Align stream wrappers with the dotnet 11 implementation (follow up to #562) by SimonCropp · Pull Request #563 · SimonCropp/Polyfill

SimonCropp · 2026-06-27T01:36:55Z

No description provided.

…562) StringStream and ReadOnlySequenceStream diverged from dotnet/runtime#126669 in ways that hurt performance on the cases those types exist for. Bring them in line and port the PR's tests. StringStream - Encode incrementally via a stateful Encoder (fast-path single-shot, spillover buffer for sub-scalar reads, final flush) instead of encoding the whole text into one buffer on first read. Peak memory is now bounded by the caller's read buffer rather than the full encoded length, matching the BCL. - Span-based Encoder.Convert/Encoding.GetBytes on net core 2.1 / netstandard 2.1+, array-based equivalents on older TFMs. ReadOnlySequenceStream - Track a SequencePosition cursor alongside the absolute position so Read/ReadByte advance from the current segment instead of re-slicing from the start every call (was O(segments^2) for multi-segment sequences). Seek/Position reposition the cursor forward from the current spot, walking from the start only for back jumps. - Override CopyTo/CopyToAsync to write segments directly to the destination. CopyTo is gated to net core 2.1 / netstandard 2.1+ (Stream.CopyTo(Stream, int) is not virtual on older TFMs); CopyToAsync applies everywhere. ReadOnlyMemoryStream / WritableMemoryStream are unchanged: they are already at parity for array-backed memory, and matching the BCL for native (non-array) memory would require MemoryStream internals unavailable outside CoreLib. Tests - Port the explicit unit tests from the PR (StringStream encodings, surrogate pairs, chunk boundaries, GetMaxByteCount overflow guard, memory-slice/char-array ctors; WritableMemoryStream capacity/seek/overwrite cases) into the TUnit suite, plus ReadOnlySequenceStream CopyTo/CopyToAsync coverage. Regenerated Split + assemblySize; public API surface unchanged. Verified across all 22 TFMs (Consume) and passing on net10.0 and net48.

SimonCropp · 2026-06-27T01:36:55Z

No description provided.

…562) StringStream and ReadOnlySequenceStream diverged from dotnet/runtime#126669 in ways that hurt performance on the cases those types exist for. Bring them in line and port the PR's tests. StringStream - Encode incrementally via a stateful Encoder (fast-path single-shot, spillover buffer for sub-scalar reads, final flush) instead of encoding the whole text into one buffer on first read. Peak memory is now bounded by the caller's read buffer rather than the full encoded length, matching the BCL. - Span-based Encoder.Convert/Encoding.GetBytes on net core 2.1 / netstandard 2.1+, array-based equivalents on older TFMs. ReadOnlySequenceStream - Track a SequencePosition cursor alongside the absolute position so Read/ReadByte advance from the current segment instead of re-slicing from the start every call (was O(segments^2) for multi-segment sequences). Seek/Position reposition the cursor forward from the current spot, walking from the start only for back jumps. - Override CopyTo/CopyToAsync to write segments directly to the destination. CopyTo is gated to net core 2.1 / netstandard 2.1+ (Stream.CopyTo(Stream, int) is not virtual on older TFMs); CopyToAsync applies everywhere. ReadOnlyMemoryStream / WritableMemoryStream are unchanged: they are already at parity for array-backed memory, and matching the BCL for native (non-array) memory would require MemoryStream internals unavailable outside CoreLib. Tests - Port the explicit unit tests from the PR (StringStream encodings, surrogate pairs, chunk boundaries, GetMaxByteCount overflow guard, memory-slice/char-array ctors; WritableMemoryStream capacity/seek/overwrite cases) into the TUnit suite, plus ReadOnlySequenceStream CopyTo/CopyToAsync coverage. Regenerated Split + assemblySize; public API surface unchanged. Verified across all 22 TFMs (Consume) and passing on net10.0 and net48.

