* Various changes so that series of image and palette subsegments can have `auto` rom addresses (as long as the first can find its rom address from the parent segment or its own definition)
### 0.16.0
* Add option `detect_redundant_function_end`. It tries to detect redundant and unreferenced functions ends and merge them together.
* This option is ignored if the compiler is not set to IDO.
* This type of codegen is only affected by flags `-g`, `-g1` and `-g2`.
* This option can also be overriden per file.
* Disable `include_macro_inc` by default for IDO projects.
* Disable `asm_emit_size_directive` by default for SN64 projects.
*`spimdisasm` 1.16.0 or above is now required.
### 0.15.4
* Try to assign a segment to an user-declared symbol if the user declared the rom address.
* Helps to disambiguate symbols for same-address overlays.
### 0.15.3
* Disabled `asm_emit_size_directive` by default for IDO projects.
* Made some modifications such that linker object paths should be simpler in some circumstances
### 0.15.0
* New options:
*`data_string_encoding` can be set at the global level (or `str_encoding` at the segment level) to specify the encoding using when guessing and disassembling strings the the data section. In spimdisasm this value defaults to ASCII.
*`rodata_string_guesser_level` changes the behaviour of the rodata string guesser. A higher value means more agressive guessing, while 0 and negative means no guessing at all. Even if the guesser feature is disabled, symbols manually marked as strings in the symbol_addrs.txt file will still be disassembled as strings. In spimdisasm this value defaults to 1.
* level 0: Completely disable the guessing feature.
* level 1: The most conservative guessing level. Imposes the following restrictions:
* Do not try to guess if the user provided a type for the symbol.
* Do no try to guess if type information for the symbol can be inferred by other means.
* A string symbol must be referenced only once.
* Strings must not be empty.
* level 2: A string no longer needs to be referenced only once to be considered a possible string. This can happen because of a deduplication optimization.
* level 3: Empty strings are allowed.
* level 4: Symbols with autodetected type information but no user type information can still be guessed as strings.
*`data_string_guesser_level` is similar to `rodata_string_guesser_level`, but for the data section instead. In spimdisasm this value defaults to 2.
*`asm_emit_size_directive` toggles the size directived emitted by the disassembler. In spimdisasm this defaults to True.
* If enabled it tells the disassembler to try disassembling functions with unknown instructions instead of falling back to disassembling as raw data
### 0.13.9
* New segment option `linker_entry` (true by default).
* If disabled, this segment will not produce entries in the linker script.
### 0.13.8
* New option `segment_end_before_align`.
* If enabled, the end symbol for each segment will be placed before the alignment directive for the segment
### 0.13.7
* Severely sped-up linker entry writing by using a dict instead of a list. Symbol headers will no longer be in any specific order (which shouldn't matter, because they're headers).
### 0.13.6
* Changed CI image processing so that their data is fetched during the scan phase, supporting palettes that come before CI images.
### 0.13.5
* An error will be produced if a symbol is declared with an unknown type in the symbol_addrs file.
* The current list of known symbols is `'func', 'label', 'jtbl', 'jtbl_label', 's8', 'u8', 's16', 'u16', 's32', 'u32', 's64', 'u64', 'f32', 'f64', 'Vec3f', 'asciz', 'char*', 'char'`.
* Custom types are allowed if they start with a capital letter.
* Renamed `follows_vram_symbol` segment option to `vram_of_symbol` to more accurately reflect what it's used for - to set the segment's vram based on a symbol.
* Refactored the `appears_after_overlays_addr` feature so that expressions are written at the latest possible moment in the linker script. This fixes errors and warnings regarding forward references to later symbols.
* Added a new symbol_addrs attribute `appears_after_overlays_addr:0x1234` which will modify the linker script such that the symbol's address is equal to the value of the end of the longest overlay starting with address 0x1234. It achieves this by writing a series of sym = MAX(sym, seg_vram_END) statements into the linker script. For some games, it's feasible to manually create such statements, but for games with hundreds of overlays at the same address, this is very tedious and prone to error. The new attribute allows you to have peace of mind that the symbol will end up after all of these overlays.
