| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| | | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
It can happen that when parsing a file puppet parses other manifests
if they get imported (this is at least true for site.pp, even in
ignoreimport=true). Thus those files are now "watched".
But puppetdoc needs to analyze all files, and since 99c101 we are now
checking if the file was already parsed to not reparse it again.
If that was the case, though, we weren't analyzing the produced code.
Thus it was possible to not produce documentation for the site.pp content.
Signed-off-by: Brice Figureau <brice-puppet@daysofwonder.com>
|
| | | |/ / / / /
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
The following manifest was crashing puppetdoc:
class test {
include "test::$operatingsystem"
}
Because the quoted string is "rendered" as a concat AST, which in turn
ended being an array when entering RDoc.
Signed-off-by: Brice Figureau <brice-puppet@daysofwonder.com>
|
| | |\ \ \ \ \ \
| | |/ / / / /
| |/| | | | | |
|
| | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
- previously, Puppet would search $LOAD_PATH and just
load applications in the first $LOAD_PATH to have
the directory puppet/application. Now multiple paths
can contain applications.
|
| | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
The commit for #2597 included a test that asserted the text resulting
from detecting a cycle. However, the cycle detection could start
randomly from any node, resulting in different text in the error. I'm
not sure what the randomization key would be based on since the test
failed consistently over dozens of runs for me, and didn't for Daniel.
Paired-with: Daniel Pittman <daniel@puppetlabs.com>
|
| | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
We only really want to verify that the code exits, but the current
implementation emits text directly; this results in messing up the tests,
which we can avoid with this tiny shim.
Reviewed-By: Nick Lewis <nick@puppetlabs.com>
|
| | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
This was tightly coupled to the code implementation; it mostly still is, but
now allows argument extension without needing to adjust the test which is
only focused on the first argument anyhow.
Reviewed-By: Nick Lewis <nick@puppetlabs.com>
|
| | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
This code was stubbing Puppet settings, which is no longer required now we
reset them between tests. Using the real thing reduces points that the rest
of the code can break, too.
As a side effect this caused Puppet[:trace] to be "true", which meant that we
emitted a huge, nasty backtrace during the testing of a specific internal
failure handling case.
Reviewed-By: Nick Lewis <nick@puppetlabs.com>
|
| | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
We replace it with an instance of the actual resource we are testing, which
reduces the number of ways this code is tied to the specific implementation.
Reviewed-By: Nick Lewis <nick@puppetlabs.com>
|
| | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
This was a helper that implemented rspec style "shared behaviour" for
Test::Unit; now that we have moved on we can use the upstream implementation
instead. This eliminates a whole bit of code we have to maintain.
Reviewed-By: Nick Lewis <nick@puppetlabs.com>
|
| | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
This is replaced with the new my_fixture{,s} methods; old fixture data is
ported into the spec tests at the same time, but left where it was against
unit tests that require it.
Reviewed-By: Nick Lewis <nick@puppetlabs.com>
|
| | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | | |
We had a combination of bad logic, and bad testing, and a nasty behaviour of
Mocha <= 0.9.10 that would result in a false pass for one of our tests.
This not only falsely passed, but hid an infinite loop retrying decompression
on an invalid data stream; it could be triggered by anything that sent an HTTP
request with an invalid compressed body, resulting in a livelock.
Paired-with: Jesse Wolfe <jesse@puppetlabs.com>
Signed-off-by: Daniel Pittman <daniel@puppetlabs.com>
|
| |\ \ \ \ \ \ \ |
|
| | | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
Reviewed-By: Nick Lewis
|
| | | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
into next"
This reverts commit 448a439f5abc3d51accececb678e9c5f547f7615, reversing
changes made to 06939c51a3f675137b53fac8a521132a4c9cfcbe.
As per discussion in http://projects.puppetlabs.com/issues/5691#note-5
|
| |\ \ \ \ \ \ \ \ |
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
This is special feature that changes the process name of the running puppet
entity to display its current activity.
It is disabled by default, and can be enabled by sending the QUIT signal
to the process in question (or calling enable through the code).
This system can work only if some "probes" are integrated in the core puppet
codebase. Since tools to visualize process names have a large refresh time
(ie more than 1s) it only makes sense to track long activities (like compilation,
transaction or file serving).
Those probes are the subject of a subsequent patch.
This system tracks every thread activity and form a strings which will
be used as the process name. Due to the way it is implemented it is
possible that it doesn't work on all platforms (I tested successfully
on osx and linux). On some systems the space available is dependent on
the original size of the full command. That's why if this string is longer
than a 50 characters, the string is scrolled (like stock market tickers).
Note: This is not intended to be a generic instrumentation system. Also, being
block based means that it can reduce performance if the instrumentation
probes are used in tight inner loops.
Signed-off-by: Brice Figureau <brice-puppet@daysofwonder.com>
|
| |\ \ \ \ \ \ \ \ \ |
|
| | | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
When the '--graph' option is specified, generate a new 'cycles.dot' file and
report the location of that to the user. This contains only the cycles, in
dot format, allowing a visual representation of the cycle to be obtained
quickly.
