Decompression by JNI failed #373

New Issue

claunia · 2026-01-29T20:42:51Z

claunia commented

2026-01-29 20:42:51 +00:00

Originally created by @wlwenming on GitHub (Feb 7, 2021).

Decompressing did not complete, expected to return 4 (OK), but returned 2 (NEEDS_MORE_INPUT).

The debugging information is as follows:

Please help to analyze the reasons. Thanks.

Originally created by @wlwenming on GitHub (Feb 7, 2021). Decompressing did not complete, expected to return 4 (OK), but returned 2 (NEEDS_MORE_INPUT). The debugging information is as follows: ![image](https://user-images.githubusercontent.com/7542421/107138857-0d853300-6952-11eb-9691-58f367e1d6a0.png) Please help to analyze the reasons. Thanks.

claunia closed this issue

2026-01-29 20:42:52 +00:00

claunia commented

2026-01-29 20:42:52 +00:00

@wlwenming commented on GitHub (Feb 8, 2021):

I found a solution.Make sure the status is correct.

JNIEXPORT jobject JNICALL
Java_org_brotli_wrapper_dec_DecoderJNI_nativePull(
    JNIEnv* env, jobject /*jobj*/, jlongArray ctx) {
  jlong context[3];
  env->GetLongArrayRegion(ctx, 0, 3, context);
  DecoderHandle* handle = getHandle(reinterpret_cast<void*>(context[0]));
  size_t data_length = 0;
  const uint8_t* data = BrotliDecoderTakeOutput(handle->state, &data_length);
  bool hasMoreOutput = !!BrotliDecoderHasMoreOutput(handle->state);
  if (hasMoreOutput) {
    context[1] = 3;
  } else if (BrotliDecoderIsFinished(handle->state)) {
    /* Bytes after stream end are not allowed. */
    context[1] = (handle->input_offset == handle->input_length) ? 1 : 0;
  } else {
    /* Pull decompressed data is ok */
    context[1] = 4;
  }
  context[2] = hasMoreOutput ? 1 : 0;
  env->SetLongArrayRegion(ctx, 0, 3, context);
  return env->NewDirectByteBuffer(const_cast<uint8_t*>(data), data_length);
}

@wlwenming commented on GitHub (Feb 8, 2021): I found a solution.Make sure the status is correct. ``` JNIEXPORT jobject JNICALL Java_org_brotli_wrapper_dec_DecoderJNI_nativePull( JNIEnv* env, jobject /*jobj*/, jlongArray ctx) { jlong context[3]; env->GetLongArrayRegion(ctx, 0, 3, context); DecoderHandle* handle = getHandle(reinterpret_cast<void*>(context[0])); size_t data_length = 0; const uint8_t* data = BrotliDecoderTakeOutput(handle->state, &data_length); bool hasMoreOutput = !!BrotliDecoderHasMoreOutput(handle->state); if (hasMoreOutput) { context[1] = 3; } else if (BrotliDecoderIsFinished(handle->state)) { /* Bytes after stream end are not allowed. */ context[1] = (handle->input_offset == handle->input_length) ? 1 : 0; } else { /* Pull decompressed data is ok */ context[1] = 4; } context[2] = hasMoreOutput ? 1 : 0; env->SetLongArrayRegion(ctx, 0, 3, context); return env->NewDirectByteBuffer(const_cast<uint8_t*>(data), data_length); } ```

claunia commented

2026-01-29 20:42:53 +00:00

@eustas commented on GitHub (Jun 23, 2021):

So the situation happens when output size is exactly the power of 2.
In this case native decoder reports that output is ready.
Decoder saves the remaining part of last byte, which contains the "no more input" section (it is just 2 bits IIRC).
"take output" does not flip "is finished" state.
Going to think a little bit more, on what the right fix should be.
Next invocation to "process" would have detected that input is over...

@eustas commented on GitHub (Jun 23, 2021): So the situation happens when output size is exactly the power of 2. In this case native decoder reports that output is ready. Decoder saves the remaining part of last byte, which contains the "no more input" section (it is just 2 bits IIRC). "take output" does not flip "is finished" state. Going to think a little bit more, on what the right fix should be. Next invocation to "process" would have detected that input is over...

claunia commented

2026-01-29 20:42:53 +00:00

@eustas commented on GitHub (Jun 24, 2021):

Pinpointed the problem. Fix is coming.

@eustas commented on GitHub (Jun 24, 2021): Pinpointed the problem. Fix is coming.

claunia commented

2026-01-29 20:42:53 +00:00

@slandelle commented on GitHub (Jul 24, 2021):

@eustas gentle ping

@slandelle commented on GitHub (Jul 24, 2021): @eustas gentle ping

claunia commented

2026-01-29 20:42:53 +00:00