Since the BufferStore implementation component is now supposed
to handle only the memory allocation, a complete end-to end data flow
must be ensured, so that allocation / deallocation functions get all
the relevant information. Notably, the `LocalTag` is now used as an
implementation-defined conduit for further (opaque) marker data.
Since all data passed over this API are simple values (even `LocalTag`
is effectively a 64bit integer), we can now avoid relying on output
parameters or similar side-effect tricks.
This change largely resolves the problems with the storage API,
so that the implementation can now be split in two disjoint segments.
One operation, `mark_emitted()`, is still misplaced on that API, yet
must be retained until the refactoring on the metadata side is complete
(this latter refactoring can not be started sensibly, before the old
»tracking« implementation is cleaned up to care for allocation solely).
With this change, the existing implementation, which combines the
tracking functionality and the allocation, has been squeezed down
into the heap-memory implementation class, yet the tracking functionality
is already detached. The remaining code down there is a huge pile of mess,
that can hopefully be dismantled soon. But at least all of this code
passes the compiler again, and so it should be possible now to focus
on getting the `DiagnosticBufferProvider` up and running, thereby
providing a new clean re-implementation of the diagnostic functions.
This is an important step to prepare a split of the existing implementation APIs
into a type-management part and a memory-allocation part; generally speaking,
a memory allocator needs to know the allocation size, preferably also for
releasing memory. Up to now, this information could be retrieved locally,
with a metadata lookup through the hash-ID; now we ensure that the
size is always passed alongside through the implementation API.
Incidentally, it looks like we might be able to form a common allocation-descriptor,
to represent this internal API in a more concise way. Note further that
all these changes are considered not relevant for the performance,
since work buffers are large and media computation dominates.
...just enough to be able to write the assertions in the first test case
(..and the test case fails still, due to the pending work on sorting out
the implementation APIs between `BufferStage` und `BufferStore`)
To push ahead with the refactoring,
I will build the test/diagnostics API completely anew,
and attach it solely to an instance of `DiagnosticBufferProvider`.
Unfortunately this ''leads us into chainsaw surgery...''
* it uncovers a host of conceptual contradictions
* actually, the `HeapMemProvider` was more an implementation
of diagnostic tracking, than it was an actual `BufferProvider`
implementation (hence the old name: `TrackingHeapBlockProivder`)
* and this lured me into using these tracking functions directly,
while the `DiagnosticBufferProvider` remained in unfinished state.
So basically I have now **disabled** all verification from the Unit-tests.
I'll have to rebuild them from scratch, thereby reproducing the logic/intention
- write some documentation
- identify some add-on API that must be retracted
- consider further how to disentangle ''diagnostics'' and ''implementation''
More or less forcibly squeeze all the existing implementation
into the new basic structure with two separate »pillars«
**Breakages**
- existing code expected the `BufferStore` to transition to ''locked'' state
- existing code expected `provideLockedBuffer(typeID)` to tail-call the
`buildHandle()` function of the enclosing `BufferProvider` base implementation
- various diagnostic usages bypassed the official APIs and called directly
into the implementation. This is no longer possible; rather, `DiagnosticBufferProvider`
needs to provide additional diagnostic API to suit those purposes
As a consequence of preceding experiments..
* I'll stick to the structure of the BufferProvider front-end interface
* also the »code behind« will remain there; a purely abstract API has no merit here
* however, the ''setup'' of the actual implementation will move into a derived class
* this derived scope class, `BufferProviderSetup` will house two implementation classes
* these two, now named `Stage` and `Store` provide the base implementation
for the two "pillars" of the implementation
Admitted, this structure is now much more elaborate, but it's the only path forward
I'm currently able to come up in order to disentangle the existing implementation mess...
The outward interface has stood the test of time quite well.
But the back-side towards the implementation is a tangled mess,
presumably due to being developed from prototyping code: the
actual implementation is completely overgrown by diagnostics
and test instrumentation, since running in unit test setup
was the only actual usage of this code ever since.
