[PATCH] Refactor parse_publication_options
Hi,
While reading the code related to creating subscriptions and publications for logical replication I found a difference between passing options to parse_subscription_options and parse_publication_options.
To parse_publication_options function we pass all options separately, while in parse_subscription_options we use SubOpts.
IMO the usage of a struct instead of each option separately seems cleaner.
I attached a patch that creates a PubOpts struct and uses it in all parse_publication_options occurrences.
--
Miłosz Bieniek
On Mon, Aug 24, 2026 at 10:16 PM Miłosz Bieniek <milosz@bieniek.cc> wrote:
Hi,
While reading the code related to creating subscriptions and publications for logical replication I found a difference between passing options to parse_subscription_options and parse_publication_options.
To parse_publication_options function we pass all options separately, while in parse_subscription_options we use SubOpts.
IMO the usage of a struct instead of each option separately seems cleaner.
I attached a patch that creates a PubOpts struct and uses it in all parse_publication_options occurrences.
IIRC the introduction of SubOpt had become a necessity because there
was an excessive number of parameters to parse_subscription_options.
OTOH, publications have a lot fewer options than subscriptions, so
this patch to introduce PubOpts seems more a matter of code
consistency than necessity.
Anyway, +1 from me for attempting this, but I expect it could receive
pushback from the "if-it-aint-broke-dont-fix-it" people.
~~~
Meanwhile, here are some other review comments for patch 0001.
======
1.
+typedef struct PubOpts
+{
+ bool publish_given;
+ PublicationActions pubactions;
+ bool publish_via_partition_root_given;
+ bool publish_via_partition_root;
+ bool publish_generated_columns_given;
+ char publish_generated_columns;
+} PubOpts;
To make this more similar to SubOpts, then you should also remove
those `xxx_given` members and instead use a `specified_opts` bitmap
member and PUBOPT_xxx macros, exactly the same as SubOpts does.
~~~
parse_publication_options:
2.
+ /* Start out with cleared opts. */
+ memset(opts, 0, sizeof(PubOpts));
+
+ /* Set default values */
+ opts->publish_given = false;
+ opts->publish_via_partition_root_given = false;
+ opts->publish_generated_columns_given = false;
IMO it's unnecessary to assign these `xxx_given` members to false,
since you've only just done the memset 0.
======
Kind Regards,
Peter Smith.
Fujitsu Australia
Hi Peter, thanks for your review.
IIRC the introduction of SubOpt had become a necessity because there
was an excessive number of parameters to parse_subscription_options.
OTOH, publications have a lot fewer options than subscriptions, so
this patch to introduce PubOpts seems more a matter of code
consistency than necessity.Anyway, +1 from me for attempting this, but I expect it could receive
pushback from the "if-it-aint-broke-dont-fix-it" people.
I understand that some people might not want to change working
code, but IMO it's worth the change for the sake of consistency.
~~~
Meanwhile, here are some other review comments for patch 0001.
======
1. +typedef struct PubOpts +{ + bool publish_given; + PublicationActions pubactions; + bool publish_via_partition_root_given; + bool publish_via_partition_root; + bool publish_generated_columns_given; + char publish_generated_columns; +} PubOpts;To make this more similar to SubOpts, then you should also remove
those `xxx_given` members and instead use a `specified_opts` bitmap
member and PUBOPT_xxx macros, exactly the same as SubOpts does.~~~
parse_publication_options:
2. + /* Start out with cleared opts. */ + memset(opts, 0, sizeof(PubOpts)); + + /* Set default values */ + opts->publish_given = false; + opts->publish_via_partition_root_given = false; + opts->publish_generated_columns_given = false;IMO it's unnecessary to assign these `xxx_given` members to false,
since you've only just done the memset 0.
You're right. I made the v2 version more similar to SubOpts.
One thing I'm not so sure about is `IsSet` macro. I copied it from the 'subscriptionscmds.c',
but maybe it should be extracted to a header file?
Kind regards,
Miłosz Bieniek
Patch v2 review comments:
1.
Oid puboid;
bool nulls[Natts_pg_publication];
Datum values[Natts_pg_publication];
+ PubOpts opts;
HeapTuple tup;
- bool publish_given;
- PublicationActions pubactions;
- bool publish_via_partition_root_given;
- bool publish_via_partition_root;
- bool publish_generated_columns_given;
- char publish_generated_columns;
Nit. Perhaps the PubOpts should come *after* the HeapTuple, just so
the nulls/values/tup all stay together like they were before.
~
Otherwise, the patch v2 LGTM
On Tue, Aug 25, 2026 at 7:43 PM Miłosz Bieniek <milosz@bieniek.cc> wrote:
...
One thing I'm not so sure about is `IsSet` macro. I copied it from the 'subscriptionscmds.c',
but maybe it should be extracted to a header file?
I would not worry about that for now. Wait to see if this patch gets
any support first.
======
Kind Regards,
Peter Smith
Fujitsu Australia