This will include up to 10 paths through the cycle in the graph, and only one
in the display to the user, to reduce information overload.
|
| | | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
Rather than reporting the cluster of vertexes in the dependency graph, which
is interesting but not entirely informative, we now calculate and report the
paths through the graph that form cycles.
This returns the most useful information, which is the exact path that the
dependency cycle has, allowing the user to (hopefully) immediately target it
and start to work out why the cycle has formed there.
We strongly prefer short paths through the dependency graph within the cycle,
which should report the most useful loops to target first; extended loops
involving more items will show up later if they are independently created.
We also limit the number of paths reported (default: 10) to avoid overwhelming
the error report with the combinatorial explosion that can easily result
from a large scale cycle. (eg: Package => User => Package or something.)
|
| | | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
In order to bypass the limitations of the C stack, which is also the Ruby
stack, we replace the simple and clear recursive Trajan implementation with an
iterative version that uses the heap as the stack.
This is somewhat harder to read, but can now run a 10,000 vertex deep linear
dependency relationship where, previously, 1,250 was about the limit on my
machine.
This should now be bounded by the size of the heap rather than the stack on
all platforms -- though it would be nice to get rid of the magic and return to
the recursive version if Ruby ever follows Perl down the sensible path of
essentially unlimited recursion by writing that code for us in the
interpreter...
|
| | | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
This implements Tarjan's algorithm for finding strongly connected components
in a directed graph, and leverages that to find cycles.
This allows us to report the minimum set of nodes in each cycle, as well as
reporting each cycle discretely if there are multiple of them.
While this can still produce overwhelming and/or unhelpful output, it
represents a large step forward in delivering useful information when a cycle
is detected.
This presently reports the set of nodes that contain the cycle, in no
particular order, rather than the set of edges connecting those nodes.
Sadly, it also suffers a limitation: the implementation of Tarjan's algorithm
used to find strongly connected components is recursive, so is limited by the
maximum Ruby stack depth to dependency chains less than 1,000 nodes deep.
While this is probably not a limit in practice, it is a nasty limitation, and
other considerations (like Ruby stack consumption across versions) could
trigger this much sooner than is desirable.
|
| | | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
Split out the reporting from a single line (often with literally hundreds or
thousands of items) into a multi-line report. This is still nasty, but at
least it is easier to use as input to other systems.
This will also auto-join to a single line when sent to targets such as syslog
that do not approve of newlines in messages; this preserves the utility of the
message without needing to lose console utility.
|
| | | |_|_|_|_|_|_|/
| |/| | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
The SimpleGraph class was reporting duplicate data when printing cycles:
Notify[c]Notify[c] => Notify[d]
Notify[a]Notify[a] => Notify[b]
This was caused by throwing the array representation of the edge into a
string, rather than just the relationship data; we only care about the later,
so now we only emit that later and have the correct text in the error.
|
| |\ \ \ \ \ \ \ \ \
| |/ / / / / / / /
|/| | | | | | | | |
|
| | | |_|/ / / / /
| |/| | | | | | |
|
| | | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
Third party scripts, and complex command line tools, depend on being able to
configure the run_mode value at runtime, not just when they fire up.
For better or worse we used to allow this sort of thing to work, but stopped,
and we have no sane, safe and consensual alternative, so we broke a bunch of
client code.
This enables the feature again, but does not add any safety catch; you can now
happily slice off your own feet with this, if you really want to.
|
| | | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
We previously had an ordering dependency in the autoflush option, which was
statically read from defaults when the log destination was configured.
We add a hook in the defaults to update the log subsystem, which in turn
updates log destinations, when autoflush is changed.
This would work as desired:
puppet agent --autoflush --logdest=file
This would not work, as autoflush would be false:
puppet agent --logdest=file --autoflush
Now those changes propagate correctly.
Paired-with: matt@puppetlabs.com
|
| |\| | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
* 2.6.x: (46 commits)
Augmentation of tests for prior commit
Fix to fix for #5755 -- backref serialization issues in zaml
Fixed #5564 - Added some more fqdn_rand documentation
Fixed #4968 - Updated list of options turned on by --test in documentation
(#5061) - allow special hostclass/define variables to be evaluated as defaults.
(#6107) Fix an error when auditing a file with empty content
Remove already initialized constant warning from file_spec.rb tests
(#5566) Treat source only File checksums as syntax errors when used with content
Rename variable used in File type validation to be more clear
Remove invalid "timestamp" and "time", and add missing "ctime" File checksum types.
Remove order dependency when specifying source and checksum on File type
Bug #5755 -- ZAML generates extra newline in some hash backreferences.
bug #5681 -- code fix to handle AIX mount output
Bug #5681 -- parse AIX mount command output.
Spec for #5681 to allow parsing of AIX mount output in mount provider
Fixed #6091 - Changed POSIX path matching to allow multiple leading slashes
Bug #6091 -- test leading double-slash in filenames are allowed.
Fixed #6071 - Fixed typo and improved exec path error message
Fixed #6061 - Allowed -1 as password min/max age
Bug #6061 -- verify that negative {min,max}_password_age are accepted.
...