* Added `ld_wildcard_sections` option (disabled by default), which adds a wildcard to the linker script for section linking. This can be helpful for modern GCC, which creates additional rodata sections such as ".rodata.xyz".
* Added `ld_use_follows` option (enabled by default), which, if disabled, makes splat ignore follows_vram and follows_symbols. This helps for fixing matching builds while being able to add infrastructure to the yaml for non-matching builds by just re-enabling the option.
### 0.13.0
* Automatically generate `INCLUDE_RODATA`/`#pragma GLOBAL_ASM` directives for non migrated rodata symbols when creating new C files.
* Non migrated rodata symbols will now only be produced if the C file has a corresponding rodata file with the same name and the C file has a `INCLUDE_RODATA`/`#pragma GLOBAL_ASM` directive referencing the symbol, similar to how functions are disassembled.
* Because of this, the `partial_migration` attribute has lost its purpose and has been removed.
* Rodata symbol files are now included in the autogenerated dependency files too.
* The gfx and vtx segments now have a `data_only` option, which, if enabled, will emit only the plain data for the type and omit the enclosing symbol definition. This mode is useful when you want to manually declare the symbol and then #include the extracted data within the declaration.
* The gfx segment has a method, `format_sym_name()`, which will allow custom overriding of the output of symbol names by extending the `gfx` segment. For example, this can be used to transform context-specific symbol names like mac_01_vtx into N(vtx), where N() is a macro that applies the current "namespace" to the symbol. Paper Mario plans to use this, so we can extract an asset once and then #include it in multiple places, while giving each inclusion unique symbol names for each component.
* A check was added to prevent segments marked with `exclusive_ram_id` have a vram address range which overlaps with segments not marked with said tag. If this happens it will be warned to the user.
* Fixed a bug involving the order of attributes in symbol_addrs preventing proper range searching during calls to `get_symbol`
### 0.12.3: Initial Gamecube Support
Initial support for Gamecube disk images has been set up! Disassembly is not currently supported, and a more comprehensive explanation of Gamecube support will come once that is finished.
* The Symbol class is now hashable
* Added the ability for segments to specify a file path (`path`) to receive that file's contents as their split input
* The `generated_s_preamble` option now will be applied to data files created by spimdisasm
* Rewrote symbol range check code to be more efficient
* Fixed bug that allowed empty top-level segments of type `code`.
* Fixed progress bars to properly update their descriptions
* Fixed bug pertaining to symbols getting assigned to segments they shouldn't if their segment is given in symbol_addrs (`segment:`)
* Fixed bug where `given_dir` was possibly not a `Path`
### 0.12.1
* The constructor for `Segment` takes far fewer arguments now, which will affect (and hopefully simplify) any custom segments that are implemented.
* The new option `string_encoding` can be set at the global or segment level and will influence the encoding for strings in rodata during disassembly. The default encoding used is EUC-JP, as it was previously.
## 0.12.0: Performance Boost
In this release, we bring many performance improvements, making splat dramatically faster. We have observed speedups of 10-20x, though your results may vary.
* Linker script `_romPos` alignment statements now take a form that is friendlier to different assemblers.
* Fixed the default value of `use_legacy_include_asm` to be what it was before 0.11.2
Additionally, any custom segments or code that needs to read options will have to accommodate the new API for doing so. Options are now fields of an object named `opts` within the existing `options` namespace. Because the options are fields, `get_` is no longer necessary. To give an example:
Before: `options.get_asm_path()`
After: `options.opts.asm_path`
The clean_up_path function in linker_entry.py now uses a cache, offering a small performance improvement during the linker script writing phase.
### 0.11.1
* The linker script now includes a `_SIZE` symbol for each segment.
* The new `create_asm_dependencies`, if enabled, will cause splat to create `.asmproc.d` files that can inform a build system which asm files a c file depends upon. If your build system is configured correctly, this can allow triggering a rebuild of a C file when its included asm files are modified.
