perf: direct-write stdout bypass StringWriter/StringBuffer allocation

[He-Pin]

Motivation

When outputting to stdout (no --output-file), sjsonnet renders JSON into a StringWriter, calls toString() to get the full string, then println() to write it. This creates a redundant copy: the JSON is materialized in memory as a String before being written. For large outputs (e.g., Kubernetes manifests), this doubles peak memory usage.

Key Design Decision

When writing to stdout (no output file), use a ByteArrayOutputStream + OutputStreamWriter pipeline instead of StringWriter. This:

1. Renders JSON directly into a byte buffer (no intermediate String)
2. Uses baos.writeTo(stdout) for a single bulk write
3. On rendering error, the buffer is simply discarded (nothing reaches stdout)

Modification

sjsonnet/src-jvm-native/sjsonnet/SjsonnetMainBase.scala:

• Added stdout: PrintStream parameter to writeToFile and renderNormal
• New code path when stdout != null and no output file: ByteArrayOutputStream(65536) → OutputStreamWriter → writeTo(stdout)
• Thread stdout through processFile → renderNormal → writeToFile

Benchmark Results

This optimization targets CLI I/O throughput, not JMH evaluation speed. The benefit is:

• Reduced memory: No intermediate String for stdout output
• Reduced copies: One writeTo call vs toString() + println()
• Measurable via CLI: time sjsonnet large_file.jsonnet > /dev/null

Analysis

• Error safety: ByteArrayOutputStream buffers all output. On error, the buffer is discarded — no partial output reaches stdout.
• Buffer size: Initial 65536 bytes (64KB) — handles most outputs without reallocation.
• No functional change: Only affects the I/O path when writing to stdout. --output-file path unchanged.

References

• ByteArrayOutputStream.writeTo(OutputStream) — zero-copy transfer to stdout
• Original StringWriter.toString allocation pattern

Result

Eliminates intermediate String allocation for stdout output. Reduces peak memory and I/O copies for CLI usage. Draft PR pending native benchmark data.