Skip to content

Latest commit

 

History

History
179 lines (138 loc) · 9.13 KB

CHANGELOG.md

File metadata and controls

179 lines (138 loc) · 9.13 KB
Raw

CHANGELOG

This document outlines changes and updates in major releases of vtkio.

Release 0.6

This release moves all IO API into the Vtk struct, which should make the documentation easier to browse and adds some consistency to the API (functions taking self export and static functions construct a Vtk file via import).

Additionally, API function names are now more consistent throughout the library (more specifically between parse and write functions).

  • parse_vtk_{be|le} is renamed to parse_legacy_{be|le}.
  • parse_vtk_buf_{be|le} is renamed to parse_legacy_buf_{be|le}.
  • import_{be|le} is renamed to import_legacy_{be|le}.

Support for compression and decompression (feature gated by the "compression" feature which is enabled by default) are also added.

  • LZMA, LZ4 and Zlib compression are now all supported for appended data blobs.
  • Compression level is currently ignored on LZ4 until either the lz4_flex crate implements support, or the lz4 crate supports LZ4 block format.
  • Note that solutions to the above problem should only cause a minor version bump.

A few new API changes have been made:

  • The VTK file was changed to include an optional file_path, which encodes the original path to the VTK file. This allows relative paths when reading in "parallel" XML files. This is how ParaView deals with "parallel" XML files for instance. Note that the "parallel" files refers to how they are defined in the VTK documentation; async file loading is not yet supported, but it is planned.
  • load_piece_data was renamed to into_loaded_piece_data to indicate that this function takes a piece by value and converts it into concrete piece data.
  • In contrast load_piece_in_place takes a Piece by mutable reference and replaces source pieces by their loaded counterparts.
  • load_all_pieces is added as a convenience function to load "parallel" XML files recursively in-place. For non-parallel files this function is a no-op.
  • The error and display traits are finally implemented on writer errors, which fixes #6.
  • Added try_into_xml_format function to model::Vtk, which accepts an additional arguments indicating the compression level. The TryFrom trait defaults to no compression as before.
  • The XML API is updated to accept an encoding info (which includes byte order and header type) when encoding/decoding data.

Release 0.5

This is a small update to v0.4 that simplifies the Polygon data topology fields. This release also fixes bugs in parsing and writing files in XML format, especially involving Polygon data.

In particular the following API was changed:

  • Polygon data topology (PolyDataTopology) was previously stored as a Vec of enums identifying the topology type (one of Vertices, Lines, Polygons, or Triangle Strips). This is removed in favour of storing the specific types of topologies in a dedicated field of the new PolyDataPiece struct. This makes it unambiguous that there is at most one of each topology sections in each piece.

  • A new parse and write API is introduced for easily reading and writing legacy and xml files from standard byte buffers (and string slices for ASCII files). Specifically the functions parse_vtk_{be|le}, parse_vtk_buf_{be|le}, parse_xml, write_legacy, write_legacy_ascii, and write_xml are added.

Release 0.4

This release most notably adds support for importing and exporting VTK files in the modern XML format.

The XML support is provided via an additional xml::VTKFile type, which stores xml specific information which can be directly serialized and deserialized with serde and quick-xml into the corresponding VTK XML file type.

This means that an additional pass is required to convert the XML type into the underlying model::Vtk type used for vtkio, which unifies legacy and xml I/O and facilitates the lazy loading of parallel XML formats. Performance sensitive applications may chose to work with the xml::VTKFile type directly, however the model::Vtk API is much more comprehensive and easier to work with. Notably, xml::VTKFile stores loaded data buffers as encoded and potentially compressed strings and byte arrays, whereas model::Vtk stores the decoded uncompressed data ready for processing.

In order to facilitate import and export for both legacy and xml formats, the underlying VTK data model (model::Vtk) was updated. The following outlines most of the critical changes that were made in version 0.4, particularly in the model module:

  • IOBuffer was simplified into a simple enum of Vecs of different predetermined scalar types. This means that a number of member functions are now unavailable (and many are unnecessary) on IOBuffers.