* Splat no longer depends directly on pypng and now instead uses [n64img](https://github.com/decompals/n64img). Currently, all image behavior uses the exact same code. Eventually, n64img will be implemented in C and support rebuilding images as well.
Spimdisasm now handles data (data, rodata, bss) disassembly in splat! This includes a few changes in behavior:
* Rodata will be migrated to c files' asm function files when a .rodata subsegment is used that corresponds with an identically-named c file. Some symbols may not be automatically migrated to functions when it is not clear if they belong to the function itself (an example of which being const arrays). In this case, the `partial_migration` option can be enabled for the given .rodata subsegment and splat will create .s files for these unmigrated rodata symbols. These files can then be included in your c files, or you can go ahead and migrate these symbols to c and disable the `partial_migration` feature.
* BSS can now be disassembled as well, and the size of a code segment's bss section can be specified with the `bss_size` option. This option will tell splat how large the bss section is in bytes so BSS can properly be handled during disassembly. For bss subsegments, the rom address will of course not change, but the vram address should still be specified. This currently can only be done in the dict form of segment representation, rather than the list form.
Thanks again to [AngheloAlf](https://github.com/AngheloAlf) for adding this functionality and continuing to improve splat's disassembler.
Linker scripts splat produces are now capable of being shift-friendly. Rom addresses will automatically shift, and ram addresses will still be hard-coded unless the new segment option `follows_vram` is specified. The value of this option should be the name of a segment (a) that this segment (b) should follow in memory. If a grows or shrinks, b's start address will also do so to accommodate it.
The `enable_ld_alignment_hack` option and corresponding behavior has been removed. This proved to add too much complexity to the linker script generation code and was becoming quite a burden to keep dealing with. Apologies for any inconvenience this may cause. But trust me: in the long run, it's good you won't be depending on that madness.
### 0.9.5
* Changes have been made to the linker script such that it is more shiftable. Rather than setting the rom position to hard-coded addresses, it increments the position by the size of the previous segment. Some projects may experience some alignment-related issues after this change. If specified, the new segment option `align: n` will add an `ALIGN(n)` directive for that section's linker segment.
### 0.9.4
* A new linker script section is now automatically created when the .bss section begins, using NOLOAD as opposed to the previous hacky rom rewinding we were previously doing. Additionally, `ld_section_labels` now includes `.rodata` by default.
### 0.9.3
* Added `add_set_gp_64` option (true by default), which allows controlling whether to add ".set gp=64" to asm/hasm files
* Thanks to [AngheloAlf](https://github.com/AngheloAlf), we now have a much better MIPS disassembler in splat! spimdisasm has much better hi/lo matching, much lower ram usage, and plenty of other goodies.
We plan to roll this out in phases. Currently, it only handles actual code disassembly. Later on, we will probably migrate our current data assembly code to use spimdisasm as well.
**NOTICE**: This integration has been tested on a variety of games and configurations. However, with any giant change to the platform like this, there are bound to be things we didn't catch. Please be patient with us as we handle these remaining issues. Though from what we've seen already, the slight bugs one may come across are totally worth the much improved disassembly.
### gfx segment type
* A new `gfx` segment type is available, which creates a c file containing a disassembled display list according to the segment's start and end offsets. Thanks to [Glank](https://github.com/glankk) and [Tharo](https://github.com/thar0/) for their work on [libgfxd](https://github.com/glankk/libgfxd) and [pygfxd](https://github.com/thar0/pygfxd/), respectively, for helping make this a possibility in splat.
### API breaking changes
* Some `Segment()` arguments have changed, which may cause extensions to break. Please see the `__init__` function for `Segment` for more details.
### symbol_addrs.txt changes
* symbol_addrs now supports the `segment:` attribute, which allows specifying the symbol's top-level segment. This can be helpful for symbol resolution when overlays use overlapping vram ranges. See `exclusive_ram_id` below for more information.