Manually Resolved Conflicts:
lib/puppet/util/zaml.rb
spec/unit/util/zaml_spec.rb
|
| | |\ \ \ \ \ \ \ |
|
| | | |\ \ \ \ \ \ \
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
* bodepd/feature/2.6.4/5910:
(#5910) Improved logging when declared classes cannot be found:
|
| | | | | |_|/ / / /
| | | |/| | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Previously, when a class could not be found, it was displaying the same error message as when a resource type could not be found. This resulted in confusing error message: Invalid resource type class, when really it should display the name of the class that could not be found.
My patch changes the error message to:
Could not find declared class #{title}
|
| | | |\ \ \ \ \ \ \
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | |
| | | | | | | | | | |
2.6.next
* 'ticket/2.6.x/5913' of git://github.com/mitchellh/puppet:
(#5913) Fix Puppet::Application.find constant lookup behavior
|
| | | | | |_|_|_|_|/
| | | |/| | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Puppet::Application.find now only looks in the Puppet::Application
namespace for the given constant.
|
| | | |\ \ \ \ \ \ \ |
|
| | | | |/ / / / / / |
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
There is no good answer to tests that depend on the order of itteration over hashes.
|
| | |/ / / / / / /
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
This commit addresses the original issue that the change reverted in the previous
commit for #5755 was intended to fix by removing the special case on labels in
emit (lables, even if they are never generated, are not "new lines" and thus
@recent_nl should always be set to false when one is emitted).
It also partially addresses a related issue wherein temporary strings generated
when field names are constructed recycle their object_id (they are temporary)
and thus cause incorrect back references. This commit "fixes" the problem by
never generating backrefs to strings (treating them as immutable).
It does not address other suspected issues such as thread safety durring serialization
due to the use of class variables to store the seen-object hash or the use of object
ids as "permanently unique" identifiers.
Paired with: Daniel Pittman
Advice & Commiseration: Jesse Wolfe
|
| | |\ \ \ \ \ \ \
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
* feature/master/5061:
(#5061) - allow special hostclass/define variables to be evaluated as defaults.
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
I have always been annoyed that special variables for defines and hostclasses can not be evaluated as param defaults. Special variables are: $name, $title, $module_name.
Code example:
class x (
foo = $name
) { notice($foo)}
should print x, and with my patch, it does.
Reviewed-by: Paul Berry <paul@puppetlabs.com>
|
| | |\ \ \ \ \ \ \ \
| | |/ / / / / / /
| |/| | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
* ticket/2.6.next/5045:
(#5045) Cleaning up some tests and code
(#5045) External node classifiers should be able to specify params for classes
(#5045) Adds support to resource/type to also accept a param hash
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Renamed some variables to be clearer, made tests use less stubbing,
added some additional tests and got rid of some unecessary logic.
Paired-with: Dan Bode
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
It facilitates the support for param classes from the ENC. It adds
support for classes to be passed as a hash to the evaluate_classes
method. If a hash of classes is specified, it also evaluates duplicates.
I also had to convert the hash to an array for tags to be applied
correctly.
Reviewed-by: Matt Robinson
|
| | | | |_|_|_|/ /
| | |/| | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
The params are added as attributes when the resource is created.
Also, even if the class already exists, if params are passed, we still try to create it.
Reviewed-by: Matt Robinson
|
| | |\ \ \ \ \ \ \
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
* ticket/2.6.next/6107:
(#6107) Fix an error when auditing a file with empty content
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
The manifest:
file { "/tmp/foo" :
ensure => present,
audit => content,
}
produced the error:
err: /Stage[main]//File[/tmp/foo]/ensure: change from absent to present
failed: Could not retrieve content for from filebucket: private method `sub' called for nil:NilClass at
/Users/matthewrobinson/work/puppet/test.pp:4
This was due to logic in content assuming that if you didn't specify
content while you were auditing it you must have specified a source.
The code paths in the file type badly need a cleanup so that these sorts
of errors aren't so difficult to track down and things are easier to
test.
Paired-with: Daniel Pittman
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
rspec pulls constants from the implementation into the test, so we don't
need to redefine it.
Reviewed-by: Jesse Wolfe <jesse@puppetlabs.com>
|
| | | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | |
| | | | | | | | | |
Certain checksum types (ctime, mtime) only make sense when used with the
'source' File parameter, since there is no way to check them on raw
strings.
Given the limitations of the current checksumming implementations, it is
likely to introduce unexpected behavior when using the 'none' checksum
type and either one of the 'source', and 'content' File parameters.
Because of this, it is now a syntax error to use a checksum of 'none' with
either parameter.
Paired-with: Jesse Wolfe <jesse@puppetlabs.com>
|
| | |/ / / / / / /
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
If source was specified after setting the checksum, it would cause the
checksum to be set back to :md5. This was completely unnecessary, because
the checksum has its own default of :md5.
Paired-with: Jesse Wolfe <jesse@puppetlabs.com>
|
| | | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | |
| | | | | | | | |
This data structure generates YAML with an extra newline that violates the
syntax rules and all:
list = [1]
{ :a => list, :b => list }.to_yaml
This breaks real client use of the YAML catalogs, not to mention our own use
of cached catalogs...
|
