This way, after o1 += o2, when o1 is linked all the outer data of o2 is
linked alongside the outer data of o1, and the inner data remains
contiguous. This is important for arrays, where we don't want the outer
data of o2 to appear before the next inner field of o1.
With this commit, an update to o2 after o1 += o2 no longer updates o1.
This wasn't a feature in the first place.
* alpha is now the last color of the palette rather than always being 0.
* alpha is not materialized in the P8 palette.
* Fixed a bug where images with more than 32/256 colors being converted
in P4/P8 with transparency would use all colors for opaque pixels,
causing alpha to randomly land on a color index that is in use.
Set the custom type with "custom-type: your_type_name" (instead of using
--custom on the command-line).
The converter.py in the current directory is still auto-detected as a
custom converter, but that won't work with CMake (which builds with the
build folder as current directory). Instead, a new option --converters=
with a semicolon-separated of Python files has been added. Use
fxconv_declare_converters() from the fxSDK-provided Fxconv module to
declare them programatically in CMake.
This change adds a new way for fxconv to discover metadata for file
conversions. This complements the existing mechanism of passing
parameters on the command-line.
The new mechanism activates when fxconv is called without a type
argument. Type information and metadata are searched in an
fxconv-metadata.txt file in the same folder as the resource. The
metadata file lists parameters, with some additional flexibility enabled
by the use of wildcards.
This way of declaring will replace command-line argument passing, which
currently read parameters from the unreadable and not-so-maitainable
project.cfg file. Both the GNU make and CMake build systems should use
it in the future. The current way is still supported only for older
projects and one-shot conversions outside of projects.
ObjectData is a stream-like object that accepts bytes, bytearrays,
references to external variables and references to other bundled data to
create structures with pointers without having to write assembly.
Internally ObjectData unfolds into static data and an assembly
instruction. Existing assembly support remains fully compatible.
* Added an ObjectData interface to ease reference generation
* Ported topti to ObjectData (instead of assembly)
* Ported libimg_fx and libimg_cg to ObjectData (idem)
Custom conversions can be used by:
* Providing a converters.py in the main directory with a convert()
function
* Specifying --custom on the fxconv command-line or using a subfolder or
assets-{fx,cg} unused by standard fxconv
Indexed images don't have an alpha channel, so the detection of
transparent pixels in fxconv failed. This change forces a conversion of
indexed images to RGBA to properly handle this.
This commit changes the output structure of topti to a format that
supports arbitrary Unicode blocks, but still only accepts the fixed set
of charsets that was defined before.
This commit introduces the libimg image format, selected with the option
type:libimg-image. To avoid confusion with the bopti image format,
options -i and --image are now deprecated and should be replaced with
--bopti-image or type:bopti-image. The fxSDK Makefile has been updated
accordingly.
To support the construction of a structure that contains a pointer in
fxconv, an assembly-code feature has been added. The structure itself is
assembled with as and then linked with the data proper. This allows the
structure to reference the data by name and have the pointer calculated
by ld at link time.
