| Age | Commit message (Collapse) | Author |
|---|
|
Using different ways of opening files means we have different behaviour
and error messages for them, so by the same for all we can have more
uniformity for users and apt developers alike.
|
|
Failing on too much data is good, but we can do better by checking for
exact filesizes as we know with hashsums how large a file should be, so
if we get a file which has a size we do not expect we can drop it
directly, regardless of if the file is larger or smaller than what we
expect which should catch most cases which would end up as hashsum
errors later now a lot sooner.
|
|
We tend to operate on rather large static files, which means we usually
get Content-Length information from the server. If we combine this
information with the filesize we are expecting (factoring in pipelining)
we can avoid reading a bunch of data we are ending up rejecting anyhow
by just closing the connection saving bandwidth and time both for the
server as well as the client.
|
|
It is highly unlikely to encounter fields which start with HTTP in
practice, but we should really be a bit more restrictive here.
|
|
The comment says this is intended, but looking at the history reveals
that the comment comes from a different era. Nowadays we don't really
need it anymore (and even back then it was disputeable) as we haven't
used that file for our update in the end and nothing really needs this
file after the update.
Triggered is this by 188f297a2af4c15cb1d502360d1e478644b5b810 which
moves various error conditions forward including this code expecting the
file to exist – but it doesn't need to as download could have failed.
We could fix that by simple checking if the file exists and only stage
it if it does, but instead we don't stage it and instead even rename it
out of the way with our conventional FAILED name (if it exists).
That restores support for partial mirrors (= in this case mirrors which
don't ship pdiff files). Note that apt heals itself even if only such a
mirror is used as the update is successful even if that error is shown.
Closes: 869425
|
|
This makes the code easier to read.
|
|
This makes it easier to see which headers includes what.
The changes were done by running
git grep -l '#\s*include' \
| grep -E '.(cc|h)$' \
| xargs sed -i -E 's/(^\s*)#(\s*)include/\1#\2 include/'
To modify all include lines by adding a space, and then running
./git-clang-format.sh.
|
|
Commit d7c92411dc1f4c6be098d1425f9c1c075e0c2154 parses the Components
section of (In)Release and attempts to detect the distribution's
supported components. While doing so, it handles component names with
slashes in a special manner, assuming that the actual component is only
the part after the final slash. This is done to handle
security.debian.org, which usually appears in sources.list as follows:
deb http://s.d.o/debian-security stretch/updates main contrib non-free
while the actual release file has:
Codename: stretch
Components: updates/main updates/contrib updates/non-free
While this special handing on APTs part indeed works for
debian-security, it emits spurious warnings on repositories that
actually use slashes in the component names *and* appear so in
sources.list.
We fix this by adding both component versions (whole and final part) to
the SupportedComponents array.
Closes: #868127
|
|
This caused spurious test failures.
|
|
The old curl based method is still available as 'curl',
'curl+http', and 'curl+https'.
|
|
When APT was trying multiple addresses, any later error
somewhere else would be reported with ConnectionRefused
or ConnectionTimedOut as the FailReason because that
was set by early connect attempts. This causes APT to
handle the failures differently, leading to some weirdly
breaking test cases (like the changed one).
Add debugging to the previously failing test case so
we can find out when something goes wrong there again.
|
|
The apt-transport-tor package operates via simple symlinks which can
result in 'http' being called as 'tor+https', so it must pick up the
right configuration pieces and trigger https support also in plus names.
|
|
This should fix some issues with dpkg normalizing such
values. Suprisingly enough apt treats the Version: field
the same, even with epoch vs without, but not when searching,
and does not strip the 0: from the output.
|
|
This gives the repository owner a chance to explain why this change was
needed – e.g. explaining the organisational changes or simply detailing
the changes in the new release made. Note that this URI is also shown
if the change is accepted, so it also draws attention to release notes
of minor updates (if users watch apt output closely).
|
|
The value of Origin, Label, Codename and co can be used in user
configuration from apts own pinning to unattended upgrades.
A repository changing this values can therefore have serious effects on
the behaviour of apt and other tools using these values.
In a first step we will generate error messages for these changes now
explaining the need for explicit confirmation and provide config options
and commandline flags to accept them.
|
|
The exception was made to give (script) users a one-release grace period
to adapt their setup to deal with apt enforcing signing of repositories.
