Skip to content

Optimize large buffer allocations with ArrayPool #78

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Tyrrrz merged 9 commits into from
Feb 17, 2026
Merged

Optimize large buffer allocations with ArrayPool #78

Show file tree
Hide file tree
Changes from 8 commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
375f042
Initial plan
Copilot Feb 17, 2026
81a3dd6
Replace array allocations with ArrayPool for memory optimization
Copilot Feb 17, 2026
b9e4bd0
Revert ArrayPool changes for small buffers - not beneficial
Copilot Feb 17, 2026
fbe7c1d
Fix TextReader.ReadAsync to correctly slice pooled buffer
Copilot Feb 17, 2026
c0c62e9
Add buffer size constants for clarity and consistency
Copilot Feb 17, 2026
a132bff
Address PR feedback: rename variable, inline expressions, use MemoryP…
Copilot Feb 17, 2026
d57680f
Address feedback: rename to buffer, append char-by-char, simplify sli…
Copilot Feb 17, 2026
65c8a88
Inline span variable access in char-by-char append loops
Copilot Feb 17, 2026
f2b8c03
Incorporate review feedback: use ArrayPool with array-based ReadAsync…
Copilot Feb 17, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 15 additions & 8 deletions PolyShim/Net50/Convert.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@
// ReSharper disable PartialTypeWithSinglePart

using System;
using System.Buffers;
using System.Diagnostics.CodeAnalysis;

[ExcludeFromCodeCoverage]
Expand Down Expand Up @@ -48,17 +49,23 @@ static int GetHexValue(char c)
// https://learn.microsoft.com/dotnet/api/system.convert.tohexstring#system-convert-tohexstring(system-byte()-system-int32-system-int32)
public static string ToHexString(byte[] value, int startIndex, int length)
{
var c = new char[length * 2];
var chars = ArrayPool<char>.Shared.Rent(length * 2);
try
{
for (var i = 0; i < length; i++)
{
var b = value[startIndex + i] >> 4;
chars[i * 2] = (char)(55 + b + (((b - 10) >> 31) & -7));
b = value[startIndex + i] & 0xF;
chars[i * 2 + 1] = (char)(55 + b + (((b - 10) >> 31) & -7));
}

for (var i = 0; i < length; i++)
return new string(chars, 0, length * 2);
}
finally
{
var b = value[startIndex + i] >> 4;
c[i * 2] = (char)(55 + b + (((b - 10) >> 31) & -7));
b = value[startIndex + i] & 0xF;
c[i * 2 + 1] = (char)(55 + b + (((b - 10) >> 31) & -7));
ArrayPool<char>.Shared.Return(chars);
}

return new string(c);
}

// https://learn.microsoft.com/dotnet/api/system.convert.tohexstring#system-convert-tohexstring(system-byte())
Expand Down
11 changes: 8 additions & 3 deletions PolyShim/Net70/TextReader.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@
// ReSharper disable PartialTypeWithSinglePart

using System;
using System.Buffers;
using System.IO;
using System.Text;
using System.Threading;
Expand All @@ -30,18 +31,22 @@ internal static class MemberPolyfills_Net70_TextReader
public async Task<string> ReadToEndAsync(CancellationToken cancellationToken)
{
var result = new StringBuilder();
var buffer = new char[4096];
using var buffer = MemoryPool<char>.Shared.Rent(4096);

while (true)
{
var charsRead = await reader
.ReadAsync(buffer, cancellationToken)
.ReadAsync(buffer.Memory, cancellationToken)
.ConfigureAwait(false);

Copilot AI Feb 17, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Using TextReader.ReadAsync(Memory, CancellationToken) here can route through PolyShim’s Memory polyfill on NETSTANDARD<2.1/NETCOREAPP<2.1, which allocates a new char[] (buffer.ToArray()) per call. For this ReadToEndAsync loop, prefer a pooled char[] with ReadAsync(char[], int, int) to avoid per-iteration allocations.

Copilot uses AI. Check for mistakes.

if (charsRead <= 0)
break;

result.Append(buffer, 0, charsRead);
// Append char by char to avoid string allocation
for (var i = 0; i < charsRead; i++)
{
result.Append(buffer.Memory.Span[i]);
}

Copilot AI Feb 17, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Appending one char at a time is much slower than appending the whole chunk. The previous approach (Append(buffer, 0, charsRead) on a char[]) avoids intermediate string allocations already; consider restoring chunk appends (or using StringBuilder.Append(ReadOnlySpan) where available).

