view tests/test-gpg.t @ 18519:ca430fb6a668 stable

tests: fix toctou race in tinyproxy.py (issue3795)

test-http-proxy.t sometimes failed with:
    File ".../tests/tinyproxy.py", line 110, in _read_write
      data = i.recv(8192)
  error: (104, 'Connection reset by peer')

This might have started showing up with a9fd11ffa13f ... but it has apparently
also been seen before. I don't see anything in a9fd11ffa13f that can explain
it. It seems to be a race in test, in the tinyproxy helper:

Tinyproxy found an incoming socket using select(). It would break the loop if
an error had been detected on the socket, but there was no error and it tried
to recv() from the socket. That failed - apparently because it had been reset
after select().

Errors in the recv() will now be caught and will break the loop like errors
detected by select() would.

(send() could also fail in a similar way ... but using the same solution there
and losing data we have read doesn't feel right.)

Test the GPG extension

  $ "$TESTDIR/hghave" gpg || exit 80
  $ cat <<EOF >> $HGRCPATH
  > [extensions]
  > gpg=
  > 
  > [gpg]
  > cmd=gpg --no-permission-warning --no-secmem-warning --no-auto-check-trustdb --homedir "$TESTDIR/gpg"
  > EOF
  $ hg init r
  $ cd r
  $ echo foo > foo
  $ hg ci -Amfoo
  adding foo

  $ hg sigs

  $ hg sign 0
  signing 0:e63c23eaa88a

  $ hg sigs
  hgtest                             0:e63c23eaa88ae77967edcf4ea194d31167c478b0

  $ hg sigcheck 0
  e63c23eaa88a is signed by:
   hgtest

verify that this test has not modified the trustdb.gpg file back in
the main hg working dir
  $ "$TESTDIR/md5sum.py" "$TESTDIR/gpg/trustdb.gpg"
  f6b9c78c65fa9536e7512bb2ceb338ae  */gpg/trustdb.gpg (glob)

  $ cd ..