Medium: fixStringEncodedNumber is O(n²) — allocates per match

[metis-sprock]

Description

fixStringEncodedNumber allocates a new []byte slice and copies the entire data buffer for every regex match found. For data with many string-encoded numbers, this results in O(n²) memory allocation and copying behavior.

Evidence

File: pkg/codec/echoreplay.go:308-330

func fixStringEncodedNumber(data []byte) []byte {
    // For each match:
    //   - allocates new []byte with make()
    //   - copies prefix + replacement + suffix
    //   - repeats for next match on the new allocation
}

Each match triggers a full-buffer copy. With hundreds of matches per frame (common in Echo VR replay data with many numeric fields), this becomes a significant hot path.

Impact

Performance degradation during conversion. For frames with many string-encoded numbers, the quadratic allocation pattern causes excessive GC pressure and CPU usage. This is measurable on large tape files.

Suggested Fix

Use bytes.Buffer or strings.Builder with a single pass, or use regexp.ReplaceAllFunc which handles the iteration internally with fewer allocations:

func fixStringEncodedNumber(data []byte) []byte {
    return re.ReplaceAllFunc(data, func(match []byte) []byte {
        // strip quotes from the number
        return match[1 : len(match)-1]
    })
}

🤖 Filed by Metis