[Metakit] more commit-aside problems

[zooko at zooko.com]

------- Forwarded Message

From: Jim McCoy <mccoy at allmydata.com>
Date: May 4, 2006 12:06:49 PM PDT
To: metakit at equi4.com
Subject: Update on the commit-aside crash

After further poking around we discovered another problem using  
commit-aside in Python.  If you have a view that has a binary  B  
property then some part of the full commit does not seem to work.   
If the view has no B properties then we can do a full commit and  
nuke the aside file with no consequences, but if there is a B  
property we can reliably crash the python process due to something  
in Mk4py.dll exploding.  The following code snippet should recreate  
the problem.

import metakit, random, os

db = metakit.storage("_f.mk", 1)
dba = metakit.storage("_f.mka", 1)
db.aside(dba)
vw = db.getas("a[i:I,b:B]")
vw.append(i=1,b=random.getrandbits(32))
vw.append(i=2, b=random.getrandbits(32))
vw.append(i=3, b=random.getrandbits(32))
dba.commit()
db.commit(1)

del vw
del db
del dba

os.remove("_f.mka")

db = metakit.storage("_f.mk", 1)
dba = metakit.storage("_f.mka", 1)
db.aside(dba)
vw = db.getas("a")
assert len(vw) == 3
# and now to crash things
assert vw[0].i == 1



If none of the view properties are binary then everything will work  
as expected (modulo the "open the views before trying a full  
commit" bit Zooko noted) but with a binary property in the mix then  
as soon as we try to use the re-opened views after doing a full  
commit and nuking the aside file everything comes crashing down.

Regards,

Jim McCoy