Add Buffered<T> to add buffering to byte streams. #3405
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
We shouldn't assert that the input file is valid.
I suspected an error in CircularDuplexStream::read(Bytes, size_t). This
does not appear to be the case, this test case is useful regardless.
The following script was used to generate the test:
import gzip
uncompressed = []
for _ in range(0x100):
uncompressed.append(1)
for _ in range(0x7e00):
uncompressed.append(0)
for _ in range(0x100):
uncompressed.append(1)
compressed = gzip.compress(bytes(uncompressed))
compressed = ", ".join(f"0x{byte:02x}" for byte in compressed)
print(f"""\
TEST_CASE(gzip_decompress_repeat_around_buffer)
{{
const u8 compressed[] = {{
{compressed}
}};
u8 uncompressed[0x8011];
Bytes{{ uncompressed, sizeof(uncompressed) }}.fill(0);
uncompressed[0x8000] = 1;
const auto decompressed = Compress::GzipDecompressor::decompress_all({{ compressed, sizeof(compressed) }});
EXPECT(compare({{ uncompressed, sizeof(uncompressed) }}, decompressed.bytes()));
}}
""", end="")
The streaming operator doesn't short-circuit, consider the following
snippet:
void foo(InputStream& stream) {
int a, b;
stream >> a >> b;
}
If the first read fails, the second is called regardless. It should be
well defined what happens in this case: nothing.
nico
added a commit
to nico/serenity
that referenced
this pull request
Sep 9, 2024
Symbols that need <= 8 bits hit a fast path as of SerenityOS#18075, but the slow path has done a full binary search over all symbols ever since this code was added in SerenityOS#2963. (SerenityOS#3405 even added a FIXME for doing this, but SerenityOS#18075 removed it.) Instead of doing a binary search over all codes for every single bit read, this implements the Moffat-Turpin approach described at https://www.hanshq.net/zip.html#huffdec, which only requires a table read per bit. hyperfine 'Build/lagom/bin/unzip ~/Downloads/enwik8.zip' 1.008 s ± 0.016 s => 957.7 ms ± 3.9 ms, 5% faster Due to issue SerenityOS#25005, we can't peek the full 15 bits at once but have to read them one-by-one. This makes the code look a bit different than in the linked article. I also tried not changing CanonicalCode::from_bytes() too much. It does 16 passes over all symbols. I think it could do it in a single pass instead. But that's for a future change. No behavior change (other than slightly faster perf).
nico
added a commit
to nico/serenity
that referenced
this pull request
Sep 10, 2024
Symbols that need <= 8 bits hit a fast path as of SerenityOS#18075, but the slow path has done a full binary search over all symbols ever since this code was added in SerenityOS#2963. (SerenityOS#3405 even added a FIXME for doing this, but SerenityOS#18075 removed it.) Instead of doing a binary search over all codes for every single bit read, this implements the Moffat-Turpin approach described at https://www.hanshq.net/zip.html#huffdec, which only requires a table read per bit. hyperfine 'Build/lagom/bin/unzip ~/Downloads/enwik8.zip' 1.008 s ± 0.016 s => 957.7 ms ± 3.9 ms, 5% faster Due to issue SerenityOS#25005, we can't peek the full 15 bits at once but have to read them one-by-one. This makes the code look a bit different than in the linked article. I also tried not changing CanonicalCode::from_bytes() too much. It does 15 passes over all symbols. I think it could do it in a single pass instead. But that's for a future change. No behavior change (other than slightly faster perf).
timschumi
pushed a commit
that referenced
this pull request
Sep 14, 2024
Symbols that need <= 8 bits hit a fast path as of #18075, but the slow path has done a full binary search over all symbols ever since this code was added in #2963. (#3405 even added a FIXME for doing this, but #18075 removed it.) Instead of doing a binary search over all codes for every single bit read, this implements the Moffat-Turpin approach described at https://www.hanshq.net/zip.html#huffdec, which only requires a table read per bit. hyperfine 'Build/lagom/bin/unzip ~/Downloads/enwik8.zip' 1.008 s ± 0.016 s => 957.7 ms ± 3.9 ms, 5% faster Due to issue #25005, we can't peek the full 15 bits at once but have to read them one-by-one. This makes the code look a bit different than in the linked article. I also tried not changing CanonicalCode::from_bytes() too much. It does 15 passes over all symbols. I think it could do it in a single pass instead. But that's for a future change. No behavior change (other than slightly faster perf).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is more of a collection of loosely related stuff.
I added a
Buffered<T>template that adds buffering to any class that inherits fromInputStreamorOutputStream.I've tried to get the Gzip implementation to a state where it can correctly decompress usual files (as opposed to files constructed to test one specific behavior.) I already found and fixed a few bugs but there seems to be a bug buried in
DeflateDecompressor::decode_distancestill.I also accumulated a few unit tests that I wrote to track down errors.
It started to become annoying to rebase so I thought I'd merge some of it.