Deal with failed clone/transaction interaction -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Deal with failed clone/transaction interaction > What is happening is that something in the transaction machinery is > causing the directory to be completely recreated. The transaction gets rolled back by its destructor. This is critical so it happens whenever an exception occurs that unwinds the stack. Unfortunately, what's happening with clone is we're trying to delete the directory during exception propagation. And a reference to the transaction is held in the exception backtrace stack frames so it still exists until the exception is completely resolved. So there's no way to do the directory delete inside the exception handling cleanly. But we can handle it similarly to the transaction itself: use an object with a destructor. manifest hash: fc38550a20d64d08333f256bbedc312493c1390b -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.0 (GNU/Linux) iD8DBQFCxDT2ywK+sNU5EO8RAjikAJ0Tej56rAutxQDfYzVbFGtT1sEC5ACgmVds /fwdQyHn+FwshugqXLemUaM= =3f78 -----END PGP SIGNATURE-----

     1

#!/bin/sh -x

     2

     3

mkdir t

     4

cd t

     5

hg init

     6

echo This is file a1 > a

     7

echo This is file b1 > b

     8

hg add a b

     9

hg commit -t "commit #0" -d "0 0" -u user

    10

echo This is file b22 > b

    11

hg commit -t"comment #1" -d "0 0" -u user

    12

hg update 0

    13

rm b

    14

hg commit -A -t"comment #2" -d "0 0" -u user

    15

# in theory, we shouldn't need the "yes k" below, but it prevents

    16

# this test from hanging when "hg update" erroneously prompts the

    17

# user for "keep or delete"

    18

yes k | hg update 1

    19

# we exit with 0 to avoid the unavoidable SIGPIPE from above causing

    20

# us to fail this test

    21

exit 0