The new `symbol_name_format` option allows specification of how symbols will be named. This can be set as a global option and also changed per-segment. `symbol_name_format_no_rom` is used when the symbol does not have a rom address (BSS).
`$ROM` - the rom address of the symbol, hex-formatted and padded to 6 characters (ABCF10, 000030, 123456) (note: only for `symbol_name_format`, usage in `symbol_name_format_no_rom` will cause an error)
The appropriate prefix string will still automatically be applied depending on the type of the symbol: `D_` for data, `jtbl_` for jump tables, and `func_` for functions. This functionality may be customizable in the future.
The `auto_all_sections` option now should be a list of section names (`[".data", ".rodata", ".bss"]` by default) indicating the sections that should be linked from .o files built from source files (.c or asm/hasm .s files), when no subsegment explicitly indicates linking this type of section.
For example, if any subsegment of a code segment is of segment type `data` or `.data`, the `.data` section from all `c`/`asm`/`hasm` subsegments will not be linked unless explicitly indicated with a relevant `.data` subsegment.
Previously, this option was a bool, and it enabled this feature for all sections specified in `section_order`. Now, the desired sections must be specified manually. The default value for this option retains previous behavior.
----
The new `mips_abi_float_regs` option allows for changing the format of float registers for MIPS disassembly. The default value does not change any prior behavior, but `o32` is heavily encouraged and may become the default option in the future. For more information, see this [great writeup](https://gist.github.com/EllipticEllipsis/27eef11205c7a59d8ea85632bc49224d).
----
The new `gfx_ucode` option allows for specifying the target for the graphics macro format, which is used in the gfx segment type. The default is `f3dex2`.
### Segment options changes
The new `exclusive_ram_id` segment option allows specifying an identifer that will prevent the segment from seeing any symbols from other segments with the same identifer. This is useful when multiple segments are mapped to the same vram address at runtime and should never be able to refer to each other's symbols. Setting all of these segments to have the same value for this option will prevent their symbols from clashing / meshing unexpectedly.
----
The `overlay` setting on segments has been removed. Please see `symbol_name_format` above for info on how to influence the names of symbols, which can be applied at the segment level as well as the global level.
* You can now use the option `section_order` to define the binary section order for your target binary. By default, this is `[".text", ".data", ".rodata", ".bss"]`. See options.py for more details
* Documented all options in options.py
* Support for SN64 games (thanks Wiseguy!)
* More consistent handling of paths (thanks Mkst!)
* Various other cleanup and fixes across the board
* Finally removed the dumb linker section alignment hack
* Added version number to output on execution
### 0.7.8
* Fixed a bug relating to a linker section alignment hack (thanks Wiseguy!)
* Fixed a bug in linker_entry.py's clean_up_path that should make this function more versatile (thanks Wiseguy!)
### 0.7.7
* Disassembly now reads the `size` property of a function in symbol_addrs.txt to disassemble `size / 4` number of instructions. Feel free to specify the size of your functions in your symbol_addrs file if splat's disassembly is chopping a function too short or making a function too long.
If you have a group segment with multiple c files and want splat to automatically create linker entries at a given position for each code object (c, asm, hasm) in the segment, you can use an `all_` type for that section. For example, you can add `[auto, all_bss]` as the last subsegment in a segment. This will direct splat to create a linker entry for each code object in the segment. This saves a lot of time when it comes to manually adding .bss subsegments for bss support, for example. The same thing can be done for data and rodata sections, but note this should probably be done later into a project when all data / rodata is migrated to c files, as the `all_` types lose the rom positioning information that's necessary for splat to do proper disassembly.
The `auto_all_sections` option, when set to true, will automatically add `all_` types into every group. This is only done for a section in a group if no other manual declarations for that section exist. For example, if you have 30 c files in a group and a .data later on for one of them, `auto_all_sections` will not interfere with your `.data` subsegment. If you remove this, however, splat will use `auto_all_sections` to implicitly `.data` subsegments for all of your code objects behind the scenes. This feature is again particualrly helpful for bss support, as it will create bss linker entries for every file in your project (assuming you don't have any manual .bss subsegments), which eliminates the need to create dummy .bss subsegments just for the sake of configuring the linker script.
