| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
|
|
This is a feature that helps ELAPI.
It makes lookup of the fields that need
to be resolved for every event a bit faster.
The idea is to be able to put a 'pin'
into a specific place while iterating
the collection and make this place a new
"wrap around" place for the collection.
This means that next time you
iterate this collection you will start
iterating from the next item and
the item you got before pin will be last
in your iteration cycle.
Here is the example:
Assume you have two collections that you need
to compare and perform some action on collection
1 based on the presense of the item in collection 2.
Collection1 = A, B, C, D, E. F
Collection2 = A, C, F
The usual approach is to try A from collection 1
against A, B, C from collection 2. "A" will be found
right away. But to find "F" it has to be compared
to "A" and "C" first. The fact that the collections
are to some extent ordered can in some cases
help to reduce the number of comparisons.
If we found "C" in the list we can put a "pin"
into the collection there causing the iterator
to warp at this "pin" point. Since "D" and "E"
are not in the second collection we will have
to make same amount of comparisons in traditional
or "pinned" case to not find them.
To find "F" in pinned case there will be just one
comparison.
Traditional case = 1 + 3 + 2 + 3 + 3 + 3 = 15
Pinned case = 1 + 3 + 1 + 3 + 3 + 1 = 12
It is a 20% comparison reduction.
|
|
Created a new module to hold functions
related to iterator and iterating
collections. Planning to add new functions
but the main collection module is already
too big. So this patch just moves code around
and fixes the build making foundation for
the next patch.
|
|
Needed item comparison functions and realized
that the easiest way to test them would be using
sorting. Since there already been a ticket #73
to do that I added function to sort collection
based on different properties of the item.
COLLECTION Fixing issues with comparisons
COLLECTION Adding do-while to macro
|
|
Necessary for RPM builds on RHEL5
|
|
|
|
The hashing logic was internal to the collection item.
But if someone wants to effectively deal with
the items and compare the property to a string
he should compare hashes first. But it was not possible
without the provided functions. As a result some
of the ELAPI modules had to take advantage of
knowledge of the item structure. This is bad.
So this patch lays foundation for refactoring
of the ELAPI code that was using internals of the item
directly (file_util.c mostly).
Also patch adds a unit test that was required for
testing new functionality and for ticket #83
|
|
This patch adds better options for
copying collections in flat mode.
It allows caller of the interface
to control prefixing of the fields
when one collection is appended to another.
It also avoids creating prefixes when the
collection is simply copied in flat mode.
Also for ELAPI I realized that the most efficient
way to deal with the "resolved" event
(event where all templeted values are actually replaced
with the real values) is to add a callback
capability to a copy collection function
so that the callback can be used to modify
the data (resolve it) while the copy operation
is in progress. This approach eliminates the need
for separate set of lookups after the event
is already copied.
|
|
This patch addresses several issues:
a) Adds capability to add or copy the collections
in flattened mode but construct names of attributes
in dotted notation. For example when you append collection
"sub" with items "foo" and "bar" previously you
could add them as "foo" and "bar" not you can
flatten them and the names will be "sub.foo"
and "sub.bar" this allows better processing of the
attributes in the elapi message.
b) Removes old implemntation of the
copy collection function.
c) Removes the col_set_timestamp,
this functionality has been moved to ELAPI
long ago.
d) Updates collection unit test.
e) Updates elapi to use new functionality
f) Updates elapi unit test
Have run under valgrind with no problems.
|
|
- replaced mailing list address
- let sssd base components read version from VERSION
|
|
Started looking at the ticket #107 related to
traverse functions. Realized that the return values
are not consistent. That ovelapped with the work
that I wanted to do for ticket #103 - errno cleanup.
So I (across collection, INI and ELAPI):
* Made the return codes consistent (where found)
* Removed errno where it is not needed
While was testing used valgrind and found a nasty
problem when the value was added to collection with
overwriting duplicates the count was decreased improperly.
Fixing collection.c to not decrease count made
valgrind happy. While I was debugging this
I also spotted several build warnings in trace
statements when the " exp ? v1 : v2 " was used.
Fixed those.
In ini_config.c there was a trace stament that used
variable after it was freed. Removed trace stament.
|
|
The targets are the destinations which
caller wants to send the events to.
The sinks are now on the second level
under targets and constitute a so called
fail over chain for a target.
Such approach eliminates the need for complex
routing function.
The dispatcher keeps the list of targets in a collection.
The element in the collection is the target context.
Also gispatcher keeps the list of the sinks
in a separate collection.
Each target context has a list of the sinks
associated with this target. But those are just
pointers (at least for now) to the sinks
form the list kept by dispatcher.
I had to add some internal debug callbacks
to be able to see that all the internals of
the dispatcher are actually in order.
See the conttent of config file for more comments.
Also see information posted on SSSD wiki.
https://fedorahosted.org/sssd/wiki/WikiPage/ELAPIInterface
|
|
Patch adds checks for NULL to the trace macros.