  • a byte_order field was added to the Vtk model to facilitate automatic determination byte order on import and export. This means that import and export functions will automatically encode data in the byte order provided in that field. Byte order can still be overridden with export_{le,be} or by changing the byte_order field in the Vtk file directly.

  • Attributes have been completely reformulated to distinguish between Field attributes which are available only in Legacy VTK formats and the rest. The distinction between the "kind" of attribute (ColorScalars, Scalars, Vectors, etc.) is offloaded to a separate enum called ElementType. Because XML files don't necessarily need a "kind" for each attribute, ElementType can take on a Generic variant, which can represent any number of components per element.

  • DataArrays as well as Attributes are equipped with convenience constructors for creating arrays with a specific type.

  • call_numeric_buffer_fn macro has been removed. A new macro that can tersely abstract over the type of the underlying vector is match_buf, which evaluates an expression with a binding to the underlying vector. As before this is useful for expressions whose return type is the same regardless of the type of the vector, otherwise the caller must match on the IOBuffer variants explicitly.

    For example, suppose we had a function

    fn display<T: std::fmt::Display>(buf: &IOBuffer) { ... }

    that printed the buffer in some way. Previously we would need to call this function as follows

    call_numeric_buffer_fn!( display<_>(&buf) or {} );

    With the current changes, we would need to rewrite the display function in terms of a Vec (or slice) instead as follows:

    fn display<T: std::fmt::Display>(slice: &[T]) { ... }

    and call it with match_buf like so:

    match_buf!(&buf, v => display(v.as_slice()));

    which translates to

    match &buf {
    IOBuffer::Bit(v) => display(v.as_slice()),
    IOBuffer::U8(v) => display(v.as_slice()),
    IOBuffer::I8(v) => display(v.as_slice()),
    IOBuffer::U16(v) => display(v.as_slice()),
    IOBuffer::I16(v) => display(v.as_slice()),
    IOBuffer::U32(v) => display(v.as_slice()),
    IOBuffer::I32(v) => display(v.as_slice()),
    IOBuffer::U64(v) => display(v.as_slice()),
    IOBuffer::I64(v) => display(v.as_slice()),
    IOBuffer::F32(v) => display(v.as_slice()),
    IOBuffer::F64(v) => display(v.as_slice()),
}

  • DataSet has been decomposed into Pieces to be compatible with the XML vtk format. Use the load_piece_data function to retrieve data corresponding to each piece. Each DataSet can now contain multiple pieces. Each piece can be stored either inline alongside the data set (as before) or they can be loaded lazily from other referenced vtk files as described in the parallel XML vtk format. Since each loaded piece is itself a DataSet, there are in total 3 variants of a Piece: Inline, Source and Loaded. Inline pieces contain the actual piece data. See the documentation for details.

  • An optional MetaData type is attached to a data set to store meta information about the pieces referenced within. This is only useful when the pieces are not loaded eagerly with the data set. The parallel XML formats provide the additional meta data, which allows users to set up the data structures without having to load the actual data.

  • DataType was renamed to ScalarType to be more consistent with modern VTK formats. The names of the variants were also renamed to resemble Rust numeric types rather than C/C++.

  • The StructuredPoints data set variant was renamed to ImageData to reflect modern vtk formats.

  • A new Extent type was introduced to abstract between Legacy and XML formats. This type describes the extent of a grid-like structure (i.e. ImageData, StructuredGrid and RectilinearGrid data set types) or piece. See the documentation for details.

  • The coordinates of the RectilinearGrid type were split into a separate struct containing the individual x, y and z coordinate data arrays.

  • Data arrays in the Vtk data structure can be stored in 3 different ways, two of which specialize the generic DataArrayBase struct and one is a plain old IOBuffer. Most data arrays can now be named, with the exception of Coordinates, Points and Cells arrays, since those are always unique.