My first attempt with decomposing into a decorator-chain failed.
Thus the new line of attack is to introduce two implementation
pillars right from top-level
- `BufferStage` : the state and metadata handling
- `BufferStore` : the storage backend
These are defined as nested interfaces, and the further refactoring
starts with subclassing everything, nested as it is.
The intention is to build the tracking functionality from scratch,
while wrapping the old implementation as delegate. Hopefully this allows
to reduce the complexity of the existing implementation step by step,
until I'm ''able to grasp what this implementation does''.
This old implementation works, but is highly problematic code.
The Git history shows that I had already problems to understand this code
while building the DiagnosticOutputSlot, one year later.
To my defense, in both cases (2011 and 2012), I had no clear idea yet
how the processing within the engine would work, and thus I considered
that Buffer Type registry a mere implementation detail, that can be
lumped into the »code behind«.
The prototype implementations of BuferProvider and OutputSlot are confusing;
in-depth analysis shows that this is related to intermingling use
of test instrumentation and features of a prototypical default.
To get anywhere with the existing code, it seems we'd need to split off
the test instrumentation (i.e. `DiagnosticBufferProvider`) altogether,
even if this causes duplication of implementation code. Many years ago,
when first implementing this prototype, I had no clear idea what the usage
in the Render Engine entails, yet I noticed that some basic buffer »tracking«
could be used as implementation primitive to simplify the prototyping work.
This might be one of the reasons, why the "code behind" in the protocol
implementations both in `BufferProvider` and `OutputSlot` became so strongly
tangled, so that I way unable to make both protocols orthogonally composable.
The implementation code for `DiagnosticOutputProvider` and `TrackingHeapBlockProvider`
look very confusing and tangled. It seems, instrumentation for tests
got mixed up with the effort to provide a simplistic prototype implementation.
What we need foremost here is the latter; most tests that import the
DiagnosticOutputProvider do not even use any instrumentation....
`EngineCtx` is considered merely an access front-end now;
After some analysis of existing code, it seems the RenderEnvironment
is the most appropriate framework to configure the global engine services
and thus to ''boot the Render Engine''.
The access to the `EngineCtx` is tricky, because of the requirements
* it will be dependency-injected into the internals of the Render Nodes
* excessive indirection should be avioded; preferrably the actual
code in the render pipeline sees a pointer (coded as language reference)
which directly refers to the buffer and cache manager.
* but on the other hand, we want configurability; simple tests shall
access "the buffer manager" without much ado, while the actual
render engine obviously has a life cycle and needs an administrative
access to the various global implementation facilities
* this both together implies some kind of ''virtual constructor''
* and since such a pattern can not be implemented in C++ with simple
inheritance (because the base constructor does not "see" the actual
implementation type at the time when it's invoked), the solution
is to rely on a second indirection through our dependency-injection
framework: thus the setup of the actual implementation is what
can be configured flexibly, and overridden by mock implementation
for some test cases
While this topic is only tangentially related to the problems
with the »Buffer Provider Protocol«, the configuration of service instances
is relevant, insofar it provides a configuration point for unit testing.
The structure of the `EngineCtx` was created as a placeholder, to get those
issues out of the way while drafting the render node code; in its current
form it is clearly both to complicated and inflexible.
My intention is to merge the `EngineCtx::Facilities`, and to prefer
configuration of more elaborate setup by subclasses.
It seems evident that the the problems with the existing BufferProvider implementation
are rooted in a lack of understanding of the actual usage situation at that time(2011).
This caused a »hexagonal design«, i.e. something entirely driven by uint testing
for the sake of unit testing.
So to identify the unsuitable decisions, some degree of deductive work is necessary,
based on our much better understanding of the Render Engine.
A crucial challenge is that buffer management must perform well
within a massively concurrent environment, and do so without excessive latency.
