This is an implementation experiment to validate the improvements
and internal clean-up. Previously, I was unable to implement such a
forwarding proxy buffer provider, as would be required for the
intended harmonisation of the `OutputSlot` interface with `BufferProvider`
With the reworked verison of the BufferProvider standard implementation
it seems straight-forward to...
* define a »Buffer Type« that marks any of the prepared buffers to pass-through
* use the resulting Buffer Descriptor together with a rigged storage backend,
that just returns the ''local tag'' as buffer adress
The first round of the [ticket:1410 BufferProvider clean-up] is roughly complete.
* a much cleaner diagnostic interface has been defined and implemented
* this allowed to modernise the `DiagnosticBufferProvider_test`
and the `BufferProviderProtocol_test`
* the implementation scheme has been split in two realms
* the old heap-memory based implementation has been re-arranged,
so that the aforementioned tests are **green** again.
Now is the point to draft a solution to get the interfaces of
`BufferProvider` and `OutputSlot` aligned, so that the latter
can expose its lifecycle through the former.
Today, I have analysed various aspects and developed a pathway
to get into that task, and next I will reshape the previously
failed `OutputProxyProvider_test` into a prototype to
validate this idea
* write down some first considerations regarding requirements
of the actual production-strength implementation of buffer management
* consider relevance of the part of the API that was retained
* write a new series of test cases to cover handling of embedded objects,
using the test support for tracking object creation
Some weeks ago, at the time where I started to dig into this topic,
I extended the BufferProviderProtocol_test to perform a buffer usage pattern
that mimics the usage in the real Render Engine, as implemented last year.
For this, I used the `TestFrame` and the chained-hash dummy computations.
Now, after reordering and clean-up of the `BufferProvider` implementation interfaces,
it can be showed with this computation scheme that
- all buffers are indeed allocated and handed out as expected
- the metadata handling is correct and points to the processed buffers
- the complete memory used in this test is uncorruped and holds correct numbers
It covers now roughly the same procedures as before,
yet the scheme how to do verification is quite different.
Furthermore, we allocate large random-sized buffers now
and implant `TestFrame` instances, and so also the verification
covers a significant storage area.
* can now replace old verification code by new diagnostic functions
* test case runs without failure
''UNRESOLVED:'' failure in verification of memory contents
In the end I refrained from applying most of the changes I had intended.
I am aware that `BufferProvider` is still quite a massive structure,
yet with the implementation refactorings done thus far, it seems
to be reasonably clear and self-contained now.
It would be possible to totally overthrow the grown structure and
expose only the two handles (BuffDescr and BuffHandle), together with
a factory function, but I cannot see how this would help to make the
code clearer and more approachable. Maybe it would make the usage
more obvious from the client's perspective — yet, to be honest,
there are not ''that much clients'' to justify such a move.
Furthermore, the implementation is far from finished: for both of
the back-end lines, a fully production-strength implementation
must be created, that is able to operate in a highly concurrent
environment; very likely some lock-free messaging and hand-over
between threads will become necessary, adding to the complexity.
Thus it seems better to note the ''design critique'', but to
postpone any further refactorings to a point where we can watch
a fully developed Render Engine under production load.
These were defined in a confusing and implementation-oriented way.
Since we have now mostly abstracted away the metadata handling from the
`BufferProvider` front-end, a more succinct definition was chosen
* a ''Descriptor'' is valid when the lookup succeeds
* a ''Handle'' is likewise valid when the underlying
state-Key can be resolved, and the memory address is given
Some aspects regarding the state remain unresolved though.
Ideally we'd want a `BuffHandle` to be valid preciselc when
it ''is in valid state to be accessed''
Now able to remove some obsoleted part of the old implementation.
The existing implementation of the ''Buffer Provider Protocol'' is
quite old and dated meanwhile; it was more like an explorative prototype
and mixed up implementation and test code.
At that time, notably the assumption was that the core render engine
will be implemented in plain-C and will push around `void*` all over the place.
The `BufferProvider` was meant as an adaptor, while questions of typing
were added as an afterthought. Especially some functionality to use
the `BuffHandle` in the way of a smart-ptr are completely obsoleted
by the reworked design of the Render Engine (2024).
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.
...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
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....
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.
...this bug was spotted as the test failed reproducibly
when built with `-fstack-protector-strong` — which adds
additional "canary" markers to some kinds of problematic
stack storage; this seemed to have the effect that now
the second test uses exactly the same location as the
preceding test, and thus finds valid data in the newly
created work buffers.
