hbisect.py: don't rely on __del__ to write the current state. This is yet another page of the "Thou shalt not do too much inside __del__ methods" book, in the "demandload and __del__ don't go well together" chapter. The bisect extension is broken in 0.9.1: $ hg bisect init $ hg bisect bad Fatal Python error: Interpreter not initialized (version mismatch?) Aborted (yes, I tripled checked my instalation to make sure the problem is not there) It's been broken since revision fe1689273f84 moved the import of the binascii module into a demandload. (In details: the first time that "hg bisect bad" (or good) is called, there are still no revisions saved in .hg/bisect/*, so bisect.__init__ doesn't call hg.bin on anything. So, when we reach __del__, the binascii module still hasn't been imported and we get that "nice" message above.)

#!/bin/sh

hg init
echo 123 > a
hg add a
hg commit -m "first" -d "1000000 0" a
mkdir sub
echo 321 > sub/b
hg add sub/b
hg commit -m "second" -d "1000000 0" sub/b
cat sub/b
hg co 0
cat sub/b 2>/dev/null || echo "sub/b not present"
ls sub 2>/dev/null || echo "sub not present"

true