LUMIERA.clone/doc/devel/rfc/RefactorLiblumieraOut.txt

Refactor Liblumiera Out
=======================

// please don't remove the //word: comments

[options="autowidth"]
|====================================
|*State*       | _Dropped_
|*Date*        | _Fr 22 Apr 2011 10:46:50 CEST_
|*Proposed by* | Christian Thaeter <ct@pipapo.org>
|====================================

********************************************************************************
.Abstract
`liblumiera` contains a lot of useful and re-useable code
which is already in use by other projects
********************************************************************************

Description
-----------
//description: add a detailed description:
Over the time we've put some efforts into the 'liblumiera'. I have added
some from my code which predates the Lumiera project and which I am using
on many other projects. This means that I would have to maintain these
sources in different unrelated projects and have to cross merge and update
several copies of the same code when I do updates and fixes somewhere.
I think its time to factor the re-useable parts out into a independent
library (similar to what Glib does for GTK). In fact, I had this plan long ago.


.What parts are eligible for a standalone library
Anything which is something tool alike and useful for other projects and not
tied to Lumiera only. This are the algorithms/datastructures, allocators, tool
macros. Additionally some of the src/common things should be moved into the
library. I give some lists below.

.How to name it
Long time ago my plan was to name it `ctlib' or `cehlib' but meanwhile there is
enough code done by others. So I'd propose a more neutral name, still
`lumieralib' or `lulib' would be appropriate. The only thing we have to account
for is that some parts which are too specific for Lumiera and should not be
integrated into this spinoff need either to stay in a lumiera-internal lib
(src/lib/) as currently or being moved to the respective subsystems using them
(src/backend, src/proc, src/common, ...), so the names should not clash.

.C, C++ ...
For myself I need the C parts, while there is C\++ code which interfaces to the
C implementations and also a lot code which does nice C\++ things on its own.
This possibly means that we should in fact make 2 packages out of this, one C
and one C\++ library (where the C++ part is more than just the wrappers, but
also the tools and tricks which are currently in 'src/lib/' and reusable).

.Who maintains it
Despite of being a spin-off I think we don't want to change anything from our
current practice and maintain it by the Lumiera developers. In part I feel
responsible for it, while it is really a part of the Lumiera codebase,
despite of being independently usable.

.How to maintain it
We need to decide about build system and documentation system. As build system
we may right start using SCons. For documentation the situation is somewhat
different since some of my code uses pipadoc/asciidoc and other uses doxygen.

.What not to do
Some of the code is currently quite specific to Lumiera while it could be made
more generic. This is _NOT_ subject of this RfC -- we may or may not do such a
refactoring but this RfC and any work resulting from this should be restricted
to simple things like necessary namespace and variable renaming and integration
in the build system.


C Parts
~~~~~~~

Library
^^^^^^^
What belongs to the library

Containers
++++++++++
 * cuckoo hashing (cuckoo.c|h)
 * linked lists (llist.h)
 * cache lists (mrucache.c|h)
 * splay trees (psplay.c|h)
 * priority queues (not done yet)

Runtime tools
+++++++++++++
 * error handling (error.h error.c) used by the other facilities too
 * clib convenience wrappers (safeclib.c|h) needs better name, maybe refactor
   into new facilities

Multithreading
++++++++++++++
 * locking, condition variables etc. (condition.c|h (rec)mutex.c|h, rwlock ...)

Memory management
+++++++++++++++++
 * Memory pools (mpool.c|h)
 * Temporary buffers (tmpbuf.c|h)

Metaprogramming
+++++++++++++++
  * preprocessor tools (ppmpl.h) move common preprocessor macros here
  * polymorphic call helper for C (vcall.h)

Interface system and module loader
++++++++++++++++++++++++++++++++++

except for some hardcoded references to `lumiera_org' and `.lum' plugin names
this is quite generic, possibly moving this over could be postponed, but might
eventually be done.

From 'src/common'
------
interface.c interfacedescriptor.h interface.h interfaceproxy.cpp
interfaceregistry.c interfaceregistry.h plugin.c plugin_dynlib.c plugin.h
------


The ``config system'' could become a candidate too if it ever gets finished and
proves itself useful, but for the time being it's better kept in Lumiera.


Not Library
^^^^^^^^^^^
Tied specifically to Lumiera:
-----
luid.c luid.h time.h
-----


C++ Parts
~~~~~~~~~

