|
|
|
/**
|
|
|
|
* Writes data from the VM step to the in-memory
|
|
|
|
* coverage map constructed by the Instrumenter.
|
|
|
|
*/
|
|
|
|
class DataCollector {
|
|
|
|
constructor(instrumentationData={}, viaIR){
|
|
|
|
this.instrumentationData = instrumentationData;
|
|
|
|
|
|
|
|
this.validOpcodes = this._getOpcodes(viaIR);
|
|
|
|
this.lastHash = null;
|
|
|
|
this.viaIR = viaIR;
|
|
|
|
this.pcZeroCounter = 0;
|
|
|
|
this.lastPcZeroCount = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* VM step event handler. Detects instrumentation hashes when they are pushed to the
|
|
|
|
* top of the stack. This runs millions of times - trying to keep it fast.
|
|
|
|
* @param {Object} info vm step info
|
|
|
|
*/
|
|
|
|
step(info){
|
|
|
|
if (info.pc === 0) this.pcZeroCounter++;
|
|
|
|
|
|
|
|
try {
|
|
|
|
if (this.validOpcodes[info.opcode.name] && info.stack.length > 0){
|
|
|
|
const idx = info.stack.length - 1;
|
|
|
|
let hash = '0x' + info.stack[idx].toString(16);
|
|
|
|
this._registerHash(hash);
|
|
|
|
}
|
|
|
|
} catch (err) { /*Ignore*/ };
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Normalizes has string and marks hit.
|
|
|
|
* @param {String} hash bytes32 hash
|
|
|
|
*/
|
|
|
|
_registerHash(hash){
|
|
|
|
hash = this._normalizeHash(hash);
|
|
|
|
|
|
|
|
if(this.instrumentationData[hash]){
|
|
|
|
// abi.encode (used to circumvent viaIR) sometimes puts the hash on the stack twice
|
|
|
|
// We should only skip duplicate hashes *within* a transaction (see issue #863)
|
|
|
|
if (this.lastHash !== hash || this.lastPcZeroCount !== this.pcZeroCounter) {
|
|
|
|
this.lastHash = hash;
|
|
|
|
this.lastPcZeroCount = this.pcZeroCounter;
|
|
|
|
this.instrumentationData[hash].hits++
|
|
|
|
}
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Left-pads zero prefixed bytes8 hashes to length 18. The '11' in the
|
|
|
|
* comparison below is arbitrary. It provides a margin for recurring zeros
|
|
|
|
* but prevents left-padding shorter irrelevant hashes
|
|
|
|
*
|
|
|
|
* @param {String} hash data hash from evm stack.
|
|
|
|
* @return {String} 0x prefixed hash of length 18.
|
|
|
|
*/
|
|
|
|
_normalizeHash(hash){
|
|
|
|
// viaIR sometimes right-pads the hashes out to 32 bytes
|
|
|
|
// but it doesn't preserve leading zeroes when it does this
|
|
|
|
if (this.viaIR && hash.length >= 18) {
|
|
|
|
hash = hash.slice(0,18);
|
|
|
|
|
|
|
|
// Detect and recover from viaIR mangled hashes by left-padding single `0`
|
|
|
|
if(!this.instrumentationData[hash]) {
|
|
|
|
hash = hash.slice(2);
|
|
|
|
hash = '0' + hash;
|
|
|
|
hash = hash.slice(0,16);
|
|
|
|
hash = '0x' + hash;
|
|
|
|
}
|
|
|
|
|
|
|
|
} else if (hash.length < 18 && hash.length > 11){
|
|
|
|
hash = hash.slice(2);
|
|
|
|
while(hash.length < 16) hash = '0' + hash;
|
|
|
|
hash = '0x' + hash
|
|
|
|
}
|
|
|
|
return hash;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Generates a list of all the opcodes to inspect for instrumentation hashes
|
|
|
|
* When viaIR is true, it includes all DUPs and PUSHs, so things are a little slower.
|
|
|
|
* @param {boolean} viaIR
|
|
|
|
*/
|
|
|
|
_getOpcodes(viaIR) {
|
|
|
|
let opcodes = {
|
|
|
|
"PUSH1": true
|
|
|
|
};
|
|
|
|
|
|
|
|
for (let i = 2; i <= 32; i++) {
|
|
|
|
const key = "PUSH" + i;
|
|
|
|
opcodes[key] = viaIR;
|
|
|
|
};
|
|
|
|
|
|
|
|
for (let i = 1; i <= 16; i++ ) {
|
|
|
|
const key = "DUP" + i;
|
|
|
|
opcodes[key] = viaIR;
|
|
|
|
}
|
|
|
|
|
|
|
|
return opcodes;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Unit test helper
|
|
|
|
* @param {Object} data Instrumenter.instrumentationData
|
|
|
|
*/
|
|
|
|
_setInstrumentationData(data){
|
|
|
|
this.instrumentationData = data;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
module.exports = DataCollector;
|