Compare commits

...

1 Commits

6 changed files with 84 additions and 5 deletions

View File

@@ -53,9 +53,9 @@ Progress
<!-- progress:start -->
| Decompiled | Matched | Linked | Data |
|:---:|:---:|:---:|:---:|
| 99.59% | 90.81% | 71.43% | 99.37% |
| 99.59% | 90.92% | 73.37% | 99.37% |
units 956/1027 complete, functions 12365/12563 matched.
units 960/1027 complete, functions 12368/12563 matched.
decompiled = code with a C/C++ implementation (objdiff fuzzy), matched = byte-exact code, linked = code actually linked into the DOL, data = byte-exact data
<!-- progress:end -->

View File

@@ -766,7 +766,7 @@ config.libs = [
]
),
IPLSection("channelSelect", [
Object(Equivalent, "scene/channelSelect/iplChannelSelect.cpp"),
Object(Matching, "scene/channelSelect/iplChannelSelect.cpp"),
Object(Matching, "scene/channelSelect/iplClock.cpp"),
Object(Equivalent, "scene/channelSelect/iplChannelObj.cpp"),
]

View File

@@ -266,6 +266,32 @@ namespace ipl {
};
#endif
#ifdef IPL_CHANNEL_SELECT_CPP
template <>
class HermiteIntp<f32> : public utility::FrameController {
public:
HermiteIntp() {}
virtual ~HermiteIntp() {}
void init(const f32& start, const f32& end, f32 maxFrame, f32 startTangent,
f32 endTangent, int playback = ANIM_TYPE_FORWARD, f32 speed = 1.0f) {
mStart = start;
mEnd = end;
utility::FrameController::init(playback, maxFrame, 0.0f, speed);
mStartTangent = startTangent;
mEndTangent = endTangent;
}
inline f32 get() const;
protected:
f32 mStart;
f32 mEnd;
f32 mStartTangent;
f32 mEndTangent;
};
#endif
#ifdef IPL_SD_CHANNEL_SELECT_CPP
} // namespace math
} // namespace ipl

View File

@@ -27,7 +27,11 @@ namespace ipl {
void init(int type, f32 maxFrame, f32 minFrame, f32 speed = 1.0f);
void initFrame();
#ifdef IPL_CHANNEL_SELECT_CPP
virtual inline void calc();
#else
virtual void calc(); // 0x0C
#endif
void setMaxFrame(f32 value) { mMaxFrame = value; }
f32 getMaxFrame() const { return mMaxFrame; }

View File

@@ -1,3 +1,5 @@
#define IPL_CHANNEL_SELECT_CPP
#include <decomp/ide.h>
#include "iplSceneUI.h"
@@ -43,8 +45,7 @@ namespace ipl {
return r;
}
template <>
f32 HermiteIntp<f32>::get() const {
inline f32 HermiteIntp<f32>::get() const {
f32 var_f27 = mFrame;
f32 var_f28 = 1.0f / mMaxFrame;
f32 r = (mStart * (1.0f + ((var_f28 * (var_f28 * (var_f28 * (var_f27 * (2.0f * var_f27 * var_f27))))) -

