[emschwartz]

Fixes #1040.

Previously, quick-xml assumed that the XML declaration could be read out of only the first 64 bytes. However, there are valid documents that have declarations that are a few bytes longer than that.

This PR changes the prefix reader so that it has a concept of whether the declaration is still open. It will keep reading the underlying stream looking for the declaration's closing >, even if it goes beyond 64 bytes.

This prevents a panic that would be caused by calling set_encoding after the 64 bytes and after the prefix bytes had been dropped.

Claude's write-up

The documented DecodingReader loop (set_encoding(e.encoder()) on Event::Decl) panicked when the XML declaration did not fit in the 64-byte detection prefix:

<?xml version="1.0"
      encoding="ISO-8859-1"
      standalone="yes"?><d>é</d>

set_encoding() called after prefix buffer was drained; encoding can only be changed while the prefix is still active

With non-ASCII content after the declaration, reading failed with a UTF-8 error before Decl was even returned.

Fix: when the prefix is used up before the UTF-8 decoder has seen a >, read another 64-byte chunk into it instead of dropping it. The prefix then always ends after the declaration, so the existing "stop at the first >" logic from #1023 and the prefix.is_some() check in set_encoding work unchanged. Only the declaration goes through the small prefix; the rest is decoded by the main path as before (input with no > at all is now read 64 bytes at a time).

UTF-16 documents that declare UTF-8: with a BOM the prefix holds 31 UTF-16 characters, so a complete declaration such as <?xml version="1.0" encoding="UTF-8"?> hit the same panic. A shorter one switched the decoder to the declared encoding and produced garbage. set_encoding now ignores an ASCII-compatible encoding on a UTF-16 stream, since the declaration could only be read as UTF-16. The issue suggested returning an error instead; ignoring is one line and lets these documents parse. Switching to an error is easy if preferred.

This does not cover the UTF-16LE/BE case from #1033 (neither encoding is ASCII-compatible). Both PRs edit the same if in set_encoding, so whichever lands second needs a small rebase.

Tests: declaration_longer_than_prefix and utf16_declared_as_utf8 in tests/encodings.rs fail on master (the first with the UTF-8 error, the second with the panic above) and pass with the fix. A unit test in src/encoding.rs checks that only the declaration is read through the prefix.

Authored by Claude (claude-opus-5-5); reviewed by the submitter before posting.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 92.85714% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 53.87%. Comparing base (4a56631) to head (329847c).
⚠️ Report is 23 commits behind head on master.

Files with missing lines	Patch %	Lines
src/encoding.rs	92.85%	2 Missing ⚠️
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1041      +/-   ##
==========================================
+ Coverage   53.71%   53.87%   +0.15%     
==========================================
  Files          51       51              
  Lines       19412    19532     +120     
==========================================
+ Hits        10428    10523      +95     
- Misses       8984     9009      +25     

Flag	Coverage Δ
unittests	53.87% <92.85%> (+0.15%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Mingun]

I also suggest to increase this to 128, so the standard prelude can be fully read in one chunk even in UTF-16 case.

<?xml version="1.0" encoding=""?>

contains 33 characters + name of the encoding (the longest name of the supported encoding is "x-mac-cyrillic" which is 14 characters).