Re: Information on savepoint requirement within transctions

On 30.01.2018 03:07, David G. Johnston wrote:
> ​So, my first pass at this.

Nice, thank you.

> + These are of particular use for client software to use when executing
> + user-supplied SQL statements and want to provide try/catch behavior
> + where failures are ignored.

Personally, I'd reword this to something like this:

> These are of particular use for client software which is executing
> user-supplied SQL statements and wants to provide try/catch behavior
> with the ability to continue to use the transaction after a failure.

Or maybe something like this:

> These are of particular use for client software which requires
> fine-grained support over failure behavior within a transaction.
> They allow to provide a try/catch behavior with the ability
> to continue to use a transaction after a failure.

Also I'd like to see something like this in the docs at roughly the same position:

> If a failure occurs during a transaction, the transaction enters
> an aborted state. An aborted or failed transaction cannot be used
> anymore to issue more commands, ROLLBACK or ROLLBACK TO must be used
> to regain control of the aborted transaction. A commit issued while
> the transaction is aborted is automatically converted into a
> <xref linkend="sql-rollback"/>.

I'm not sure about the terminology here, though, because the Transaction
Tutorial (https://www.postgresql.org/docs/9.6/static/tutorial-transactions.html)
speaks of "aborted" transactions, while you use the term "failed" here.