ERROR: invalid value for parameter "search_path"

[winix]

System Information:

• OS: Archlinux
• Python: 2.7.2
• python2-psycopg2 2.4.2
• Database: postgresql 9.1.1

I run this command to create trac environment:

sudo trac-admin test initenv test 'postgres://user:pass@localhost/trac?schema=test' 'svn' /home/subversion/repos/test

Initenv filed, and python traceback:

Traceback (most recent call last):
  File "/usr/lib/python2.7/site-packages/trac/admin/console.py", line 422, in do_initenv
    options=options)
  File "/usr/lib/python2.7/site-packages/trac/env.py", line 213, in __init__
    self.create(options)
  File "/usr/lib/python2.7/site-packages/trac/env.py", line 401, in create
    DatabaseManager(self).init_db()
  File "/usr/lib/python2.7/site-packages/trac/db/api.py", line 146, in init_db
    connector.init_db(**args)
  File "/usr/lib/python2.7/site-packages/trac/db/postgres_backend.py", line 98, in init_db
    params)
  File "/usr/lib/python2.7/site-packages/trac/db/postgres_backend.py", line 87, in get_connection
    params)
  File "/usr/lib/python2.7/site-packages/trac/db/postgres_backend.py", line 212, in __init__
    cnx.cursor().execute('SET search_path TO %s', (self.schema,))
DataError: invalid value for parameter "search_path": "test"
DETAIL:  schema "test" does not exist

With postgresql 9.1, if schema does not exist, 'SET search_path' will raise DataError, not ProgrammingError (/db/postgres_backend.py)

[Remy Blank]

Thanks for the report. I don't have a 9.1 server readily available (I should take the time to upgrade my 8.2), but I'll post a patch.

[Remy Blank]

Catch the right exception for PostgreSQL 9.1.

[Remy Blank]

Can you please test 10406-search-path-r10830.patch​?

To be honest, I don't really understand why we even catch this error. Currently, if SET search_path fails, we leave the search path unchanged, which means that we are going to access the public schema. I think it would be better to abort than to access the wrong schema.

Oh, now I see. It's only to allow a connection to be created for the initial CREATE SCHEMA. This could be done better. Let's see…

[Remy Blank]

Mmh… The EnvironmentStub has some weird initialization sequence that actually requires catching the "missing schema" error, so I would rather leave it as-is. So yes, please test the patch above.

[winix]

The patch works for me.

I think, before init_db call get_connection function, should remove schema value from params temporary, after get_connection, if schema exists, then create schema and set search_path.

[Remy Blank]

Thanks for testing. Patch applied in [10831].

About the weird initialization of the connector, we have never had a defect report, so I'd rather leave it as-is at this point. Maybe I'll refactor it on trunk.