As we are now at the start of a new release cycle its as good a time as
any to lift it now.
Removes-Exception: 952ee63b0af14a534c0aca00c11d1a99be6b22b2
|
|
|
|
The test keeps failing continously on Ubuntu, so let's
fix it for now.
|
|
Gbp-Dch: ignore
|
|
Gbp-Dch: ignore
|
|
On Travis CI running tests with code coverage enabled sometimes
generates profiling lines, which we filter out for a while now,
but that misses lines generated showing progress still causing test
failures, so more sed logic is added in the hopes to ignore them.
Extends: 58608941e6b58a46109b7cd875716b3d8054c4bf
Gbp-Dch: Ignore
|
|
An unknown code should be handled the same as the x00 code of this
group, but for redirections we used to treat 300 (and a few others)
as an error while unknown codes were considered redirections.
Instead we check now explicitly for the redirection codes we support for
redirecting (and add the 308 defined in RFC 7538) to avoid future
problems if new 3xx codes are added expecting certain behaviours.
Potentially strange would have been e.g. "305 Use Proxy" sending a
Location for the proxy to use – which wouldn't have worked and resulted
in an error anyhow, but probably confused users in the process.
|
|
There are very many HTTP errorcodes which indicate that the repository
isn't available at the moment or the connection has some kind of
problem. Given that we do not require Release files the result was that
these errors were ignored and the user presented with a message like
"Repository is no longer signed" which sends the user in the wrong
direction.
Instead of trying to figure out which http errorcodes indicate a global
problem we accept only 404 for ignoring and consider all the rest as
hard errors now causing us to stop instantly after the InRelease file
and print the errorcode (with short description from server) received.
|
|
Moving the code responsible for parsing the Index file from ::Done into
the slightly earlier ::VerifyDone allows us to still "fail" the download
if we can't make use of the Index for whatever reason, so that the
progress log correctly displays "Ign" instead of "Get" for the file.
This also makes quiet a few debug messages proper error messages (but
those are still hidden by default for Ign lines).
|
|
If we couldn't find an entry for a Sources file we would generate an
error while for a Packages file we would silently skip it due to
assuming it is missing because it is empty. We can do better by checking
if the repository declares that it supports a component we want to get
the file from and if not say so and hint at the user making a typo.
An example were this helps is mozilla.debian.net which dropped the
firefox-aurora component (as upstream did) meaning no upgrades until the
user notices manually that the repository doesn't provide packages
anymore. With this commit warnings are raised hopefully causing the user
to investigate what is wrong (sooner).
|
|
Changes in the past to the buildsystem and the testing framework broke
this little helper script – lets fix those problems to restore
functionality.
Gbp-Dch: Ignore
|
|
As the proxy commands are not executed as root, a user can run into
permission errors (s)he isn't expecting – as our switching is an
implementation detail – so the error message in that case should really
be better than a generic "error code 100" sending the user in the wrong
direction as that implies the command was executed, but errored out.
Closes: 857885
|
|
Gbp-Dch: Ignore
|
|
We setup a "horrible" environment in the apt-key testcase to check all
kinds of things, but we really should be making also at least a simple
apt update call, as that in turn will call apt-key which is how apt-key
is used in the non-testcase world, so that calling should be able to
deal with such environments as well.
Gbp-Dch: Ignore
|
|
Gbp-Dch: Ignore
|
|
We want to kill the agent if its home directory exists at that location,
not if it isn't there (leaving an army of processes around).
Gbp-Dch: Ignore
|
|
If the last alternative(s) of an Or group is ignored, because it does
not match an architecture list, we would end up keeping the or flag,
effectively making the next AND an OR.
For example, when parsing (on amd64):
debhelper (>= 9), libnacl-dev [amd64] | libnacl-dev [i386]
=> debhelper (>= 9), libnacl-dev |
Which can cause python-apt to crash.
Even worse:
debhelper (>= 9), libnacl-dev [amd64] | libnacl-dev [i386], foobar
=> debhelper (>= 9), libnacl-dev [amd64] | foobar
By setting the previous alternatives Or flag to the current Or flag
if the current alternative is ignored, we solve the issue.
LP: #1694697
|
|
We are in a dilemma here: The regression of sorts was introduced in 2013
with commit d8a8f9d7f0 allowing pkg modifiers for the upgrade commands.
