Rebuilding indexes should be done at the en of update iterators

[faltet]

Right now, when updating indexed columns in situations like:

for row in table:
    row['mycol1'] = XXX
    row['mycol2'] = YYY
    row.update()
table.flush()

the existing indexes are re-computed when the output buffer is full. This is too expensive, and should be enough to just re-index when the update iterator has finished.

[faltet]

Fixed in r3506.

[faltet]

Unfortunately, this exposes a bug in the next test:

tables/tests/test_indexes.py ManyNodesTestCase.test00

Until this is is fixed, I'm disabling this optimization in r3525.

Note that ticket #160 only has sense with this optimization on.

[faltet]

Re-applied r3506 in r3859 (pro-trunk) and r3860 (std-trunk) and it works now. I don't know why it works now, but this is probably due to the recent improvements in node management.

Closing this ticket.

[faltet]

I'm having problems again with the test_indexes.py:ManyNodesTestCase. They appear only on Windows :-/ This should be further investigated.

[faltet]

Fixed in r3886. The problem was that calling a Table.reIndexDirty() during a kill phase was loading more nodes (related with indexes), and that's always dangerous. Now, the call is only made when manual calls are made (else, a warning is issued).