For most of the C++ parts I am not sure, _ichthyo_ should decide upon these
(please edit this here)

Library
^^^^^^^
These look 'generic' or wrap the C parts:
------
singleton-factory.hpp singleton.hpp singleton-policies.hpp
singleton-preconfigure.hpp singleton-ref.hpp singleton-subclass.hpp
sync-classlock.hpp sync.cpp sync.hpp thread-local.hpp
typed-allocation-manager.hpp typed-counter.hpp util.cpp util-foreach.hpp
util.hpp variant.hpp
------

Not Sure
^^^^^^^^
------
access-casted.hpp advice advice.hpp allocation-cluster.cpp
allocation-cluster.hpp bool-checkable.hpp cmdline.cpp cmdline.hpp del-stash.hpp
diagnostic-context.hpp element-tracker.hpp error.hpp (currently too
lumiera specific) exception.cpp (as before) factory.hpp format.hpp
frameid.hpp functor-util.hpp handle.hpp hash-indexed.hpp iter-adapter.hpp
iter-adapter-stl.hpp iter-source.hpp itertools.hpp iter-type-binding.hpp
lifecycle.cpp lifecycleregistry.hpp lumitime-fmt.hpp lumitime.hpp
multifact-arg.hpp multifact.hpp meta/* null-value.hpp observable-list.hpp
opaque-holder.hpp optional-ref.hpp p.hpp query.cpp query.hpp ref-array.hpp
ref-array-impl.hpp result.hpp scoped-holder.hpp scoped-holder-transfer.hpp
scoped-ptrvect.hpp searchpath.cpp searchpath.hpp sub-id.hpp symbol.hpp
symbol-impl.cpp visitor-dispatcher.hpp visitor.hpp visitor-policies.hpp
wrapper.hpp wrapperptr.hpp appstate.cpp appstate.hpp basic-setup.cpp
basic-setup.hpp DIR_INFO external guifacade.cpp instancehandle.hpp option.cpp
option.hpp query subsys.cpp subsys.hpp subsystem-runner.hpp
----


Not Library
^^^^^^^^^^^
------
logging.cpp nobug-init.cpp nobug-init.hpp streamtype.cpp streamtype.hpp test/*
time/* time.cpp tree.hpp
-----



Tasks
~~~~~
// List what needs to be done to implement this Proposal:
// * first step ([green]#✔ done#)

 * decide on name, namespaces [,yellow]#WIP#
 * create git repository, setup boilerplate (build system, legalise)
   [,yellow]#WIP#
 * move all code over into the git repos, refactor (namespace renames) ()
   [,yellow]#WIP#
 * make Lumiera use the new lib [,yellow]#WIP#

Discussion
~~~~~~~~~~

Pros
^^^^
// add a fact list/enumeration which make this suitable:

 * I am already reuse much of its code, making it independent makes maintaining
   it less burden


Cons
^^^^
// fact list of the known/considered bad implications:

 * new packages, new dependencies for Lumiera instead 'batteries included'

Alternatives
^^^^^^^^^^^^
//alternatives: explain alternatives and tell why they are not viable:
Do nothing and handle fixes on a case by case base.




//Conclusion
//----------
//conclusion: When approved (this proposal becomes a Final)
//            write some conclusions about its process:




Comments
--------
//comments: append below

.State -> Dropped
Ongoing development over the following years (after 2011) helped to clarify the situation
regarding the support library in Lumiera. The improved language standards, most notably C++14,
obsoleted most of the ad-hoc solutions in our library, since many aspects of system
integration are now handled by language and STL facilities in a satisfactory way.
On the other hand, by using the heightened capabilities of the language, some solution
patterns very specific to Lumiera could be established, gradually turning the support
library into some kind of an application framework.

It is quite common in the field of software development to put great hopes into code
reuse -- yet it is this very tendency towards a framework which prevents that ideal
from playing out as anticipated, in practice. What seems to work best is to look
at some well-made piece of software, understand how it works, and then to copy
and extract parts in order to transform it into _your own solution._

Ichthyostega:: '2025-09-16'

//endof_comments:


''''
Back to link:/x/DesignProcess.html[Lumiera Design Process overview]