* Significantly better performance, especially when using the cache feature (`--use-cache` CLI arg).
* BREAKING: Some cli args for splat have been renamed. Please consult the usage output (-h or no args) for more information.
*`--new` has been renamed to `--use-cache`
*`--modes` arg changes:
* Image modes have been combined into the `img` mode
* Code and ASM modes have been combined into the `code` mode
* BREAKING: The `name` attribute of a segment now should no longer be a subdirectory but rather a meaningful name for the segment which will be used as the name of the linker section. If your `name` was previously a directory, please change it into a `dir`.
* BREAKING: `subsections` has been renamed to `subsegments`
* New `dir` segment attribute specifies a subdirectory into which files will be saved. You can combine `dir` ("foo") with a subsegment name containing a subdirectory ("bar/out"), and the paths will be joined (foo/bar/out.c)
* If the `dir` attribute is specified but the `name` isn't, the `name` becomes `dir` with directory separation slashes replaced with underscores (foo/bar/baz -> foo_bar_baz)
* BREAKING: Many configuration options have been renamed. `_dir` options have been changed to the suffix `_path`.
* BREAKING: Assets (non-code, like `bin` and images) are now placed in the directory `asset_path` (defaults to `assets`).
* Linker symbol header generation. Set the `linker_symbol_header_path` option to use.
*`typedef u8[] Addr;` is recommended in your `common.h` header.
* You can now provide `auto` as the `start` attribute for a segment, e.g. `[auto, c, my_file]`. This causes the segment to not be extracted, but linked. This feature is intended for modding.
* Providing just a ROM address but no type or name for a segment is now valid anywhere in `segments` or `subsegments` rather than just at the end of the ROM. It specifies the end of the previous segment for types that need it (`palette`, `bin`, `Yay0`) and causes the linker to simply write padding until that address.
* The linker script file is left untouched if the contents have not changed since the previous split.
* You can now group together segments with `type: group` (similar to `code`). Note that any ASM or C segments must live under a `type: code` segment, not a basic `group`.
**Breaking Change**: The command line args to split.py have changed. Currently, only the config path is now a required argument to splat. The old `rom` and `outdir` parameters are now optional (`--rom`, `--outdir`). Now, you can add rom and out directory paths in the yaml.
The `out_dir` option specifies a directory relative to the config file. If your config file is in a subdirectory of the main repo, you can set `out_dir: ../`, for example.
The `target_path` option spcifies a path to the binary file to split, relative to the `out_dir`. If your `baserom.z64` is in the top-level of the repo, you can set `target_path: baserom.z64`, for example.
I've begun a refactor of the code "files" code, which makes everything cleaner and easier to extend.
There's also a new option, `create_new_c_files`, which disables the creation of nonexistent c files. This behavior is on by default, but if you want to disable it for any reason, you now have the option to do so.
I am also working on adding bss support as well. It should almost be all set, aside from the changes needed in the linker script.
This release adds a new `assets_dir` option in `splat.yaml`s that allows you to override the default `img`, `bin`, and other directories that segments output to.
Internally, there's a new Symbol class which stores information about a symbol and is stored in a couple places during disassembly. Many things should be improved, such as reconciling symbols within overlays, things being named functions vs data symbols, and more.
**Breaking change**: The format to symbol_addrs.txt has been updated. After specifying the name and address of a symbol (`symbol = addr;`), optional properties of symbols can be set via inline comment, space delimited, in any order. The properties are of the format `name:value`
*`type:` supports `func` mostly right now but will support `label` and `data` later on. Internally, `jtbl` is used as well, for jump tables. Splat uses type information during disassembly to disambiguate symbols with the same addresses.
*`rom:` is for the hex rom address of the symbol, beginning with `0x`. If available, this information is extremely valuable for use in disambiguating symbols.
*`size:` specifies the size of the symbol, which splat will use to generate offsets during disassembly. Uses the same format as `rom:`