That calls the autoremover as a sideeffect through and with it comes the
option to remove the garbage packages in these commands (similar to aptitude).
Having the option on the commandline is no problem – people aren't going
to request what they don't want (or so I hope), but the documentation
explicitly states that this option only effects install/remove and
mentions a config knob users might use and expect to not suddenly apply
(especially without documentation) to more commands.
Just reverting the commit is out of question, completely ignoring the
option breaks the workflow of every user who happened to use
--autoremove on the commandline for upgrade and expects that to work
given that it was accepted and worked in a stable release. Changing the
documentation to reflect reality while perhaps the simplest and cleanest
option contradicts freeze and is a surprising change we tend to avoid
like the plague while just leaving it be confuses all users who end up
believing the documentation even if was different in the last 3 years.
So what we do is a tricky compromise: The configuration option if read
from a file does apply only for install/remove as documented, while if
the option is encountered on the commandline it is accepted and applies
to the upgrade which should make 99% of the users happy. The rest has to
wait for us to figure out for buster how to get that documented and
implemented in a saner way.
Closes: #855891
|
|
In the intended usecase where this serves as a hack there is no problem
with double/single quotes being present as we write it to a log file
only, but nowadays our calling of apt-key produces a temporary config
file containing this "setting" as well and suddently quoting is
important as the config file syntax is allergic to it.
So the fix is to ignore all quoting whatsoever in the input and just
quote (with singles) the option values with spaces. That gives us 99% of
the time the correct result and the 1% where the quote is an integral
element of the option … doesn't exist – or has bigger problems than a
log file not containing the quote. Same goes for newlines in values.
LP: #1672710
|
|
In the case of build-dep and other commands where a file can be
passed we must make sure not to normalize the path name as that
can have odd side effects, or well, cause the operation to do
nothing.
Test for build-dep-file is adjusted to perform the vcard check
once as "vcard" and once as "VCard", thus testing that this
solves the reported bug.
We inline the std::transform() and optimize it a bit to not
write anything in the common case (package names are defined
to be lowercase, the whole transformation is just for names
that should not exist...) to counter the performance hit of
the added find() call (it's about 0.15% more instructions
than with the existing transform, but we save about 0.67%
in writes...).
Closes: #854794
|
|
$ uname -m
i686
$ date -d '0-12-25'
date: invalid date '0-12-25'
Test-Regression-In: 25a14d4ccfceb2698edce01092bc6a1dbe9fb217
|
|
This hides errors in the test suite because it will exit
with 0 here. Instead, just do exit 1 in most traps, and
do just the cleanup in the QUIT hook.
This fixes a regression introduced with the caching of the
GPG home directory in 4ce2f35248123ff2366c8c365ad6a94945578d66.
|
|
Most of them in (old) code comments. The two instances of user visible
string changes the po files of the manpages are fixed up as well.
Gbp-Dch: Ignore
Reported-By: spellintian
|
|
Normal cows moo every time they feel like it and it might be a "moo",
"moo!" or "moo?". This is completely unacceptable behaviour in our super
cow through as as a superior being it has to show its superiority over
the common cows and the meager easter eggs by being fully reproducible!
The second version of Chris' patch is modified to include an array of
tests for this feature – which doubles as explanation for some of the
moo lines by giving more exact dates – and falling back to current time
if the environment is invalid + passing time around instead of having an
invalid environment be an unrecoverable error (aka: Guru Meditation) as
that is more inline with how apt usually behaves: The wisdom of super cow
should be available to everyone, even to the most misfortune users not
capable of having a valid environment variable.
That makes the code slightly more ugly, so instead of requiring a young
follower to produce a third version a high priest of the cult applied the
finishing touches as he is used to the pain by now – and another round
with the slowpoke high priest would have been a serious threat to the
Debian release schedule which the cow would not approve.
Closes: #848721
Signed-off-by: Super Cow
Thanks: Chris Lamb for initial patch and guru meditation
|
|
If apt renames a file to .FAILED it leaves its namespace and is never
touched again – expect since 1.1~exp4 in which "apt clean" will remove
those files. The usefulness of these files rapidly degrades if you don't
keep the update log itself (together with debug output in the best case)
through and on 99% of all system they will be kept around forever just
to collect dust over time and eat up space.