View File

@@ -0,0 +1,48 @@
# ChannelSelect native linking
Baseline: HEAD f7b1b9b0, Equivalent, DOL SHA1 26116613f624061ba99c8d1a299aaa6efa85670d. Unit code 25668/25668, data 2336/2336, functions 102/102. Existing untracked fresh1.attempts.md left alone.
1. Set only ChannelSelect to Matching to locate the first linked-byte divergence.
Matching SHA1 f045a95026c3ca34039d0d9064fad33832ce581a. First DOL mismatch at file 0x5ba, address 0x8133009a in __start; relocation points to data shifted by 0x40. nm and ELF comparison show the first text layout change at createChannelThumbnails, as recorded in the exact symbol table below. Extra weak Interporation<float> destructor is 0x40 bytes, plus its vtable; neither exists in the target symbols. .text grows 0x40. Pool identical, 98 strings.
/tmp/link1.baseline.elf: createChannelThumbnails__Q33ipl5scene13ChannelSelectFv 0x813abcc4 size 0x5c.
/tmp/link1.matching.main.elf: createChannelThumbnails__Q33ipl5scene13ChannelSelectFv 0x813abd04 size 0x5c.
/tmp/link1.matching.main.elf: __dt__Q33ipl4math16Interporation<f>Fv 0x813abcc4 size 0x40.
2. Model the scalar Hermite specialization as a direct FrameController subclass, with its destructor declared out of line, matching the existing scalar specialization used by ChannelTitle and the target ownership at 0x81363c10. Scope the new specialization to IPL_CHANNEL_SELECT_CPP. Preserve inline init and get calculations. This avoids instantiating a nonexistent Interporation<float> destructor or vtable.
Result: all 102 functions and data still 100%; other unit reports unchanged. DOL SHA1 acdfd504572753800958ecd275825fdc19bb92c1. Text now correct but .data is 32 bytes short: scalar table and base table reservations were removed. Real data/vtable pointers up through ChannelSelect table still identical.
3. Keep the original Hermite primary template and specialize only scalar Interporation as a novtable base, retaining its real start/end fields and methods. The target inlined constructor stores only the derived vtable; the extra base destructor had no target owner. Test whether suppressing base vtable use lets the linker discard this extra destructor while retaining ordinary compiler-emitted data reservations.
Result: SHA1 f7cdb8937b2638367aa0479f20a880cd063d06bb. Removing the scalar base vtable use strips the 64-byte extra destructor, but removes a 16-byte data reservation; createDiskLayout remains instruction exact.
4. Restore a direct scalar Hermite specialization, now with an inline empty destructor as in the target scalar destructor. Declare FrameController::calc inline only in this TU to test normal weak base-vtable emission. This preserves the ordinary scalar and FrameController vtables without defining or retaining any fake object. Target scalar construction only sets the derived table, destructor is empty; the target data has 32 bytes of discarded weak tables after the ChannelSelect vtable and before the GUI table reservations.
Result: DOL SHA1 26116613f624061ba99c8d1a299aaa6efa85670d with ChannelSelect Matching. Scalar and FrameController weak tables are compiler-emitted and deduplicated against their existing earlier owners; no Interporation<float> symbol remains. All 102 objdiff functions exact, code 25668/25668, data 2336/2336. Every other unit report is byte-for-byte equal as JSON to the initial live baseline. pool_diff identical, 98 strings; createDiskLayout 168/168 instructions, diffs 0; VEC3 Hermite destructor 16/16 instructions, diffs 0.
Validation: running the full clean gate, full build/43U/ok, and independent checks of all 102 function instruction streams. Inline FrameController declaration keeps the implementation in its existing owning TU and requests normal weak-vtable emission here. No forced sections, data padding, new assembly, label aliases, or force-active changes.
Final clean gate output:
```text
full build: ok
main.dol sha1: 26116613f624061ba99c8d1a299aaa6efa85670d
[src/scene/channelSelect/iplChannelSelect] pool: IDENTICAL
[src/scene/channelSelect/iplChannelSelect] objdiff: code 25668/25668 data 2336/2336 functions 102/102 fuzzy 100.0000 linked code 25668
[src/scene/channelSelect/iplChannelSelect] instruction-exact functions: 102/102
[src/scene/channelSelect/iplChannelSelect] section .data size 2144 match 100.0
[src/scene/channelSelect/iplChannelSelect] section .rodata size 16 match 100.0
[src/scene/channelSelect/iplChannelSelect] section .sbss size 8 match 100.0
[src/scene/channelSelect/iplChannelSelect] section .sdata size 80 match 100.0
[src/scene/channelSelect/iplChannelSelect] section .sdata2 size 88 match 100.0
[src/scene/channelSelect/iplChannelSelect] section .text size 25668 match 100.0
[src/scene/channelSelect/iplChannelSelect] baseline: code 25668/25668 data 2336 functions 102 fuzzy 100.0000
regressions vs baseline: 0
global matched_code_percent: 90.80962 -> 90.80962
global fuzzy_match_percent: 99.58569 -> 99.58569
global complete_code_percent: 72.50912 -> 73.36610
global matched_data_percent: 99.36639 -> 99.36639
forbidden patterns added (net, per file): 0
readability warnings (net, per file; must be 0 in the final result): 0
note: config touched: configure.py (orchestrator reviews every config/symbols change)
GATE PASS
```
Explicit ninja build/43U/ok passes. Regenerated live report: all other unit records unchanged from the initial snapshot. All 102 ctxdiff.py runs report identical instruction counts and diffs 0; no open function remains. DOL bytes identical to orig/43U/00000008.app, SHA1 26116613f624061ba99c8d1a299aaa6efa85670d. Functions 102/102 -> 102/102, code 25668/25668 -> 25668/25668, data 2336/2336 -> 2336/2336, linked code 0 -> 25668.