[Yum-devel] speeding up searchNevra in rpmsack

[seth vidal]

Hey folks,
 I looked at ways of speeding up searchNevra in rpmsack and I started
testing them.

The shortest way I found was by creating a simple dict of

pkgtup == index_of_hdr_in_rpmdb

then we could simply look up the hdr by index and that would make
returning headers very efficient. That was written last night. It did,
in fact, result in a massive speed up for remove operations.

It also ended up knocking rpm over everytime.

If you want to see the cause try this:

import rpm

for num in range(1, 1000):
    ts = rpm.TransactionSet()
    print num
    mi = ts.dbMatch(0, 20)
    h = mi.next()
    ts.closeDB()
    del ts



The line:
 mi = ts.dbMatch(0, 20) is the retrieval by index of a header. In this
case it is retrieving the header found at index 20.

Give that code a whirl, as root, inside a python or ipython interactive
shell. See if it causes a rather unceremonious exit. It does for me on
FC5.

The only way I could find to not make that happen when working on
searchNevra is to open a ts and keep it open. Which, of course, breaks
ctrl-c handling.

I just wanted to let people know that I have the code to speed up
searchNevra - I just can't put it in b/c it makes the world break.

-sv