With this commit an update call will remove all FAILED files of previous
runs, so that the FAILED files you have on disk are always only the ones
related to the last apt run stopping apt from hoarding files.
Closes: 846476
|
|
The mode wasn't working at all if not used together with --fix-missing
which while likely to come in pairs its legal to use standalone.
Regression-in: eb1f04dda07c2b69549ad9fd793cca0e91841b3e
|
|
Unlikely that anyone is actually running into this, but if we asked to
not generate a cache and avoid it in the first step we shouldn't create
one implicitly anyway by displaying the statistics.
|
|
This prevents CI failures from happening in 1.3 and 1.2 and
might actually be more complete.
Gbp-Dch: ignore
|
|
This will avoid people from thinking that they have to do nothing
when they change the set of files.
Gbp-Dch: ignore
|
|
Curl requires URLs to be urlencoded. We are however giving it
undecoded URLs. This causes it go completely nuts if there is
a space in the URI, producing requests like:
GET /a file HTTP/1.1
which the servers then interpret as a GET request for "/a" with
HTTP version "file" or some other non-sense.
This works around the issue by encoding the path component of
the URL. I'm not sure if we should encode other parts of the URL
as well, this one seems to do the trick for the actual issue at
hand.
A more correct fix is to avoid the dequoting and (re-)quoting
of URLs when a redirect occurs / a new request is sent. That's
been on the radar for probably a year or two now, but nobody
bothered implementing that yet.
LP: #1651923
|
|
Our implementation of wildcards was rudimentary. It worked for some
common ones, but it was also broken: For example, armel matched any-armel,
but should match any-arm.
With this commit, we load the correct tables from dpkg. Supported are
both triplets and quadruplet tables (the latter introduced in dpkg 1.18.11).
There are some odd things we have to deal with in the cache filter for
historical and API reasons:
* The character "*" must be accepted as an alternative to any - in fact
it may appear anywhere in the wildcard as we also allow fnmatch() style
wildcard matching on the commandline.
* The code might get passed an arch with a minus at the end, for example
the cmdline "install apt:any-arm-" will first try to check if any-arm-
is a valid architecture. We deal with this by rejecting any wildcard
ending in a minus.
* Triplets are actually implemented by extending them to faux quadruplets
- by prepending a "base" component for the architecture tuple, and "any"
if there is a wildcard component.
Once we have constructed a wildcard, it is transformed into an fnmatch()
expression for historical reasons. In the future, we should really get a
tuple class and implement matching in a better, more explicit way.
This does for now though - it passes all the test cases and accepts all
things it should accept.
Closes: #748936
Thanks: James Clarke <jrtc27@jrtc27.com> for the initial patch
|
|
Run the test for kfreebsd-i386 and amd64 and pass "amd64" as
an additional argument to the function. This tests that the
argument is used and thus ParseDepends returns the amd64
results even on a different architecture like i386.
|
|
The idea is simple: Each¹ Find*( call starts with a call check if the
given option (with the requested type) exists in the whitelist. The
whitelist is specified via our configure-index file so that we have
a better chance at keeping it current. the whitelist is loaded via a
special (undocumented for now) configuration stanza and if none is
loaded the empty whitelist will make it so that no warnings are shown.
Much needs to be done still, but that is as good a time as any to take a
snapshot of the current state and release it into the wild given that it
found some bugs already and has no practical effect on users.
¹ not all in this iteration, but many
|
|
Users end up believing that this is a --force mode as -f is common for
that, but apt doesn't have such a mode and --fix-broken is really not
about forcing something but actually trying to fix the breakage which
tends to be the result of a user forcing something on its system via
low-level forced dpkg calls.
Example: The "common" pattern of "dpkg -i ./foo.deb; apt install -f" is
nowadays far better dealt with via "apt install ./foo.deb".
And while at it the two places handing out this suggestion are changed
to use the same strings to avoid needless translation work in the future
and the suggestion uses 'apt' instead of 'apt-get' as this will be run
interactively by a user, so its a good opportunity to showcase what we
can do and will allow us to be more helpful to the user.
Closes: #709092
Thanks: Kristian Glass for initial patch!
|
|
The config list APT::Build-Essential gets a similar treatment to other
lists now by having the value of the option itself be an override for
the list allowing to disable build-essentials entirely as well as
adding/overriding as usual by now in other lists.
Reported-By: Johannes 'josch' Schauer on IRC
|
