| Commit message (Collapse) | Author | Age |
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
| |
This is not set here, which is just as well because it would have been
quite wrong.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
before (5.3)
1044.27user 126.77system 6:26.89elapsed 302%CPU (0avgtext+0avgdata 402420maxresident)k
1778718inputs+5123720outputs (2370major+57693524minor)pagefaults 0swaps
after (5.3 + these changes)
1072.81user 131.38system 6:25.02elapsed 312%CPU (0avgtext+0avgdata 402448maxresident)k
1777558inputs+5124952outputs (2370major+59176578minor)pagefaults 0swaps
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
| |
Tolerate failure of the error log reporting arrangement. That way the
retry loop is effective even if we don't have a tty.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
| |
This was supposed to sleep for 5 seconds _instead_. This is a
kind-of-expected error and it's best to carry on a bit quicker.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
| |
apt-key seems to feed gpg --import an empty pipe. This makes gpg exit
with status 2. We think this is trouble, but it isn't.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Recognise these with an ad-hoc pattern on the command line arguments.
When they occur, save stdin to a temporary file.
We can't do this generally, because stdin might be some thing that
gpg's caller doesn't expect gpg to eat.
When we do this saving, print the size in bytes of the input file.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
| |
This is also an attempt to reduce the impact of the gnupg races.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
|
|
| |
The retry strategy is, empirically, fairly successful. But the
failure probability is still high enough to be a real nuisance. Let's
try running it a 3rd time if the second fails. Maybe this will reduce
the overall impact.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
|
|
| |
Work around gnupg agent connection races by having our stunt gpg
wrapper simply try running gpg again, once, if it exits 2.
This does not fully suppress the bug but it does significantly reduce
the probability.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
|
|
| |
This avoids lintian complaining about our stupid test packages.
We need the stunt debuild because debuild likes to launder the
environment, removing our stunt lintian from PATH !
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
| |
Like official one does, nowadays.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
| |
Some programs might treat that as a path of implicitly ".".
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Specifically:
* Properly dereference the variable when looking for the stunt
dir to be there already
* Do not add "." to the path if the path starts out empty.
* Crash _with the right message_ in dpkg-parsechangelog if PERLLIB is
unset. (PERLLIB mustn't be unset because we're trying to strip
our own value from it.)
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
| |
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
|
|
|
|
|
|
|
| |
Fall back to system implementation if command line options are
requested that we don't understand. This is a bit fiddly - we need to
strip the stunt entry out of PATH and also out of PERLLIB.
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
|
| |
|
|
|
|
|
|
| |
This saves a further 13s. Overalll the stunt changelog machinery
saves 28s out of 100s (in my example case, the current
debpolicy-newreject).
|
|
This program takes 16ms instead of ~700ms on my computer. This saves
15s out of 100s for the debpolicy-newreject test (in its current
state).
Currently this is only used by various things in devscripts because
dpkg-source uses /usr/lib/dpkg/parsechangelog/debian directly via a
Perl module Dpkg::Changelog::Parse.
|