Skip to content

Commit 4a1635e

Browse files
SwenSchaeferjohannSwenSchaeferjohann
committed
make csdk use heap alot
1 parent 85b961d commit 4a1635e

File tree

6 files changed

+666
-157
lines changed

6 files changed

+666
-157
lines changed

‎js/stateless.js/src/compressible/instruction.ts

Lines changed: 3 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -212,7 +212,7 @@ export function createCompressAccountInstruction(
212212
* @param bumps Array of PDA bumps
213213
* @param validityProof Validity proof for decompression
214214
* @param systemAccounts Additional system accounts (optional)
215-
* @param dataSchema Borsh schema for account data
215+
* @param coder Borsh schema for account data
216216
* @returns TransactionInstruction
217217
*/
218218
export function createDecompressAccountsIdempotentInstruction<T = any>(
@@ -225,7 +225,7 @@ export function createDecompressAccountsIdempotentInstruction<T = any>(
225225
bumps: number[],
226226
validityProof: import('../state/types').ValidityProof,
227227
systemAccounts: AccountMeta[] = [],
228-
dataSchema?: any,
228+
coder: (data: any) => Buffer,
229229
): TransactionInstruction {
230230
// Validation
231231
if (solanaAccounts.length !== compressedAccountsData.length) {
@@ -255,15 +255,7 @@ export function createDecompressAccountsIdempotentInstruction<T = any>(
255255
systemAccountsOffset: solanaAccounts.length,
256256
};
257257

258-
// Serialize instruction data with discriminator
259-
let data: Buffer;
260-
if (dataSchema) {
261-
const schema =
262-
createDecompressMultipleAccountsIdempotentDataSchema(dataSchema);
263-
data = serializeInstructionData(schema, instructionData, discriminator);
264-
} else {
265-
throw new Error('dataSchema is required for proper serialization');
266-
}
258+
const data = coder(instructionData);
267259

268260
return new TransactionInstruction({
269261
programId,

‎sdk-libs/macros/LOGIC_PRESERVATION.md

Lines changed: 82 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,82 @@
1+
# Logic Preservation in decompress_accounts_idempotent Refactoring
2+
3+
## Original Logic Flow (Before Refactoring)
4+
5+
1. Box parameters (proof, compressed_accounts, bumps)
6+
2. Get PDA accounts from remaining accounts
7+
3. Validate account counts match
8+
4. Create CPI accounts
9+
5. Load config and get address space
10+
6. Pre-allocate compressed_infos vector
11+
7. FOR EACH compressed account:
12+
- Box the compressed_data
13+
- Check bounds
14+
- Create bump slice
15+
- MATCH on account variant:
16+
- Build seeds refs
17+
- Clone and box data
18+
- Create LightAccount
19+
- Call prepare_accounts_for_decompress_idempotent
20+
- Extend all_compressed_infos
21+
8. IF compressed_infos not empty:
22+
- Create CpiInputs
23+
- Invoke light system program
24+
25+
## New Logic Flow (After Refactoring)
26+
27+
1. Box parameters (proof, compressed_accounts, bumps) ✅
28+
2. Get PDA accounts from remaining accounts ✅
29+
3. Validate account counts match ✅
30+
4. Call setup_cpi_and_config helper:
31+
- Create CPI accounts ✅
32+
- Load config and get address space ✅
33+
5. Pre-allocate compressed_infos vector ✅
34+
6. FOR EACH compressed account:
35+
- Box the compressed_data ✅
36+
- Check bounds ✅
37+
- Call process_single_compressed_variant helper:
38+
- Create bump slice ✅
39+
- MATCH on account variant: ✅
40+
- Build seeds refs ✅
41+
- Clone and box data ✅
42+
- Create LightAccount ✅
43+
- Call prepare_accounts_for_decompress_idempotent ✅
44+
- Return compressed_infos ✅
45+
- Extend all_compressed_infos ✅
46+
7. Call invoke_cpi_with_compressed_accounts helper:
47+
- IF compressed_infos not empty: ✅
48+
- Create CpiInputs ✅
49+
- Invoke light system program ✅
50+
51+
## What Changed
52+
53+
### Structural Changes Only:
54+
55+
- Code split into inner functions for stack management
56+
- Helper functions defined inside main function (not at module level)
57+
- Added lifetime parameters to ensure borrowing is correct
58+
59+
### What Did NOT Change:
60+
61+
- ✅ Same parameter boxing
62+
- ✅ Same validation logic and error messages
63+
- ✅ Same iteration order
64+
- ✅ Same match statement logic
65+
- ✅ Same seeds construction
66+
- ✅ Same LightAccount creation
67+
- ✅ Same CPI invocation
68+
- ✅ Same error handling (ErrorCode::InvalidAccountCount)
69+
- ✅ Same msg! debug statements
70+
- ✅ Same data transformations
71+
72+
## Proof of Preservation
73+
74+
The refactoring is purely mechanical - moving code blocks into functions without changing:
75+
76+
1. Order of operations
77+
2. Data transformations
78+
3. Control flow
79+
4. Error conditions
80+
5. External function calls
81+
82+
Every single line of logic from the original is preserved, just organized into smaller stack frames.