There is no reason for using uninitialised storage in this
test (I can recall that I wanted to build a generic helper
and intended to use that as ''virtual overlay'' over existing
memory — but that usage never took place, and the `struct Buffer`
is not a general-purpose tool now, but only made for this
specific test. Thus memory can be easily zero-initialised.
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...)
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)
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.
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.
Now looking into largely obsolete library facilities...
Starting from `ScopedHolder`, I found a surprising problem with ''perfect forwarding....''
...which however turned out to be the result of ''sloppy programming'' on my side.
At that time, in 2017, seemingly I was under pressure to define a Session-Command,
which captures a Time-entity as »State Memento«. Which turned out to be impossible,
since the Time entities were conceived as being immutable -- a questionable design
decision (see #1261), which already generated quite some additional complexities.
In the course of this »exercise«, I could again clarify that the implementation
of perfect forwarding works as expected on modern compilers — confusion may arrise
sometimes due to ''copy elision'' (which the compiler is required to perform
since C++17, even when the elided constructor has observable side effects).
And it can be derailed when (as was the case here) a »grab everything« constructor
accidentally ends up generating a copy- or move-constructor for the container class
itself. This is an instance of the problem documented with #963 ...
.... and the best solution is to abide by the rule-of-five (and a half)
and to put that `ReplacableItem` to where it belongs...
Indeed — this change set is kind of sad.
Because I still admire the design of the GAVL library,
and would love to use it for processing of raw video.
However, up to now, we never got to the point of actually
doing so. For the future, I am not sure if there remains
room to rely on lib-GAVL, since FFmpeg roughly covers
a similar ground (and a lot beyond that). And providing
a plug-in for FFmpeg is unavoidable, practically speaking.
So I still retain the nominal dependency on lib-GAVL
in the Build system (since it is still packaged in Debian).
But it is pointless to rely on this library just for an
external type-def `gavl_time_t`. We owe much to this
inspiration, but it can be expected that we'll wrap
these raw time-values into a dedicated marker type
soon, and we certainly won't be exposing any C-style
interface for time calculations in future, since
we do not want anyone to side-step the Lumiera
time handling framework in favour of working
„just with plain numbers“
NOTE: lib-GAVL hompage has moved to Github:
https://github.com/bplaum/gavl
- remove obsolete configuration settings
- walk through all settings according to the documentation
https://www.doxygen.nl/manual/config.html
- now try to use the new feature to rely on Clang for C++ parsing
- walk through the doxygen-warnings.txt and fix some obvious misspellings
and structural problems in the documentation comments.
With Debian-Trixie, we are now using Doxygen 1.9.8 —
which produces massively better results in various fine points.
However, there are still problems with automatic cross links,
especially from implementation to the corresponding test classes.
- conversion from pointer to bool now counts as ''narrowing conversion''
- constructor names must not include template arguments (enforced with C++20)
- better use std::array for some dummy test code
Several further warnings are due to known obsoleted or questionable constructs
and were left as-is (e.g. for ScopedHolder) or just commented for later referral
Future C++ versions will no longer generate default copy operations
once any single one was defined explicitly. So the goal is to kind-of
''enforce the rule of five'' (if you define one, define them all).
However, sometimes one of these special operators must be defined for a different reason,
e.g. because it is defined as protected, yet should not be exposed on the public API.
In such cases, any other copy operation which still is valid in the default form
must be declared explicitly ''as defaulted''
Overall this seems to be quite an improvement --
and it highlights (again) some known instances of questionable design,
which are mostly obsoleted and require clean-up anyway, or (as in the case of the
Placements) indicate »placeholder code« where the actual solution still needs to be worked out
Based on the building blocks developed thus far,
it was possible to assemble a typical media processing chain
* two source nodes
* one of these passes data through a filter
* a mixer node on top to combine both chains
* time-based automation for processing parameters
As actual computation, hash-chaining on blocks of
reproducible random data was used, allowing to verify
for every data word that expected computations were
carried out, in the expected order.
Using basically the same topology as in the preceding test, which focused on connectivity. However, in this case we retrieve actual processing functions from the »Test-Rand« ontology in order to perform hash-chaining computations on full data blocks. And, in addition, a »Param Agent Node« is used.
While initially intended as introductory test, it meanwhile
focuses on intricate technical details on the level of
basic building blocks, notably the `FeedManifold`
Now I have added a simple end-to-end demonstration example
how a Render Node is built from scratch, leaving out all
technical details and all convenience front-ends like
the `NodeBuilder` — just one dummy port invoked directly.
NodeBase_test demonstrates the building blocks of a Render Node,
and verifies low-level mechanics of those building blocks, which
can be quite technical. At the top of this test however are some
very basic interactions, which serve as an introduction.