This picks up the work for [ticket:1395 resolve integration of BufferProvider with DataSink (#1395)]:
Almost a year ago, before the long hiatus due to important general clean-up
and maintenance work, and the research regarding Video output —
a discrepancy was discovered between the »Buffer Provider Protocol«
and the »Output Slot Protocol«. This mismatch prevents us from building
a simple preliminary `BufferProvider` to be used for the engine integration
as part of the [ticket:1228 »Playback Vertical Slice«].
After re-establisihing the work context, I have read through the new
Render Node code built last year, and distilled the sequence of
buffer management operations that actually happen during a recursive
`pull()` call. With this kowledge, I am now able to fill in the
''standard case for rendering'' — which I was unable to formulate
during the first attempt of building the Render Engine in 2011.
__Next Steps__: Redesign of the standard implementation base of the
`BufferProvider` to prepare for a massively concurrent usage.
Some recherche regarding the requirements and conditions of the GPL
made it clear that we are obliged to have at least an summary statement
of all authors and the years of their contributions embedded somewhere
in the source itself. The Git history alone is not sufficient to fulfil
this requirement, since a code base could be packaged as Tarball
and passed-on outside of any Git repository.
Thus I propose the following policy
- add yourself to the copyright header whenever you made
some ''significant'' contribution to a code file
- maintain a list of relevant authors at top-level,
indicating a time range of their contribution
- use the same information also in the debian/copyright (DEB package)
- clarify in the LICENSE that authorship can be extracted from Git
__Remark__: Anton Yakovlev made a single code contribution in 2009 (87e528bd),
which was never used anywhere and removed as part of a general clean-up in 2023.
Thus I do not list him as a code author, but he is mentioned in credits.txt
...funny enough I never noticed this obvious mistake,
since I never install software directly into my system,
while the DEB-build does not use absolute paths...
Solution:
* SCons has this speical convention that a path prefixed with '#'
is resolved relative to the root of the build (where the SConstruct resides)
* now we apply this automatically to the two relevant settings
** INSTALLDIR
** TARGDIR
* but only (conditionally) if the configured path is relative, not absolute
As a consequence, most other hard-coded usages of the '#'-prefix can then be dropped
Based on the additional insight gained through the FrOSCon talk,
an attempt was made to investigate and fix the garbled display of the
docking panel header directly above our VideoDisplayWidget.
However, that turned out to be a problem with insufficeint support
of the XVideo standard with my (very old, 2011) NVidia graphic card.
Indeed, the hardware+software stack needs to support this Auto-Keying
extension of XVideo, so that the video display can be integrated
seamlessly into the rest of the UI. In my case, the driver seemingly
just fills the complete XWindow with the key-marker and thus the
decorations of the docking panel, which happens to use the same XWindow,
will just not be flushed to the display.
This unsuccessful research indicates that upholding support for XVideo
could become problematic...
- reorganise the navigation tab structure
- place all further index lists into a common "Index" tab
- include a list of concepts
- remove the various "member" sub-tabs, these are not helpful
- the "modules" tab is now called "topics" (since C++ has now Modules!)
- generally re-sync DoxygenLayout.xml with a current pristine template
- comb though the Doxygen Warnings and fix a lot of small problems
The »Outer Space« and »Inner Core« documents present a comprehensive
overview of the vision, architecture and essential parts of the implementation.
In the light of changes to policy, design and implementation approaches,
some updates were necessary to align these crucial texts with the current
state of planning and implementation. Notably I have added a recent Screenshot
of the UI, showing a nested track structure pushed up by Diff from the core.
Over the last weeks, I conducted an extended analysis of the inception time
of the Lumiera project, which gave me a sharpened understanding of the goals,
going forward. I have also written an in-depth essay "Complexity and Flexibility".
See the corresponding commits in the Website repository.
The RfC documents were written to complement discussions of the Lumiera developers;
yet since the time where ''Ichthyo'' is working basically alone on the project,
this kind of discussions have ceased. During the following years, some ideas
promoted by the existing RfC documents became rather detached from the
actual state of development in the code base.