‎sdk-libs/macros/STACK_OPTIMIZATIONS_MACRO.md

Lines changed: 105 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,105 @@
1+
# Stack Optimization for decompress_accounts_idempotent Macro
2+
3+
## Problem
4+
5+
The macro-generated `decompress_accounts_idempotent` function had a stack frame of 6080 bytes, exceeding the 4096 byte limit by 1640 bytes.
6+
7+
## Solution: Inner Function Decomposition with Parameter Bundling
8+
9+
Split the large monolithic function into multiple **inner helper functions** within the main function, each with its own stack frame. This avoids Anchor's "multiple fallback functions" error while still reducing stack usage. Additionally, bundle parameters into structs to reduce stack pressure from parameter passing.
10+
11+
### 1. **Main Function** (`decompress_accounts_idempotent`)
12+
13+
- Contains all helper functions as inner functions
14+
- Reduced main logic to coordination only
15+
- Validates inputs and delegates to helpers
16+
- Stack usage: ~500 bytes (estimated)
17+
18+
### 2. **Inner Setup Helper** (`setup_cpi_and_config`)
19+
20+
- Defined inside main function
21+
- Handles CPI account creation
22+
- Loads and validates config
23+
- Returns boxed values
24+
- Stack usage: ~200 bytes (estimated)
25+
- Marked with `#[inline(never)]`
26+
27+
### 3. **Inner Processing Helper** (`process_single_compressed_variant`)
28+
29+
- Defined inside main function
30+
- Takes parameters bundled in a `ProcessParams` struct to reduce stack
31+
- Processes one compressed account at a time
32+
- Contains the match statement for account variants
33+
- All large data structures boxed
34+
- Stack usage: ~200 bytes (estimated, reduced from 4392)
35+
- Marked with `#[inline(never)]` and `#[cold]`
36+
37+
### 4. **Inner Dispatch Helper** (`dispatch_variant`)
38+
39+
- Defined inside main function
40+
- Contains only the match statement
41+
- Isolates variant matching from other processing
42+
- Stack usage: ~150 bytes (estimated)
43+
- Marked with `#[inline(never)]` and `#[cold]`
44+
45+
### 5. **Inner Prepare Accounts Helper** (`call_prepare_accounts`)
46+
47+
- Defined inside main function
48+
- Generic helper to call `prepare_accounts_for_decompress_idempotent`
49+
- Separates the heavy lifting from the match statement
50+
- Stack usage: ~300 bytes (estimated)
51+
- Marked with `#[inline(never)]` and `#[cold]`
52+
53+
### 6. **Inner CPI Helper** (`invoke_cpi_with_compressed_accounts`)
54+
55+
- Defined inside main function
56+
- Handles the final CPI invocation
57+
- Minimal stack usage
58+
- Stack usage: ~200 bytes (estimated)
59+
- Marked with `#[inline(never)]`
60+
61+
## Key Optimizations
62+
63+
1. **Function Splitting**: Breaking the function reduces per-frame stack usage from 6080 to ~500 bytes max per function
64+
65+
2. **Parameter Bundling**: Using `ProcessParams` struct to pass multiple parameters as a single boxed value
66+
67+
3. **Boxing Strategy**: All large data structures are immediately boxed:
68+
69+
- `Box::new(proof)`
70+
- `Box::new(compressed_accounts)`
71+
- `Box::new(bumps)`
72+
- `Box::new(Vec::with_capacity(...))`
73+
74+
4. **Iterator Optimization**: Removed iterator chaining that could create temporary stack allocations
75+
76+
5. **Cold Path Marking**: Helper functions marked with `#[cold]` to optimize for the common path
77+
78+
6. **No Inline**: All helpers use `#[inline(never)]` to ensure separate stack frames
79+
80+
## Benefits
81+
82+
- **Stack Safety**: Each function now uses well under the 4096 byte limit
83+
- **Maintainability**: Smaller, focused functions are easier to understand
84+
- **Debuggability**: Stack traces will show which helper failed
85+
- **Flexibility**: Individual helpers can be further optimized if needed
86+
87+
## Estimated Stack Usage
88+
89+
| Function | Before | After V1 | After V2 | After V3 |
90+
| ----------------------------------- | ---------- | ---------- | ---------- | ---------- |
91+
| decompress_accounts_idempotent | 6080 bytes | ~500 bytes | ~500 bytes | ~500 bytes |
92+
| setup_cpi_and_config | N/A | ~200 bytes | ~200 bytes | ~200 bytes |
93+
| process_single_compressed_variant | N/A | 4392 bytes | 4312 bytes | ~150 bytes |
94+
| dispatch_variant | N/A | N/A | N/A | ~150 bytes |
95+
| call_prepare_accounts | N/A | N/A | ~300 bytes | ~300 bytes |
96+
| invoke_cpi_with_compressed_accounts | N/A | ~200 bytes | ~200 bytes | ~200 bytes |
97+
98+
Total maximum stack depth: ~1500 bytes (well under 4096 limit)
99+
100+
## Testing Recommendations
101+
102+
1. Test with maximum number of compressed accounts
103+
2. Verify stack usage with `solana-stack-check` tool
104+
3. Profile with different account types
105+
4. Test error paths to ensure stack safety in all cases

0 commit comments

Comments
 (0)