SimonCropp · 2026-06-27T01:36:55Z

No description provided.

…562) StringStream and ReadOnlySequenceStream diverged from dotnet/runtime#126669 in ways that hurt performance on the cases those types exist for. Bring them in line and port the PR's tests. StringStream - Encode incrementally via a stateful Encoder (fast-path single-shot, spillover buffer for sub-scalar reads, final flush) instead of encoding the whole text into one buffer on first read. Peak memory is now bounded by the caller's read buffer rather than the full encoded length, matching the BCL. - Span-based Encoder.Convert/Encoding.GetBytes on net core 2.1 / netstandard 2.1+, array-based equivalents on older TFMs. ReadOnlySequenceStream - Track a SequencePosition cursor alongside the absolute position so Read/ReadByte advance from the current segment instead of re-slicing from the start every call (was O(segments^2) for multi-segment sequences). Seek/Position reposition the cursor forward from the current spot, walking from the start only for back jumps. - Override CopyTo/CopyToAsync to write segments directly to the destination. CopyTo is gated to net core 2.1 / netstandard 2.1+ (Stream.CopyTo(Stream, int) is not virtual on older TFMs); CopyToAsync applies everywhere. ReadOnlyMemoryStream / WritableMemoryStream are unchanged: they are already at parity for array-backed memory, and matching the BCL for native (non-array) memory would require MemoryStream internals unavailable outside CoreLib. Tests - Port the explicit unit tests from the PR (StringStream encodings, surrogate pairs, chunk boundaries, GetMaxByteCount overflow guard, memory-slice/char-array ctors; WritableMemoryStream capacity/seek/overwrite cases) into the TUnit suite, plus ReadOnlySequenceStream CopyTo/CopyToAsync coverage. Regenerated Split + assemblySize; public API surface unchanged. Verified across all 22 TFMs (Consume) and passing on net10.0 and net48.

SimonCropp · 2026-06-27T01:36:55Z

No description provided.

…562) StringStream and ReadOnlySequenceStream diverged from dotnet/runtime#126669 in ways that hurt performance on the cases those types exist for. Bring them in line and port the PR's tests. StringStream - Encode incrementally via a stateful Encoder (fast-path single-shot, spillover buffer for sub-scalar reads, final flush) instead of encoding the whole text into one buffer on first read. Peak memory is now bounded by the caller's read buffer rather than the full encoded length, matching the BCL. - Span-based Encoder.Convert/Encoding.GetBytes on net core 2.1 / netstandard 2.1+, array-based equivalents on older TFMs. ReadOnlySequenceStream - Track a SequencePosition cursor alongside the absolute position so Read/ReadByte advance from the current segment instead of re-slicing from the start every call (was O(segments^2) for multi-segment sequences). Seek/Position reposition the cursor forward from the current spot, walking from the start only for back jumps. - Override CopyTo/CopyToAsync to write segments directly to the destination. CopyTo is gated to net core 2.1 / netstandard 2.1+ (Stream.CopyTo(Stream, int) is not virtual on older TFMs); CopyToAsync applies everywhere. ReadOnlyMemoryStream / WritableMemoryStream are unchanged: they are already at parity for array-backed memory, and matching the BCL for native (non-array) memory would require MemoryStream internals unavailable outside CoreLib. Tests - Port the explicit unit tests from the PR (StringStream encodings, surrogate pairs, chunk boundaries, GetMaxByteCount overflow guard, memory-slice/char-array ctors; WritableMemoryStream capacity/seek/overwrite cases) into the TUnit suite, plus ReadOnlySequenceStream CopyTo/CopyToAsync coverage. Regenerated Split + assemblySize; public API surface unchanged. Verified across all 22 TFMs (Consume) and passing on net10.0 and net48.

SimonCropp · 2026-06-27T01:36:55Z

No description provided.

