fix(decoder): decode multi-member gzip data

[lizeyan]

Summary

#3269

Allow GZipDecoder to decode multi-member gzip data

Checklist

• I understand that this PR may be closed in case there was no previous discussion. (This doesn't apply to typos!)
• I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
• I've updated the documentation accordingly.

[tomchristie]

Rather than using a private import, could we switch this to use the same style as the other test cases here.

[lizeyan]

I changed the style of the test case

[tomchristie]

I'm wondering if there's an implementation using something like...

self._buffer = io.BytesIO()
self._reader = gzip.GzipFile(fileobj=self._buffer, mode='r')

Would using GzipFile be more robust here / catch other cases?

[lizeyan]

It seems that zlib is slightly faster than gzip

import gzip
import zlib
import time
import random
import string

def generate_random_data(size):
    return ''.join(random.choices(string.ascii_letters + string.digits, k=size)).encode()

def test_gzip_decompress(data, iterations):
    compressed = gzip.compress(data)
    start_time = time.time()
    for _ in range(iterations):
        gzip.decompress(compressed)
    end_time = time.time()
    return end_time - start_time

def test_zlib_decompress(data, iterations):
    compressed = zlib.compress(data)
    start_time = time.time()
    for _ in range(iterations):
        zlib.decompress(compressed)
    end_time = time.time()
    return end_time - start_time

# 测试参数
data_sizes = [1000, 10000, 100000]
iterations = 10000

print("Testing gzip and zlib decompression performance:")
print("------------------------------------------------")

for size in data_sizes:
    print(f"\nTesting with data size: {size} bytes")
    data = generate_random_data(size)
    
    gzip_time = test_gzip_decompress(data, iterations)
    zlib_time = test_zlib_decompress(data, iterations)
    
    print(f"gzip decompression time: {gzip_time:.4f} seconds")
    print(f"zlib decompression time: {zlib_time:.4f} seconds")
    
    if gzip_time < zlib_time:
        print(f"gzip is faster by {(zlib_time / gzip_time - 1) * 100:.2f}%")
    else:
        print(f"zlib is faster by {(gzip_time / zlib_time - 1) * 100:.2f}%")

print("\nTest completed.")

Testing gzip and zlib decompression performance:
------------------------------------------------

Testing with data size: 1000 bytes
gzip decompression time: 0.0503 seconds
zlib decompression time: 0.0386 seconds
zlib is faster by 30.47%

Testing with data size: 10000 bytes
gzip decompression time: 0.2463 seconds
zlib decompression time: 0.2337 seconds
zlib is faster by 5.38%

Testing with data size: 100000 bytes
gzip decompression time: 2.3218 seconds
zlib decompression time: 2.3094 seconds
zlib is faster by 0.54%

Test completed.

[tomchristie]

Thanks for your time looking into this.

We'll prefer following Chrome/Safari behavior here... #3269 (reply in thread)

[rafalkrupinski]

We'll prefer following Chrome/Safari behavior here.

seriously, why?

[tomchristie]

Because introducing complexity to support a use-case that browsers don't support is a trade-off I'd rather not make.

(More broadly speaking... httpx should prefer a design aesthetic of minimal complexity, where possible.)