MIME-Version: 1.0
References: 
 <CAN4CZFMmDZMH56O9vb_g7vHqAk8ryWFxBMV19C39PFghENg8kA@mail.gmail.com>
 <CAOYmi+k_et3yXpJ8op71-95j7OYg-kX5bWLgW9YTV_5G7f+O1A@mail.gmail.com>
 <CAOYmi+kivcSnazEJA=KWknd3azGYnU3mMq9SUvht5Zq74qNcYQ@mail.gmail.com>
 <CAN4CZFMKCB2OXPGW0R_hCSu4Gg==B7dBSrv6Mf-YuFcrUncADg@mail.gmail.com>
 <CAOYmi+kCYZ3YiOu+oSv1gVW6LXQaNg4BcEpskYizWgfV1z12kA@mail.gmail.com>
 <CAN4CZFPNL4xuNAwmD1-TUR-z+C84axb+cdhip26k3YBFYo9r1g@mail.gmail.com>
In-Reply-To: 
 <CAN4CZFPNL4xuNAwmD1-TUR-z+C84axb+cdhip26k3YBFYo9r1g@mail.gmail.com>
From: Jacob Champion <jacob.champion@enterprisedb.com>
Date: Wed, 1 Apr 2026 10:09:48 -0700
Message-ID: 
 <CAOYmi+mN2OeFotwLKwwSDAuXHJ6xmrn3AtkUoHRcfAEENSFbMA@mail.gmail.com>
Subject: Re: [oauth] Split and extend PGOAUTHDEBUG
To: Zsolt Parragi <zsolt.parragi@percona.com>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: 
 <https://www.postgresql.org/message-id/CAOYmi%2BmN2OeFotwLKwwSDAuXHJ6xmrn3AtkUoHRcfAEENSFbMA%40mail.gmail.com>
Precedence: bulk

On Wed, Apr 1, 2026 at 2:35=E2=80=AFAM Zsolt Parragi <zsolt.parragi@percona=
.com> wrote:
> +/* all safe and unsafe flags, for the legacy UNSAFE behavior */
> +#define OAUTHDEBUG_UNSAFE_ALL ((uint32) ~0)
>
> The name of this variable is a bit confusing, it's not only about
> unsafe settings. I understand why you added the unsafe to it, but
> before checking the value/comment I thought this will be the bitmask
> of all unsafe options.

I agree.

> On the other hand I don't have a better idea
> other than simply using ALL.

OAUTHDEBUG_LEGACY_UNSAFE?

> +oauth_get_debug_flags(void)
> +{
> + uint32 flags =3D 0;
> + const char *env =3D getenv("PGOAUTHDEBUG");
> ...
>
> One of the reasons why I implemented this in a C file is because I
> wanted to avoid reparsing and warning spam/duplication.

I think I'm missing something; how does the choice of .c/.h change
things? There's no static tracking in v1 of the patchset (nor should
there be without locking or atomics, which is not maintenance I really
want to introduce for a debug feature).

> Reparsing
> shouldn't be a major issue since this is a debug feature, but this
> approach causes a warning to print twice in a few corner cases.

Which new corner cases? v1 also prints duplicates (e.g. with
`UNSAFE:blah,http`). I didn't intend to introduce any new calls to
oauth_get_debug_flags() over those already done in v1/v2; if I did
that's a bug.

--Jacob