…562) StringStream and ReadOnlySequenceStream diverged from dotnet/runtime#126669 in ways that hurt performance on the cases those types exist for. Bring them in line and port the PR's tests. StringStream - Encode incrementally via a stateful Encoder (fast-path single-shot, spillover buffer for sub-scalar reads, final flush) instead of encoding the whole text into one buffer on first read. Peak memory is now bounded by the caller's read buffer rather than the full encoded length, matching the BCL. - Span-based Encoder.Convert/Encoding.GetBytes on net core 2.1 / netstandard 2.1+, array-based equivalents on older TFMs. ReadOnlySequenceStream - Track a SequencePosition cursor alongside the absolute position so Read/ReadByte advance from the current segment instead of re-slicing from the start every call (was O(segments^2) for multi-segment sequences). Seek/Position reposition the cursor forward from the current spot, walking from the start only for back jumps. - Override CopyTo/CopyToAsync to write segments directly to the destination. CopyTo is gated to net core 2.1 / netstandard 2.1+ (Stream.CopyTo(Stream, int) is not virtual on older TFMs); CopyToAsync applies everywhere. ReadOnlyMemoryStream / WritableMemoryStream are unchanged: they are already at parity for array-backed memory, and matching the BCL for native (non-array) memory would require MemoryStream internals unavailable outside CoreLib. Tests - Port the explicit unit tests from the PR (StringStream encodings, surrogate pairs, chunk boundaries, GetMaxByteCount overflow guard, memory-slice/char-array ctors; WritableMemoryStream capacity/seek/overwrite cases) into the TUnit suite, plus ReadOnlySequenceStream CopyTo/CopyToAsync coverage. Regenerated Split + assemblySize; public API surface unchanged. Verified across all 22 TFMs (Consume) and passing on net10.0 and net48.

SimonCropp · 2026-06-27T01:36:55Z

No description provided.

…562) StringStream and ReadOnlySequenceStream diverged from dotnet/runtime#126669 in ways that hurt performance on the cases those types exist for. Bring them in line and port the PR's tests. StringStream - Encode incrementally via a stateful Encoder (fast-path single-shot, spillover buffer for sub-scalar reads, final flush) instead of encoding the whole text into one buffer on first read. Peak memory is now bounded by the caller's read buffer rather than the full encoded length, matching the BCL. - Span-based Encoder.Convert/Encoding.GetBytes on net core 2.1 / netstandard 2.1+, array-based equivalents on older TFMs. ReadOnlySequenceStream - Track a SequencePosition cursor alongside the absolute position so Read/ReadByte advance from the current segment instead of re-slicing from the start every call (was O(segments^2) for multi-segment sequences). Seek/Position reposition the cursor forward from the current spot, walking from the start only for back jumps. - Override CopyTo/CopyToAsync to write segments directly to the destination. CopyTo is gated to net core 2.1 / netstandard 2.1+ (Stream.CopyTo(Stream, int) is not virtual on older TFMs); CopyToAsync applies everywhere. ReadOnlyMemoryStream / WritableMemoryStream are unchanged: they are already at parity for array-backed memory, and matching the BCL for native (non-array) memory would require MemoryStream internals unavailable outside CoreLib. Tests - Port the explicit unit tests from the PR (StringStream encodings, surrogate pairs, chunk boundaries, GetMaxByteCount overflow guard, memory-slice/char-array ctors; WritableMemoryStream capacity/seek/overwrite cases) into the TUnit suite, plus ReadOnlySequenceStream CopyTo/CopyToAsync coverage. Regenerated Split + assemblySize; public API surface unchanged. Verified across all 22 TFMs (Consume) and passing on net10.0 and net48.

SimonCropp · 2026-06-27T01:36:55Z

No description provided.

