Mediocre gzip compression using flate2 with miniz_oxide

[Fishrock123]

Hi there, I'm using flate2 with the default backend (apparently miniz_oxide) via the async_compression crate.

I have a "file" like this (in a test):

const TEXT: &'static str = concat![
    "Chunk one\n",
    "data data\n",
    "\n",
    "Chunk two\n",
    "data data\n",
    "\n",
    "Chunk three\n",
    "data data\n",
];

flate2 appears to compress that to the following (hex dump from binary):

00000000: 1f8b 0800 0000 0000 00ff 6dca b109 0030  ..........m....0
00000010: 0805 d1fe 4fe1 2e4e 2244 1002 0ac1 90f5  ....O..N"D......
00000020: 4352 59d8 5cf1 38b6 ed93 c215 4352 e805  CRY.\.8.....CR..
00000030: e08f 79a2 415b 5adf 0bde f2d7 8140 0000  ..y.A[Z......@..
00000040: 00                                       .

However, this test is being partially ported from Node.js, where I did something similar a while ago. Node.js uses zlib. The hex resulting from that binary is this:

00000000: 1f8b 0800 0000 0000 0013 73ce 28cd cb56  ..........s.(..V
00000010: c8cf 4be5 4a49 2c49 5400 115c 5cce 60c1  ..K.JI,IT..\\.`.
00000020: 92f2 7c2c 8219 45a9 c86a 01de f2d7 8140  ..|,..E..j.....@
00000030: 0000 00                                  ...

The zlib one here is quite a bit smaller.

Both of these do appear to be valid, and both decompress fine with other tools.

Addendum: this is with the "default" quality setting (zlib level 6 or 7 (whichever it is), default flate2). Changing flate2 to "best" does not appear to make a difference.

[alexcrichton]

This might best be an issue for miniz-oxide? That's just a straight port of miniz though, so I'm not sure that much can be done. I'd recomend enabling the zlib backend if you get better results with that

[oyvindln]

Yeah this would likely be an issue with miniz_oxide, provided you are sure you are providing the same input data. The output ought to be very similar, especially for a short input like this. Difference between default and max compression levels wont come into play when compressing a small file like this.

[joshtriplett]

Closing this issue in favor of the issue on miniz_oxide.