycvar.QRO.7.76.6.1911061231420.46@tvgsbejvaqbjf.bet>
- In-Reply-To
- <xmqqbltpwxyd.fsf@gitster-ct.c.googlers.com>
Hi,
On Wed, 6 Nov 2019, Junio C Hamano wrote:
Show 43 quoted lines
> Elia Pinto <gitter.spiros@gmail.com> writes:
>
> Did I miss the first 29 patches (with what I see in this patch, I
> do not know if I want to see them immediately, though ;-))?
>
> > Fix the LGTM warning of the rule that finds uses of the assignment
> > operator = in places where the equality operator == would
> > make more sense.
>
> I know you did not mean that existing
>
> } else if ((email = query_user_email()) && email[0]) {
>
> better reads if it were written like so:
>
> } else if ((email == query_user_email()) && email[0]) {
>
> but that is the only way how that sentence can be read (at least to
> me) without looking at what the patch actually does.
>
> As "email" has already been assigned to at this point in the
> codeflow, I agree that, to an eye that does not (and is not willing
> to spend cycles to) understand what the code is doing, the latter do
> look more natural: "If the value of the variable is the same as the
> return value of the query_user_email() function, and ...". And if
> "email" were a simpler arithmetic type it would have been even more
> (iow, it is clear "email" is a character string from "&& email[0]",
> so it is unlikely that "email == que()" is what the user intended).
>
> So I am somewhat sympathetic to the "warnings" here, but not all
> that much, especially if squelching makes the codeflow harder to
> follow by introducing otherwise unnecessary nesting levels (like
> this patch did). I suspect that it might be possible to futher
> restructure the code in such a way that we do not have to do an
> assignment in a conditional without making the code deeply nested,
> and that may perhaps be worth doing.
>
> But the thing is, assignment in a cascading conditional is so useful
> in avoiding pointless nesting of the code (imagine a reverse patch
> of this one---which is easy to sell as cleaning-up and streamlining
> the code).
>
> So, I dunno.For what it's worth, my reaction was exactly the same: I understand how some developers might deem the assignment inside an `if ()` condition undesirable, in Git's context I do strongly prefer the current code over the version proposed in this patch.
Thanks, Johannes
Show 32 quoted lines
>
> > Signed-off-by: Elia Pinto <gitter.spiros@gmail.com>
> > ---
> > ident.c | 13 ++++++++-----
> > 1 file changed, 8 insertions(+), 5 deletions(-)
> >
> > diff --git a/ident.c b/ident.c
> > index e666ee4e59..07f2f03b0a 100644
> > --- a/ident.c
> > +++ b/ident.c
> > @@ -172,12 +172,15 @@ const char *ident_default_email(void)
> > strbuf_addstr(&git_default_email, email);
> > committer_ident_explicitly_given |= IDENT_MAIL_GIVEN;
> > author_ident_explicitly_given |= IDENT_MAIL_GIVEN;
> > - } else if ((email = query_user_email()) && email[0]) {
> > - strbuf_addstr(&git_default_email, email);
> > - free((char *)email);
> > - } else
> > - copy_email(xgetpwuid_self(&default_email_is_bogus),
> > + } else {
> > + email = query_user_email();
> > + if (email && email[0]) {
> > + strbuf_addstr(&git_default_email, email);
> > + free((char *)email);
> > + } else
> > + copy_email(xgetpwuid_self(&default_email_is_bogus),
> > &git_default_email, &default_email_is_bogus);
> > + }
> > strbuf_trim(&git_default_email);
> > }
> > return git_default_email.buf;
>