Psycopg psycopg

Comments and changes to this ticket

• You flagged this item as spam.
  

  Trevor McKay December 5th, 2011 @ 03:52 PM

  Note, the test program assumes a local database with the dsn ""dbname=cumin user=cumin host=localhost" just to get a connection so a cursor can be created. Modify as necessary, the call to mogrify is the important part.

  • psycopg_leak.py 905 Bytes

• You flagged this item as spam.
  

  Trevor McKay December 9th, 2011 @ 08:59 PM

  Note,

  I found a patch for this already in version 2.3.2, but it appears not to be in the 2.4 tree (fyi).

• 

  Daniele Varrazzo December 11th, 2011 @ 02:53 AM

  • State changed from “new” to “resolved”

  Good catch, thank you very much. Fixed in my devel.

  What do you mean as "found a patch in 2.3.2"?

• You flagged this item as spam.
  

  Trevor McKay December 12th, 2011 @ 02:09 PM

  What do you mean as "found a patch in 2.3.2"

  Sorry for the confusion on this. It appears that at least some Fedora and/or Red Hat distributions of psycopg2 have applied the fix for this in the past, but maybe the fix wasn't pushed back to upstream (or maybe it was lost, another possibility?). I didn't realize this until after I posted the comment :)

  Still trying to sort out the history on our end, thank you for the fix.

  T.

• You flagged this item as spam.
  

  Trevor McKay December 12th, 2011 @ 03:26 PM

  Okay,

  Just to be clear for posterity. I was mistaken about any patch in 2.3.2, the error is there is well.

  The code in 2.3.2 is using "item = PyObject_GetItem(n, key)" to check for presence of the value in the temporary dictionary, and so it rightly contains "Py_DECREF(item)" in the case where the value is found to already be in the dictionary. What it is missing is another call, Py_DECREF(value).

  In newer versions, "item = PyObject_GetItem(n, key)" is replaced with "0 == PyDict_Contains(n, key)", so there is no "item". It turns out that the call "Py_DECREF(value)" goes in the code at the same place where "Py_DECREF(item)" used to be in older versions :)

  So, after looking at newer code and applying the fix, I looked at 2.3.2 and my brain saw what it was expecting to see instead of what was. Being late afternoon on Friday made it worse... haha

• 

  Daniele Varrazzo December 12th, 2011 @ 03:53 PM

  Oh, I see what you mean. I've checked if the error was introduced in the commit 99b3c723 (where the PyDict_Contains was introduced) but it's been instead in psycopg since ever... It was probably in psycopg 1 too.

  Thank you again for the catch, and it was just in time for the release of 2.4.3: great!