Many of the existing RfC documents require some commentary to place them
into context, and some of the decisions taken in the early stage of the
project should be **re-assessed**. This includes the decision to reject
some proposals, which initially might have seemed desirable, yet could not
be reconciled with the understanding of the matter and topic in question,
as was gained through the ongoing analysis and development.
Notably some links on the **Gnome documentation** pages are problematic,
since there was seemingly an infrastructure change -- which unfortunately
leads to loosing several deep-links into the GTK-3 related documentation
and tutorials.
While the differences to GTK-4 are often rather minimal, I would still prefer
to link to those documentation pages used as reference at implementation time
of our related library and GUI functionality.
In several cases I have thus looked up the old URLs at Archive.org
The topic of this talk was ''Video output from a Linux desktop application''
after extended research we built a demo application in four flavours
(XVideo, SDL, OpenGL legacy / modern).
This research project was set off by our immediate needs for a setup
to show computed video pixel data in a viewer window in the GUI; the
investigated technologies should work at present and in the near term
future (yet leaving out Vulkan and Wayland)
Yet another chainsaw massacre.
One of the most obnoxious annoyances with C++ metaprogramming
is the need to insert `typename` and `template` qualifiers into
most definitions, to help the compiler to cope with the syntax,
which is not context-free.
The recent standards adds several clarifications, so that most
of these qualifiers are redundant now, at least at places where
it is unambiguously clear that only a type can be given.
GCC already supports most of these relaxing rules
(Clang unfortunately lags way behind with support of newer language features...)
`lib::Result` can invoke, capture the result and thereby
represent ''either'' a result or a failure.
The old implementation required a delegate, due to the complexities
of integrating the `void` case. With C++23, `invoke_r` from the Stdlib
handles those issues, allowing a cleaner formulation, with directly
capturing the result into `lib::ItemWrapper`
Now able to remove most complicated comparison operators and most usages of boost::operators...
In most cases it is sufficient just to define one ''spaceship operator'',
and often even that one can be synthesised.
However — we still use boost::operators for arithmetic types,
notably the `lib::time::TimeValue`, which is addable and mutipliable
Only minor rearrangements necessary to make that possible with C++20
And while at this change (which requires a full rebuild of Lumiera)
- simplify the defined comparison operators, as C++20 can infer most variations
- also mark various usages of `const char*` either as Literal or CStr
Remark: regarding copyright, up to now this is entirely my work,
with two major creation steps in 2008 (conception) and
in 2017 (introduction of a symbol table)
This changeset removes various heuristics and marker-traits
by a constraint to tuple_like types. Furthermore, several usages
of `apply` can thereby be generalised to work on any tuple_like.
This generalisation is essential for the passing generic data blocks
via `FeedManifold` into the node invocation
- integrate the concept definition into tuple-helper.hpp
- use it to replace the `is_Structured` traits check
- do not need `enable_if_TupleProtocol` any more
Integrate test coverage of the concept metafunctions
and the generalised get accessor
''This changeset was made at LAC 2025 in Lyon, France''
...to rely on the new formulation and the extended template `WithIdxSeq`
This is in preparation to use this new iteration scheme also from the tuple_like concept
Motivated by the difficulties encountered with `std::apply` —
which basically forced us to define our own alternative with
conceptually more adequate limitations....
...so these are the first attempts towards building a C++20 concept.
Compilation failure with GCC-14.2 with the following code
class Base
{
protected:
Base() = default;
};
struct Feed
: Base
{ };
int
main (int, char**)
{
Feed f1;
// Feed f2{}; /// does not compile with GCC 14.2
return 0;
}
In the actual code base this can be triggered when instantiating
classes with the `NonCopyable`-mix-in; seemingly the compiler attempts
to invoke the base class ctor directly, while it should invoke a
(synthesised) default ctor for the derived class.
The problem could not be reproduced with other compiler versions at Godbolt.org
The Lumiera »Reference Platform« is now upgraded to Debian/Buster, which provides GCC-14 and Clang-20.
Thus the compiler support for C++20 language features seems solid enough, and C++23,
while still in ''experimental stage'' can be seen as a complement and addendum.