It also eliminates the unused trace.h in the collection directory.
|
|
I started to cleanup the unit tests from the type cust around NULL and
found several problems that I had to address:
1) The choice of the "." as a search separator
turned out to be a poor choice.
The problem was that the file name has "."
and INI was relaying on files to be used as
property names.
I corrected that part in the INI but after discussion
with Simo we decided to switch from "." to "!"
as special symbol anyways.
2) Found that the property rename was not
reinitializing the hash. Corrected.
Added ticket to add unit tests around it (#83).
|
|
Addressing ticket #71.
The searches were not taking advantage of the hashes,
now they are.
|
|
|
|
The remove functions extract and remove
items, they act differently from the way
the delete_property function works.
The new functions allow deletion with
the disposition while the delete_property
only deletes specified property.
The delete_property function is left as is
since there are some use cases when it is
more efficient to use it rather than new
remove_item_xxx ones.
|
|
This is just a part of the interface,
a beginning. Most likely some of the functions
will be altered but it is a starting point.
For example in future there will be
a way to override some of the parts of the
default template using the application configuration file.
Removed obfuscation of the data types based on
discussion with Simo.
|
|
During a review of the previous patch the two issues
were found:
a) The col_iterator_up function was not implemented properly
so it got reworked. New implementation changes
the way error condition is handled. Comments were updated accordingly.
b) There was a mising check for validity of the argument in
the col_insert_into_current function. Check was added.
c) Unit test modified to reflect the change in functionality.
|
|
The collection is hearachical. The flattening
of the collection was not implemented before
both for traversal and copying. This patch
introduces functionality to traverse or
iterate through collection as flat set
and also copy collection into another flattening
it and automatically resolving conflicts.
Also imptoved tracability and fixed memory leak
in unbind iterator code.
|
|
Previously, only the SSSD server components were being built with
compile-time warnings enabled. This patch will ensure that all
components in common and sss_client are also built the same way.
|
|
There was a static global structure used in iteration and in
traversing the collection. It has been removed and replaced with
a better solution.
|
|
Fix consts and function declarations
|
|
Patch prepends prefix "col_" to all
functions related to collection.
This caused some formatiing issues
so the alignement was addressed too.
|
|
Adds a unit test for deletion and re-adding
of the elements to collection.
Small syntactical fix in "stack".
|
|
Patch adds two interfaces for stack and queue using
collection as underlaying implementation.
It also fixes a bug in the insterion code -
the unsigned properties were created as long.
|
|
This patch adds formatting changes to
better follow the style guidelines in
the collection unit test.
No logical changes to the code.
I was planning to do it for a while
per Simo's comment when he accepted
the core of the collection code but
indicated that the unit test should be
cleaned later.
Later has come.
|
|
Add was always insterting at the end of the collection.
With this change one can control where the item is
inserted and deal with the duplicates too.
Also one now can extract items from collection
using absolute and relative disposition.
Using more advanced hashing function.
|
|
If automake 1.11 or higher is available, use of the configure
option '--enable-silent-rules' will suppress most build messages
and replace them with lines such as
CC sysdb.o
This affects libtool as well.
|
|
LT_INIT is supported only on Libtool >= 2.0, so I reverted it to
using AC_PROG_LIBTOOL.
Also reorganized how the common libraries were being built. Now
they are treated as libtool convenience libraries instead of
installable libraries (the --with-singlelib configure flag can
still be used to generate a combined, installable DSO)
I cleaned up the set of files being installed by automake, so the
list of things we need to remove before packaging the RPM is now
only the .la files associated with our own plugins.
|
|
Ensures that the common libraries build statically.
Also ensures that the sssd.spec creates the SYSV init script with
the appropriate permissions.
Cleans up the useless rpmdist target that was obsolete.
Adds the *.so.N.0.0 files to the RPM, as the .so and .so.N files
were dangling symlinks.
|
|
|
|
|
|
|
|
Allow configuration and build of individual SSSD utility libraries
from their own subdirectories. Building all utilities is still
possible recursively from the common root.
|
|
The read_line() function used an internal buffer allocated on stack
as temporary storage for a line read from file, then returned it.
read_line() now gets a buffer from the caller.
Fixed memory leaks in INI and Collection found by valgrind.
|
|
We were missing several BuildRequires for the autotools. Also, we
were linking against two external libraries in the common code
that we do not actually use.
|
|
Allows building shared or static libraries using autotools and
provides a pkg-config file to simplify inclusion into other parts
of the project (or other projects in the future)
For now, we will statically link the collection library and INI
parser.
|
|
|
|
Added a few new functions.
Cleaned code that was subject to conditional build.
Fixed the floating point conversion.
Keep const values as const.
|
|
|
|
|
|
|
|
Adds ini subdirectory so it will be built, adds some clarification
to the README, makes the configure --help more clear about the
trace level and enables -Wall reporting.
|
|
|
