Age | Commit message (Collapse) | Author | Files | Lines |
|
There needs to be some whitspace after the FIFO tag, otherwise k4dirstat
(silently!) stops parsing at this entry.
This was wrong in the initial commit to add FIFO support to kdirstat
4ce858d7c593 ("kdirstat: Handled named pipes (fifos)"), manual testing at the
time apparently only verified that the file was generated, not that it was
valid, sorry.
I updated the test yarns to also check for FIFO, however this is only run
"ASSUMING extended attributes are allowed for users", which does not appear to
be the case on my system. Also the testshould probably test for all sorts of
files, not just regular and FIFO.
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
|
|
This used to assume things of gpg's output, which break when gpg
becomes version 2.1.something. This changed version of the step should
hopefully work with either version of gpg.
|
|
|
|
This used to work, but a recent change to yarn broke it. The change is
that yarn now supports (or its next release will do so) arbitrary
language for IMPLEMENTS, and as a result yarn no longer puts empty
lines between files it concatenates and feeds to the shell.
|
|
|
|
Signed-off-by: Ben Boeckel <mathstuf@gmail.com>
|
|
|
|
|
|
|
|
|
|
On Sat, 2014-12-06 at 19:15 +0200, Lars Wirzenius wrote:
> On Sun, Nov 16, 2014 at 11:24:11AM +0000, Ian Campbell wrote:
> > This implements an ls variant whose output format is compatible with the
> > kdirstat cache format. The best reference I can find for this is from an old fork:
> > https://github.com/thomas-joiner/k4dirstat/blob/master/kdirstat/cache-file-format.txt
> >
> > Recent versions of k4dirstat appear to be able to read the files produced just
> > fine, although this document is not present in their source AFAICT.
> >
> > I find k4dirstat to be an excellent way to visualise what is included in a
> > backup, e.g. for the purposes of excluding file types or marking directories
> > with a CACHEDIR.TAG. My previous solution involved post-processing the output
> > of ls which require a ssh to the machine with the files (for stat() purposes).
> > This approach pulls all of the data out of the obnam repo, so is much faster,
> > less hacky and gets the status of the actual generation and not the current
> > files etc.
>
> Thank you, Ian. I've applied the patch and pushed the changes to
> master on git.liw.fi.
Thanks.
> It would be nice if someone wrote a test case for this (in yarn form),
> so that I'll notice when I break it.
Your wish etc etc... (it even found a bug!)
8<----------------
>From 8f1aecfee8ec34e5d060f81ee47f9fd8aacdc201 Mon Sep 17 00:00:00 2001
From: Ian Campbell <ijc@hellion.org.uk>
Date: Sun, 7 Dec 2014 15:19:57 +0000
Subject: [PATCH] Add a yarn for testing kdirstat cachefile generation.
Rather simplistic, but caught one bug (unwanted newline at start of file) which
is also fixed here.
|
|
|
|
|
|
|
|
|
|
Add test case. Fix code.
Reported-by: Junyx
|
|
|
|
Doing this exposed a bug in yarn, which meant that the test needs a
newer yarn to run. (Not yet released.)
|
|
Merge branch 'liw/enc-cmdtest-conv'
|
|
|
|
|
|
|
|
Remove use of extended attributes from some tests.
Make the remaining tests be conditional on whether
extended attributes actually work in the temporary
directory (in the user.* namespace).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This step failed under CI, but I haven't been able to reproduce
the failure otherwise. I'm hoping this extra output (which will
normally be hidden, but shown when a failure happens) will help
me find the problem.
|
|
Originally, we would create L/wanted, then do a manifest, then create
L/cache, then backup, restore, verify. This would sometimes fail, if
the full-second timestamp of L would change when L/cache was created.
So we (well, I) changed it so that we fiddled with the manifest just
before making a backup, but this was weird and confusing and would
still sometimes fail. This should now be much clearer and therefore
more robust.
|
|
The single file restore test had a bug, in that it was depending on
various test steps to execute fast enough that the wall clock did not
advance to the next second. This has been fixed by allowing manifests
to be done on single files instead of just directories.
Add a test for an extended attribute value changing, and that being
the only change in the live data. The backup needs to notice that it
has changed. This was intended to be the original change in this
commit, but I ran into the test bug above, so I had to fix that too.
I was too lazy to detangle the two sets of changes for commits.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This sets up a whole lot of infrastructure, too.
|
|
As part of this, make every repository operation step require an
explicit client name. The v6 reference formats have a client
name embedded that is the hostname of my laptop, so make those tests
use that name. This fixes what would not have run previously on any
system that isn't named havelock. Oops.
|
|
Remove cmdtest test for this.
To make things clearer in the new test, also change all the steps that
use a backup repository to have the user specify it explicitly.
|
|
|
|
|
|
|
|
It is just as, if not more, clear, and not much more tedious, to be
explicit about this.
|