This changeset
* upgrades the compile switches for the build system
* provides all the necessary adjustments to keep the code base compilable
Notable changes:
* λ-capture by value now requires explicit qualification how to handle `this`
* comparison operators are now handled transparently by the core language,
largely obsoleting boost::operators. This change incurs several changes
to implicit handling rules and causes lots of ambiguities — which typically
pinpoint some long standing design issues, especially related to MObjects
and the ''time entities''. Most tweaks done here can be ''considered preliminary''
* unfortunately the upgraded standard ''fails'' to handle **tuple-like** entities
in a satisfactory way — rather an ''exposition-only'' concept is introduced,
which applies solely to some containers from the STL, thereby breaking some
very crucial code in the render entities, which was built upon the notion of
''tuple-like'' entities and the ''tuple protocol''. The solution is to
abandon the STL in this respect and **provide an alternative implementation**
of the `apply` function and related elements.
This resolves an intricate problem related to metaprogramming with
variadic templates and function signatures. Due to exceptional complexity,
a direct solution was blocked for several years, and required a better
organisation of the support code involved; several workarounds were
developed, gradually leading to a transition path, which could now
be completed in an focused clean-up effort over the last week.
Metaprogramming with sequences of types is organised into three layers:
- simple tasks can be solved with the standard facilities of the language,
using pattern match with variadic template specialisations
- the ''type-sequence'' construct `Types<T...>` takes the centre stage
for the explicit definition of collections of types; it can be re-bound
to other variadic templates and supports simple direct manipulation
- for more elaborate and advanced processing tasks, a ''Loki-style type list''
can be obtained from a type-sequence, allowing to perform recursive
list processing task with a technique similar to LISP.
after all the relevant library components do support both kinds of
type sequences transparently, any usages in core code can now be
switched over to the new, variadic type sequences.
A very performance relevant shortcoming of the existing implementation
of partial function closure is that the result is always wrapped into a
std::function, which typically causes a heap allocation when more than
a single pre-bound argument must be stored — which is annoying,
since the underlying Binder provides inline storage and thus
could be handled directly as a value object.
However, returning the Binder directly is also problematic, since
this object is outfitted with several overloaded function call operators,
which defeats most techniques to detect a function signature. Notably,
relevant down-stream metaprogramming code, like the tuple-closure used
in the `NodeBuilder` would break when being confronted directly with
a binder object.
An investigation shows that there is no direct remedy, short of
wrapping the binder into another functor. This can be accomplished
with a helper template, that generates a wrapper; however, this
wrapper builder must be supplied with explicit type information
regarding the function arguments (precisely because this type
signature can not be picked up from the Binder object itself)
This is a rather intricate and technical change, but allows in the end
to switch back all usages to a main implementation patch, which is now
based on `func::BindToArgument` — so this could become the final
implementation core and replace the old `PApply` template eventually...
Largely, these changes are related to allow for ''perfect forwarding''
of the functor and the argument values to be closed; these will be
copied into the ''Binder object'' created by `std::bind`.
Notably the `TupleConstructor` was changed to perfect-forward its »source«
into the specialised `ElmMapper`; this is possible since the latter
already receives a `SRC` template parameter, which can be supplied
with whatever base type the `std::forward` invocation will expose.
In the specialisation relevant here, template `PartiallyInitTuple`,
now an ''universal reference'' is stored and passed to `std::get`,
so that (depending on the input used), either a LValue or an
RValue reference is used for the extracted data elements.
After these changes, all existing usages of `applyFirst()` or `applyLast()`
can be replaced by this modernised implementation back-end, thus obsoleting
the various hard-coded workaround added during the last years.
A lot of repetitive, pre C++11 metaprogramming code can now be removed,
and several helper constructs, which were needed to handle generic
function application and passing a tuple of values to create a binder.
Note however, the highly complex and technical core of this header
still remains intact; which is to create a ''partial closure'' over
some arguments of a function, while keeping the remaining arguments
open as parameters for invocation.
