Re: Bug #899: pg_dumpall produce erroneous files if the databbase

Hi Tom,

First of all, I appologies for my english : it's not
my native language so, it's not so easy for me :-(

> > ... yes but it is not the probleme : as no user is
> > attached w/ procedural languages (it's clearly
> > commented in the source code of pg_dump), it
> should
> > generate ACL of then in a scope of a normal user,
> and
> > not as super user. Again, I solve this issues in
> my
> > patches.
>
> I don't like the parts of this that are concerned
> with rearranging dump
> order; it's just adding complexity that won't do
> much except confuse
> people working on the code.

I try to make it as simple as possible, and my goal
was to avoid huge changes in the original source code.
And the easiest, and safer, way was to change dump
order.
With my modifications, pg_dump create data in the same
order as if you have to create a new database by hand,
nothing more :

1/ database creation
2/ schema creation
3/ procedural languages declaration
4/ PL ACL

Step 1 & 2 was already done in the original code. My
changes only add point 3 and 4

/* Now sort the output nicely */
SortTocByOID(g_fout);
+ MoveToStart(g_fout, "ACL LANGUAGE");
+ MoveToStart(g_fout, "PROCEDURAL LANGUAGE");
+ MoveToStart(g_fout, "FUNC PROCEDURAL LANGUAGE");
MoveToStart(g_fout, "SCHEMA");
MoveToStart(g_fout, "DATABASE");
MoveToEnd(g_fout, "TABLE DATA");

> The one and only true
> solution for this
> class of problems is to make pg_dump pay attention
> to pg_depend
> dependencies.

Sure, but I was facing a trouble that made my backup
unusable, because psql won't accept this file as it
is. It's not acceptable for any live application
where, if something going wrong, you must be fast as
possible : it's impossible if you have many manuals
operations to do.
So perhasp my solution is not the better one, but it
solve this issue w/o risks (because, as I said,
pg_dump follows the same procedure as if you do it by
hand) and it's the only one for the moment ;-D

> Now that the raw data is available, I
> think we're past
> the point where we should keep adding heuristic
> kluges that try to
> substitute for dependency analysis. I fear they're
> as likely to break
> corner cases as fix 'em :-(

2 solutions :
1/ ASCII dump file is now obsolete, and remove this
capability for pg_dump,
2/ ASCII dump file is still usefull and make it solide
as a rock.

If you choose the 1st solution, please remember that
is very important for companies to have something
"open", specialy w/ free software. My manager asks me
to choose a database that doesn't use private backup
format : WE HAVE TO TRUST in backup file ; even if we
have to change our database engine, data must be
EASILY importable in another database engine. Some
works has to be done, but it's quite easy w/ ASCII
files comparst to a binary format.

> As for the ACL changes, I don't see the point of 'em
> any more than Bruce
> does. Explain again please?

pg_dump.c file, line 4860 : 'usename' is the owner,
NULL if there is no owner (for languages).

-> No SET SESSION AUTHORIZATION before doing this
task, so they are done by the current active user in
the dump file. In the orginal pg_dump, it could be
done ANYWHERE in the file, and, in my case "anywhere"
was juste after some operations made by a normal user.
So as only the super user can grant language, it fails
:-(

So, I put languages ACL juste after language creation
to avoid such problem.

Have a nice W.E.

Laurent

=====
The misspelling master is on the Web.
_________ 100 % Dictionnary Free !
/ /(
/ Dico / / Pleins d'autres fautes sur
/________/ / http://go.to/destroyedlolo
(#######( / http://destroyedlolo.homeunix.org:8080
Quoi, des fautes d'orthographe! Pas possible ;-D.

___________________________________________________________
Do You Yahoo!? -- Une adresse @yahoo.fr gratuite et en français !
Yahoo! Mail : http://fr.mail.yahoo.com