__Remark__: renamed the low-level technical dispatch-access
for the parameter-accessors in `TurnoutSystem` to be more obvious,
and added comment (I was confused myself how to use them properly)
This is a crucial feature, discovered only late, while building
an overall integration test: it is quite common for processing functionality
to require both a technical, and an artistic parametrisation. Obviously,
both are configured from quite different sources, and thus we need a way
to pre-configure ''some parameter values,'' while addressing other ones
later by an automation function. Probably there will be further similar
requirements, regarding the combination of automation and fixed
user-provided settings (but I'll leave that for later to settle).
On a technical level, wiring such independent sources of information
can be quite a challenging organisational problem — which however can be
decomposed using ''partial function closure'' (as building a value tuple
can be packaged into a builder function). Thus in the end I was able to
delegate a highly technical problem to an existing generic library function.
With these additions, all conceivable cases are basically addressed.
Take this as opportunity to investigate how the existing implementation
transports values into the Binder, where they will be stored as data fields.
Notably the mechanism of the `TupleConstructor` / `ElmMapper` indeed
''essentially requires'' to pass the initialisers ''by-reference'',
because otherwise there would be limitations on possible mappings.
This implies that not much can be done for ''perfect forwarding'' of initialisers,
but at least the `BindToArgument` can be simplified to take the value directly.
What emerges here, seems to be a generic helper to handle
partial closure of ''tuple-like'' data records. In any case,
this is highly technical meta-programming code and mandates
extraction into a separate header — simplifying `NodeBuilder`
...on top of the parameter-decorating functionality developed thus far.
The idea is to allow in the `NodeBuilder` to supply ''some parameters''
directly, while the remaining parameters will be drawn from automation.
Several years ago, I developed some helpers for partial function closure.
Unfortunately these utils are somewhat limited, and rely on some pre-C++11
constructs, yet seem to be usable for the task at hand, since parameters
are always expected as value objects by definition.
This changeset shows a working proof-of concept for left-closing a
parameter tuple with 5 elements; this turns out to surprisingly difficult
due to the full genericity of the acceptable parameter-aggregates...
seemingly the definition can not be much simplified,
since there is no way around handling several definition flavours
of the processing-functor distinctly.
However, the definitions can be rearranged to be clearer,
the resulting type of the `FeedPrototype` can be deduced from the
builder function, and more stringent assertions can be added
...the idea is to limit the scope of possible changes
and rather directly accept a functor to transform the parameters.
We need then to account for the possible flexibility in processing-functor
arguments, while in fact only two cases must be actually handled.
''This proof-of-concept works in test setup''
It seemed that the integration test will end up as a dull repetition
of already coded stuff, just with more ports and thus more boilerplate;
and so I reconsidered what an actually relevant integration test might encompass
- getting parameters from the invocation
- translating and wiring parameters
- which entails to adapt / partially close a processing function!
Thus — surprise — there is a new feature not yet supported by the `NodeBuilder`,
which would be very likely to be used in many real-world use cases: which is
to adapt the parameter tuple expected by the binding from the library.
Obviously we want this, since many »raw« processing functions will expose a mix
of technical and artistic parameters; and we'd like to ''close'' the technical ones.
Such a feature ''should be implementable,'' based on the already developed
technique with the »cross builder«, which implies to switch the template arguments
from within a builder expression. We already do this very thing for adapting
parameter functor, and thus the main difficulty would be to compose an
adaptor functor to the correct argument of the processing functor...
Which is... (well, it is nasty and technical, yet feasible).
Just wanted to use a helper function to build a source-data node.
However, the resulting node had a corrupted Node-ID spec.
Investigation with the debugger showed that the ID was still valid
while in construction and shows up corrupted after returning from the
helper function.
As it turned out, the reason is related to the de-duplication of ProcID data.
While the de-duplicated strings themselves are ''not'' affected, the corruption
happened by an intermediate instance of ProcID, which was inadvertently created
and bound by-value to the builder-λ. The created Port then picks up a reference
to this temporary, leading to the use-after-free of the string_view obejcts.
Obviously, `ProcID` must not be instantiated other than through the static
front-end `ProcID::describe`. Due to the private constructor, I can not make this
object non-copyable (because then the hash-set would not be allowed to emplace it).
But making it at least move-only will provoke a compiler error whenever binding
to a lambda capture by value, which hopefully helps to pinpoint this
insidious problem in the future...
The scheme to provide preconfigured nodes with random `TestFrame` data
seems to be suitable and easy to extend to further cases; should however
always document the setup through a dedicated case in `NodeDevel_test`