TODO: Even in the remaining code there is a lot of redundancy
and helper construct which are no longer necessary
...because swapping in the new standards-based implementation
leads to compile failures on tests to cover out-of-bounds cases.
Under the (wrong) assumption, that some mistake must be hidden in
the Splice-metafunction, I first provided a complete test coverage;
while the actual problem was right below my nose, and quite obvious...
The old implementation, being based on a case distinction over the argument count,
simply was not able even to notice excess arguments; other the new implementation,
based on variadics and `std::apply`, which is fully generic and thus
passes excess arguments to `std::bind` when a position beyond the actual
argument list is specified to be closed.
The old behaviour was to silently ignore such an out-of-bounds spec,
and this can be reinstated by explicitly capping the prepared tuple
of binders and actual arguments passed to `std::bind`
Another question of course is, if being tolerant here is a good idea.
And beyond that, function-closure.hpp is still terrifyingly complex,
unorganised and use-case driven, to start with....
...can now be formulated in a single function,
based on the apply-to-λ technique invented by David Vandervoorde.
WARNING: the rewritten version of BindToArgument<...>::reduced()
does not compile in the out-of-bounds case, revealing a possibly
long standing defect in the typelist-metafunction Splice
This library header was developed at a time, where C++ had no built-in support
for so called "invokables"; `std::invoke` and `std::apply` were added much later;
So in that early version that was a significant technical hurdle to overcome.
seems like it might be possible to get rid of the TupleApplicator alltogether?
This is one of the most problematic headers, because it is highly complex
and comprises tightly interwoven definitions (in functional programming style),
which in turn are used deep within other features.
What concerns me is that this header is very much tangled
and pushes me (as the author) to my mental limits.
And on top of this comes that this code has to deal with intricate aspects
like perfect forwarding, and proper handling of binder instances and
function argument copying (which basically should be left to `std::bind`)
Fortunately, the changes ''for this specific topic'' are transparent:
Type sequences are not used on the API for function closure and composition,
but only as an internal tool to assemble argument tuples used for either
binding or invocation of the resulting (partially closed) function.
To bootstrap this tricky refactoring, initially a bridge definition
was used, with a variadic argument pack, but delegating to the old
non-variadic type sequence and from there further into LISP style
list processing of types and meta definitions, as pioneered by the
Loki libarary. Luimiera uses this technique since a long time to
perform the complex tasks sometimes required for code generation
and generic function and type adaptation.
with this changeset, a direct variadics based entrance into
type list processing is provided, so that the old definition
is now completely separate and can be removed eventually.
Most of the type-list and type-sequence related eccosystem can be
just switched over, after having added the conversion variants for
the new-style variadic type sequences
Again this was used as opportunity to improve readability of related tests
As expected, these work on the new-style variadic type sequences
equally well than on the old ones (tail-filled with `Nil` markers).
On that occasion, a complete makeover of the huge test case was carried out,
now relying on `ExpectString` instead of printing to STDOUT. This has the
benefit of showing the expectation immediately next to the code to be tested,
and thus makes it much easier to ''actually see'' how these meta-functions
operate on their parameters (which in fact are types in a type list)
- provide complete conversion paths old-style ⟷ new-style
- switch the basic tests to the new variadic sequences
- modernise the code; replace typedefs by `using`
- change some struct-style ''meta-functions'' into
constexpr or compile-time constants
Since I've convinced myself during the last years that this kind
of typelist programming is ''not a workaround'' — it is even
superior to pattern matching on variadics for certain kinds
of tasks — the empty struct defined as `NullType` got into
more widespread use as a marker type in the Lumiera code base.
It seems adequate though to give it a much more evocative name
Attempting to reduce the remaining pre-C++11 workarounds before upgrade to C++20...
As a first step: rename the old type-sequence implementation into `TyOLD`
to make it clearly distinguishable; a new variadic implementation `TySeq`
was already introduced as partial workaround, and the next steps
will be to switch over essential parts of the type-sequence library.