…562) StringStream and ReadOnlySequenceStream diverged from dotnet/runtime#126669 in ways that hurt performance on the cases those types exist for. Bring them in line and port the PR's tests. StringStream - Encode incrementally via a stateful Encoder (fast-path single-shot, spillover buffer for sub-scalar reads, final flush) instead of encoding the whole text into one buffer on first read. Peak memory is now bounded by the caller's read buffer rather than the full encoded length, matching the BCL. - Span-based Encoder.Convert/Encoding.GetBytes on net core 2.1 / netstandard 2.1+, array-based equivalents on older TFMs. ReadOnlySequenceStream - Track a SequencePosition cursor alongside the absolute position so Read/ReadByte advance from the current segment instead of re-slicing from the start every call (was O(segments^2) for multi-segment sequences). Seek/Position reposition the cursor forward from the current spot, walking from the start only for back jumps. - Override CopyTo/CopyToAsync to write segments directly to the destination. CopyTo is gated to net core 2.1 / netstandard 2.1+ (Stream.CopyTo(Stream, int) is not virtual on older TFMs); CopyToAsync applies everywhere. ReadOnlyMemoryStream / WritableMemoryStream are unchanged: they are already at parity for array-backed memory, and matching the BCL for native (non-array) memory would require MemoryStream internals unavailable outside CoreLib. Tests - Port the explicit unit tests from the PR (StringStream encodings, surrogate pairs, chunk boundaries, GetMaxByteCount overflow guard, memory-slice/char-array ctors; WritableMemoryStream capacity/seek/overwrite cases) into the TUnit suite, plus ReadOnlySequenceStream CopyTo/CopyToAsync coverage. Regenerated Split + assemblySize; public API surface unchanged. Verified across all 22 TFMs (Consume) and passing on net10.0 and net48.

SimonCropp · 2026-06-27T01:36:55Z

No description provided.

…562) StringStream and ReadOnlySequenceStream diverged from dotnet/runtime#126669 in ways that hurt performance on the cases those types exist for. Bring them in line and port the PR's tests. StringStream - Encode incrementally via a stateful Encoder (fast-path single-shot, spillover buffer for sub-scalar reads, final flush) instead of encoding the whole text into one buffer on first read. Peak memory is now bounded by the caller's read buffer rather than the full encoded length, matching the BCL. - Span-based Encoder.Convert/Encoding.GetBytes on net core 2.1 / netstandard 2.1+, array-based equivalents on older TFMs. ReadOnlySequenceStream - Track a SequencePosition cursor alongside the absolute position so Read/ReadByte advance from the current segment instead of re-slicing from the start every call (was O(segments^2) for multi-segment sequences). Seek/Position reposition the cursor forward from the current spot, walking from the start only for back jumps. - Override CopyTo/CopyToAsync to write segments directly to the destination. CopyTo is gated to net core 2.1 / netstandard 2.1+ (Stream.CopyTo(Stream, int) is not virtual on older TFMs); CopyToAsync applies everywhere. ReadOnlyMemoryStream / WritableMemoryStream are unchanged: they are already at parity for array-backed memory, and matching the BCL for native (non-array) memory would require MemoryStream internals unavailable outside CoreLib. Tests - Port the explicit unit tests from the PR (StringStream encodings, surrogate pairs, chunk boundaries, GetMaxByteCount overflow guard, memory-slice/char-array ctors; WritableMemoryStream capacity/seek/overwrite cases) into the TUnit suite, plus ReadOnlySequenceStream CopyTo/CopyToAsync coverage. Regenerated Split + assemblySize; public API surface unchanged. Verified across all 22 TFMs (Consume) and passing on net10.0 and net48.

SimonCropp added 2 commits June 27, 2026 11:23

Update Directory.Build.props
894697a

SimonCropp added this to the 10.11.1 milestone Jun 27, 2026

SimonCropp merged commit e479ffc into mainJun 27, 2026
4 of 6 checks passed

SimonCropp deleted the Align-stream-wrappers-with-the-dotnet-11-implementation-(follow-up-to-#562) branch June 27, 2026 02:01