fix: compress request bodies with GZIPOutputStream instead of Okio GzipSink

.changeset/gzip-without-okio.md

@@ -0,0 +1,5 @@
+---
+"posthog": patch
+---
+
+Compress request bodies with `java.util.zip.GZIPOutputStream` instead of Okio `GzipSink`, so apps that resolve an Okio version older than 3.11.0 no longer send corrupt gzip bodies

posthog/src/main/java/com/posthog/internal/GzipRequestInterceptor.kt

@@ -22,15 +22,14 @@ import com.posthog.PostHogCompression
 import com.posthog.PostHogConfig
 import com.posthog.PostHogInternal
 import okhttp3.Interceptor
-import okhttp3.MediaType
 import okhttp3.MultipartBody
 import okhttp3.RequestBody
+import okhttp3.RequestBody.Companion.toRequestBody
 import okhttp3.Response
 import okio.Buffer
-import okio.BufferedSink
-import okio.GzipSink
-import okio.buffer
+import java.io.ByteArrayOutputStream
 import java.io.IOException
+import java.util.zip.GZIPOutputStream
 
 // https://square.github.io/okhttp/features/interceptors/
 
@@ -56,7 +55,7 @@ public class GzipRequestInterceptor(private val config: PostHogConfig) : Interce
                 try {
                     originalRequest.newBuilder()
                         .header("Content-Encoding", "gzip")
-                        .method(originalRequest.method, forceContentLength(gzip(body)))
+                        .method(originalRequest.method, gzip(body))
                         .build()
                 } catch (e: Throwable) {
                     config.logger.log("Failed to gzip the request body: $e.")
@@ -67,44 +66,17 @@ public class GzipRequestInterceptor(private val config: PostHogConfig) : Interce
         }
     }
 
-    private fun gzip(body: RequestBody): RequestBody {
-        return object : RequestBody() {
-            override fun contentType(): MediaType? {
-                return body.contentType()
-            }
-
-            override fun contentLength(): Long {
-                return -1 // We don't know the compressed length in advance!
-            }
-
-            @Throws(IOException::class)
-            override fun writeTo(sink: BufferedSink) {
-                val gzipSink = GzipSink(sink).buffer()
-                body.writeTo(gzipSink)
-                gzipSink.close()
-            }
-        }
-    }
-
-    // https://github.com/square/okhttp/issues/350
+    // Okio's GzipSink can emit corrupt output with Okio versions before 3.11.0 (square/okio#1608).
+    // Apps resolve their own Okio version, so compress with the JDK instead.
+    // Buffering the result also sets Content-Length, see https://github.com/square/okhttp/issues/350
     @Throws(IOException::class)
-    private fun forceContentLength(body: RequestBody): RequestBody {
-        val buffer = Buffer()
-        body.writeTo(buffer)
-
-        return object : RequestBody() {
-            override fun contentType(): MediaType? {
-                return body.contentType()
-            }
+    private fun gzip(body: RequestBody): RequestBody {
+        val uncompressed = Buffer()
+        body.writeTo(uncompressed)
 
-            override fun contentLength(): Long {
-                return buffer.size
-            }
+        val compressed = ByteArrayOutputStream()
+        GZIPOutputStream(compressed).use { uncompressed.writeTo(it) }
 
-            @Throws(IOException::class)
-            override fun writeTo(sink: BufferedSink) {
-                sink.write(buffer.snapshot())
-            }
-        }
+        return compressed.toByteArray().toRequestBody(body.contentType())
     }
 }

posthog/src/test/java/com/posthog/internal/GzipRequestInterceptorTest.kt

@@ -0,0 +1,69 @@
+package com.posthog.internal
+
+import com.posthog.API_KEY
+import com.posthog.PostHogConfig
+import com.posthog.TestHttpServers
+import okhttp3.MediaType.Companion.toMediaType
+import okhttp3.OkHttpClient
+import okhttp3.Request
+import okhttp3.RequestBody.Companion.toRequestBody
+import okhttp3.mockwebserver.RecordedRequest
+import org.junit.Rule
+import java.util.Random
+import java.util.zip.GZIPInputStream
+import kotlin.test.Test
+import kotlin.test.assertEquals
+
+internal class GzipRequestInterceptorTest {
+    @get:Rule
+    val httpServers = TestHttpServers()
+
+    private fun post(payload: String): RecordedRequest {
+        val http = httpServers.mockHttp()
+        val client =
+            OkHttpClient.Builder()
+                .addInterceptor(GzipRequestInterceptor(PostHogConfig(API_KEY)))
+                .build()
+        val request =
+            Request.Builder()
+                .url(http.url("/batch"))
+                .post(payload.toRequestBody("application/json; charset=utf-8".toMediaType()))
+                .build()
+        client.newCall(request).execute().close()
+        return http.takeRequest()
+    }
+
+    private fun largePayload(): String {
+        val random = Random(42)
+        return buildString {
+            append("[")
+            repeat(20_000) { index ->
+                if (index > 0) append(",")
+                append("{\"event\":\"event_$index\",\"properties\":{\"value\":\"")
+                repeat(64) { append('a' + random.nextInt(26)) }
+                append("\"}}")
+            }
+            append("]")
+        }
+    }
+
+    @Test
+    fun `compressed body of a large payload decompresses to the original payload`() {
+        val payload = largePayload()
+
+        val request = post(payload)
+
+        assertEquals("gzip", request.getHeader("Content-Encoding"))
+        val compressed = request.body.readByteArray()
+        assertEquals(compressed.size.toLong(), request.getHeader("Content-Length")?.toLong())
+        val decompressed = GZIPInputStream(compressed.inputStream()).use { it.readBytes().toString(Charsets.UTF_8) }
+        assertEquals(payload, decompressed)
+    }
+
+    @Test
+    fun `compressed body keeps the original content type`() {
+        val request = post("{}")
+
+        assertEquals("application/json; charset=utf-8", request.getHeader("Content-Type"))
+    }
+}