On Thu, Aug 25, 2011 at 5:12 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> While I'm looking at this ... the current implementation has got a
> number of very inconsistent behaviors with respect to when it will
> expand a variable reference within a psql meta-command argument.
> Observe:
>
> regression=# \set foo 'value of foo'
> regression=# \set bar 'value of bar'
> regression=# \echo :foo
> value of foo
> regression=# \echo :foo@bar
> value of foo @bar
>
> (there shouldn't be a space before the @, IMO --- there is because this
> gets treated as two separate arguments, which seems bizarre)
>
> regression=# \echo :foo:bar
> value of foo value of bar
>
> (again, why is this two arguments not one?)
>
> regression=# \echo :foo@:bar
> value of foo @:bar
>
> (why isn't :bar expanded here, when it is in the previous case?)
>
> regression=# \echo foo:foo@:bar
> foo:foo@:bar
>
> (and now neither one gets expanded)
>
> ISTM the general rule ought to be that we attempt to substitute for a
> colon-construct regardless of where it appears within an argument, as
> long as it's not within quotes.
>
> Thoughts?
My main thought is that I remember this code being pretty awful -
especially with respect to error handling - when I looked at it. A
lot of dubious behaviors were more or less compelled by the
impossibility of bailing out at an arbitrary point a la ereport(). At
least, barring a drastic refactoring of the code, which might not be a
bad idea either.
No objection if you want to clean some of it up, but you may find it's
a larger sinkhole than you anticipate.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company