[MDA2AV]

Fixes #58

Why

TryExtractFullHeader* (and the response and FlexibleParser equivalents) report one less than the bytes they consumed, and the XML doc said the opposite. Every caller compensates with bytesRead + 1, including GenHTTP and our own example.

This takes option 2 from the issue. Fixing the value in place (option 1) would break every one of those callers silently: no compile error and no exception. The first body byte would be read as header, and a Content-Length body would arrive one byte short, which desyncs a keep-alive connection. Renaming the correct methods and marking the old ones [Obsolete] moves that break to compile time, the only place anyone can see it.

What changes

New entry points. Each returns consumed: the length of the header block including its CRLFCRLF, so the body starts at input[consumed]. It is 0 when the method returns false.

Obsolete (reports length − 1, or −1)	Replacement (reports the length)
UltraHardenedParser.TryExtractFullHeaderROM(ref ReadOnlyMemory<byte>, …)	UltraHardenedParser.TryParseRequestHeader(ref ReadOnlyMemory<byte>, …)
UltraHardenedParser.TryExtractFullHeaderValidated(ref ReadOnlySequence<byte>, …)	UltraHardenedParser.TryParseRequestHeader(ref ReadOnlySequence<byte>, …)
UltraHardenedParser.TryExtractFullResponseHeaderROM(…)	UltraHardenedParser.TryParseResponseHeader(ref ReadOnlyMemory<byte>, …)
UltraHardenedParser.TryExtractFullResponseHeaderValidated(…)	UltraHardenedParser.TryParseResponseHeader(ref ReadOnlySequence<byte>, …)
FlexibleParser.TryExtractFullHeaderReadOnlyMemory(…)	FlexibleParser.TryParseRequestHeader(ref ReadOnlyMemory<byte>, …)
FlexibleParser.TryExtractFullHeader(…)	FlexibleParser.TryParseRequestHeader(ref ReadOnlySequence<byte>, …)

Migration: switch to the new name and drop the + 1. The parameters are otherwise identical.

• The methods are renamed in place: same files, same bodies. Only the signature, the docs and the two lines that assign the count change.
• The old names live in UltraHardenedParser.Obsolete.cs / FlexibleParser.Obsolete.cs as thin wrappers that return exactly the legacy values (consumed - 1, or -1 when incomplete). Their docs now say what they actually return.
• [Obsolete] is a warning (CS0618), not an error. The message names the difference and the replacement. These members are slated for removal in a later release.
• Docs move to the new API: README (which also corrects "advance your reader by bytesRead"), SECURITY.md, site/glyph11.html (including the PipeReader sample that advanced by bytesRead + 1), and Examples/Glyph11.Example.
• Glyph11 version goes to 0.4.0.

Not touched, on purpose:

• PicoParser, core/ and the diff harness: bindings are being removed in a separate PR.
• Benchmarks/: the bindings PR (Drop the bindings and the C core: Glyph11 is managed-only #84) rewrites that file, so its two Glyph11 calls stay on the obsolete names here to avoid a conflict. They build with two CS0618 warnings, and once both PRs are merged they need a two-line rename.

GenHTTP builds with TreatWarningsAsErrors, so it needs a migration (new names, no + 1): Kaliumhexacyanoferrat/GenHTTP#972. The draft PR Kaliumhexacyanoferrat/GenHTTP#973 does it. Its full acceptance suite passes on net10.0 and net11.0 against a local pack of this repo with every open PR merged.

Decisions worth reviewing

• New names rather than overloads. An overload that differs only in what the out int means cannot coexist with the old one: same name, same parameter list. Changing it to out long would make out var ambiguous. New names keep the migration a rename.
• The test project suppresses CS0618. The existing suites reach the legacy entry points through a handful of helpers (17 call sites), and they keep doing so on purpose: those entry points still ship and must keep their exact behaviour until removed. Migrating the helpers would also collide with sibling PRs that edit the same test files.

Overlap with sibling PRs

#59, #64, #65, #68, #71 and #73 edit the bodies of these same methods, which keep their files and structure here. Expect small textual conflicts around the signature and the bytesReadCount lines. Whichever lands second must also rename any new line it adds that assigns bytesReadCount to consumed, keeping the new meaning (the full length, or 0 when incomplete).

Tests

New tests/Tests/ConsumedCount.Tests.cs, 221 cases. Every new entry point is fed four shapes: contiguous memory, a single-segment sequence, three segments, and one byte per segment.

• consumed equals the header block length, and what follows it (a body or the next pipelined message) starts exactly there.
• Incomplete input returns false with consumed == 0.
• The obsolete entry points return consumed - 1 on the same corpus, and -1 when incomplete.

A mutation check (consumed = total - 1) fails 35 tests. Full suite: 624 passing, 0 failing (net10.0). dotnet build src/Glyph11.sln is clean; the only CS0618 warnings are the two in Benchmarks/ noted above.

SonarCloud

Two findings here are expected, and the code stays as is:

• S1133 on the six obsolete wrappers: the reminder to remove deprecated code, which is what they are for.
• S3776 (cognitive complexity) on the three renamed parse methods. Their complexity is unchanged (34, 109 and 76). The rename touches the line the existing issue sits on, so SonarCloud reports it as new here.