Suggested change
// Append char by char to avoid string allocation
for (var i = 0; i < charsRead; i++)
{
result.Append(buffer.Memory.Span[i]);
}
// Append the chunk directly to avoid intermediate string allocations
result.Append(buffer.Memory.Span.Slice(0, charsRead));

Copilot uses AI. Check for mistakes.
}

return result.ToString();
Expand Down
16 changes: 12 additions & 4 deletions PolyShim/Net80/RandomNumberGenerator.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@
// ReSharper disable PartialTypeWithSinglePart

using System;
using System.Buffers;
using System.Text;
using System.Diagnostics.CodeAnalysis;
using System.Security.Cryptography;
Expand Down Expand Up @@ -46,11 +47,18 @@ public static void Shuffle<T>(Span<T> items)
// https://learn.microsoft.com/dotnet/api/system.security.cryptography.randomnumbergenerator.gethexstring#system-security-cryptography-randomnumbergenerator-gethexstring(system-int32-system-boolean)
public static string GetHexString(int stringLength, bool lowercase = false)
{
var bytes = new byte[(stringLength + 1) / 2];
RandomNumberGenerator.Fill(bytes);
var bytes = ArrayPool<byte>.Shared.Rent((stringLength + 1) / 2);
try
{
RandomNumberGenerator.Fill(bytes);

var hex = lowercase ? Convert.ToHexStringLower(bytes) : Convert.ToHexString(bytes);
return hex.Substring(0, stringLength);
var hex = lowercase ? Convert.ToHexStringLower(bytes) : Convert.ToHexString(bytes);
return hex.Substring(0, stringLength);

Copilot AI Feb 17, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ArrayPool.Rent may return a buffer larger than requested; filling/converting the entire rented array makes Convert.ToHexString/ToHexStringLower do extra work and allocate a larger intermediate string before the Substring. Use the exact requested length ((stringLength + 1) / 2) by filling only that slice and calling the Convert.ToHexString overload that takes (byte[], startIndex, length).

Copilot uses AI. Check for mistakes.
}
finally
{
ArrayPool<byte>.Shared.Return(bytes);

Copilot AI Feb 17, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Consider returning the rented byte[] with clearArray: true. This buffer contains cryptographically strong random bytes that are commonly used as secrets; leaving them in the shared pool increases the risk of data exposure between unrelated callers.

Suggested change
ArrayPool<byte>.Shared.Return(bytes);
ArrayPool<byte>.Shared.Return(bytes, clearArray: true);

Copilot uses AI. Check for mistakes.
}
}

// https://learn.microsoft.com/dotnet/api/system.security.cryptography.randomnumbergenerator.getstring
Expand Down
20 changes: 14 additions & 6 deletions PolyShim/NetCore10/Stream.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@
// ReSharper disable InconsistentNaming
// ReSharper disable PartialTypeWithSinglePart

using System.Buffers;
using System.IO;
using System.Diagnostics.CodeAnalysis;

Expand All @@ -16,14 +17,21 @@ internal static class MemberPolyfills_NetCore10_Stream
// https://learn.microsoft.com/dotnet/api/system.io.stream.copyto#system-io-stream-copyto(system-io-stream-system-int32)
public void CopyTo(Stream destination, int bufferSize)
{
var buffer = new byte[bufferSize];
while (true)
var buffer = ArrayPool<byte>.Shared.Rent(bufferSize);
try
{
var bytesRead = source.Read(buffer, 0, buffer.Length);
if (bytesRead <= 0)
break;
while (true)
{
var bytesRead = source.Read(buffer, 0, buffer.Length);
if (bytesRead <= 0)
break;

destination.Write(buffer, 0, bytesRead);
destination.Write(buffer, 0, bytesRead);
}
}
finally
{
ArrayPool<byte>.Shared.Return(buffer);
}
}

