This commit adds two classes:
* ProgressTracker
* ProgressScope
The first is for users to implement, and to instantiate when they desire
to get informed about the overall progress.
The second is to be added to all functions that may take a considerable
amount of time, such that they can report back how far along they are.
These are much more convenient to use than the existing ProgressHandler.
ProgressScope is designed such that it only requires "local knowledge"
about upcoming and finished work. Scopes are nested and combined to
form the final global progress.
The active ProgressTracker is stored in a thread_local pointer.
This is a consicius decision since in assimp there is often no 'context'
passed through. The ProgressTracker may be needed anywhere, and it would
be tedious and a huge change to pass it through to every function.
Therefore, using a thread_local variable makes it accessible everywhere,
without a major interface change. Since assimmp is single-threaded,
but may be run in parallel on multiple threads, a thread_local is a
good trade-off, in my opinion.
This change only adds few uses of ProgressScope, to generally show how
it would be used. Also for our use cases these where the most pressing
places to add progress reporting, so this already covers loading from FBX
files pretty well.
This commit does not add or remove c’tors or d’tors, so it is *not* ABI-breaking.
If a c’tor/d’tor does nothing else than the default behavior, this commit replaces it with “= default”.
If an initializer list entry does nothing else than the default behavior, this commit removes it. First and foremost, remove default c’tor calls of base classes (always called by the compiler if no other base c’tor is explicitly called) and c’tor calls of members with complex types (e.g. “std::vector”).
In a few instances, user-defined copy c’tors / move c’tors / assignment operators / move assignment operators were replaced with “= default”, too. I only did this if I had a clear understanding of what’s going on.
BaseImporter::SearchFileHeaderForToken() expected a pointer to a non-const token list. This was probably an oversight, as nobody would realistically expect the function to change the list. Furthermore, it prevented token lists from being compiled to read-only memory, in some cases even causing the compiler to generate thread-safe initialization.
The list is now const and all callers declare their token lists static const, thus compiling them to read-only memory.
The search for a matching importer had a few issues, see #3791. There were two different mechanisms to determine whether an importer accepts a specific file extension:
1. `aiImporterDesc::mFileExtensions`, which was forwarded to the UI via `BaseImporter::GetExtensionList()`.
2. `BaseImporter::CanRead()` when called with `checkSig == false`, which determines whether to actually use that importer.
Both were redundant and got out of sync repeatedly. I removed 2. completely and replaced it with 1., thereby syncing UI/import and shortening all `BaseImporter::CanRead()` implementations.
Further bugfixes:
- fixed glTF2 importer throwing exceptions when checking whether it can load a file
- removed `BaseImporter::SimpleExtensionCheck()` because it is no longer used and had a bug with case sensitivity
Since the `checkSig` parameter in `BaseImporter::CanRead()` is now useless, it can be removed completely. I’m not sure if this would break ABI compatiblity, so I’ll submit it with a later pull request.
Kim Kulling
Kim Kulling
Jan Krassnigg
Kim Kulling
Krishty
Florian Born
Krishty
Malcolm Tyrrell
Joshua Hyatt