Expand Down
29 changes: 21 additions & 8 deletions PolyShim/NetCore20/File.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@
// ReSharper disable InconsistentNaming
// ReSharper disable PartialTypeWithSinglePart

using System.Buffers;
using System.Collections.Generic;
using System.IO;
using System.Text;
Expand Down Expand Up @@ -130,19 +131,25 @@ public static async Task<string[]> ReadAllLinesAsync(string path, Encoding encod
// https://learn.microsoft.com/dotnet/api/system.io.file.readalltextasync#system-io-file-readalltextasync(system-string-system-threading-cancellationtoken)
public static async Task<string> ReadAllTextAsync(string path, CancellationToken cancellationToken = default)
{
using var stream = new FileStream(path, FileMode.Open, FileAccess.Read, FileShare.Read, 4096, true);
const int bufferSize = 4096;
using var stream = new FileStream(path, FileMode.Open, FileAccess.Read, FileShare.Read, bufferSize, true);
using var reader = new StreamReader(stream);

var content = new StringBuilder();
var buffer = new char[4096];
using var buffer = MemoryPool<char>.Shared.Rent(bufferSize);

cancellationToken.ThrowIfCancellationRequested();

int charsRead;
while ((charsRead = await reader.ReadAsync(buffer, 0, buffer.Length).ConfigureAwait(false)) > 0)
while ((charsRead = await reader.ReadAsync(buffer.Memory, cancellationToken).ConfigureAwait(false)) > 0)
{

Copilot AI Feb 17, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In this NetCore20 polyfill, calling StreamReader.ReadAsync(Memory, CancellationToken) will resolve to PolyShim’s TextReader.ReadAsync(Memory, ...) polyfill on NETSTANDARD<2.1/NETCOREAPP<2.1, which allocates a new char[] via buffer.ToArray() on every loop iteration (see PolyShim/NetCore21/TextReader.cs). That defeats the goal of reducing allocations; consider using ArrayPool (or a rented char[] via MemoryMarshal.TryGetArray) and the array-based ReadAsync(char[], int, int) overload instead.

Copilot uses AI. Check for mistakes.
cancellationToken.ThrowIfCancellationRequested();
content.Append(buffer, 0, charsRead);

// Append char by char to avoid string allocation
for (var i = 0; i < charsRead; i++)
{
content.Append(buffer.Memory.Span[i]);
}

Copilot AI Feb 17, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The per-character append loop is a significant slowdown and the comment is misleading: StringBuilder.Append(char[], int, int) does not allocate intermediate strings. If you switch back to an array-backed buffer, you can append the chunk in one call (or, on newer TFMs, use StringBuilder.Append(ReadOnlySpan) with a slice) to keep performance comparable to the BCL implementation.

Copilot uses AI. Check for mistakes.
}

return content.ToString();
Expand All @@ -151,19 +158,25 @@ public static async Task<string> ReadAllTextAsync(string path, CancellationToken
// https://learn.microsoft.com/dotnet/api/system.io.file.readalltextasync#system-io-file-readalltextasync(system-string-system-text-encoding-system-threading-cancellationtoken)
public static async Task<string> ReadAllTextAsync(string path, Encoding encoding, CancellationToken cancellationToken = default)
{
using var stream = new FileStream(path, FileMode.Open, FileAccess.Read, FileShare.Read, 4096, true);
const int bufferSize = 4096;
using var stream = new FileStream(path, FileMode.Open, FileAccess.Read, FileShare.Read, bufferSize, true);
using var reader = new StreamReader(stream, encoding);

var content = new StringBuilder();
var buffer = new char[4096];
using var buffer = MemoryPool<char>.Shared.Rent(bufferSize);

cancellationToken.ThrowIfCancellationRequested();

int charsRead;
while ((charsRead = await reader.ReadAsync(buffer, 0, buffer.Length).ConfigureAwait(false)) > 0)
while ((charsRead = await reader.ReadAsync(buffer.Memory, cancellationToken).ConfigureAwait(false)) > 0)
{

Copilot AI Feb 17, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same issue in the encoding overload: StreamReader.ReadAsync(Memory, CancellationToken) on older TFMs routes through the PolyShim TextReader.ReadAsync(Memory, ...) polyfill, which allocates via buffer.ToArray() each call. To actually reduce allocations, prefer a pooled char[] + ReadAsync(char[], int, int) here.

Copilot uses AI. Check for mistakes.
cancellationToken.ThrowIfCancellationRequested();
content.Append(buffer, 0, charsRead);

// Append char by char to avoid string allocation
for (var i = 0; i < charsRead; i++)
{
content.Append(buffer.Memory.Span[i]);
}

Copilot AI Feb 17, 2026

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The per-character append loop is likely to be much slower than appending the chunk directly (and Append(char[], int, int) doesn’t allocate a string). Consider using chunk append rather than appending one char at a time.

Copilot uses AI. Check for mistakes.
}

return content.ToString();
Expand Down
Loading