# Daniel Lemire

Daniel Lemire is a software performance expert. He ranks among the top 2% of scientists globally (Stanford/Elsevier 2025) and is one of GitHub's top 1000 most followed developers.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Decoding base64 email attachments: a speed ranking of programming languages

DevFeed: [Decoding base64 email attachments: a speed ranking of programming languages](<https://devfeed.tech/articles/decoding-base64-email-attachments-a-speed-ranking-of-programming-languages-146498.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/11/20/decoding-base64-email-attachments/>)

Author: Daniel Lemire

Published: 2026-11-20T12:37:52Z

Content type: news

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Decoding](<https://devfeed.tech/topics/decoding.md>), [Programming](<https://devfeed.tech/topics/programming.md>), [Programming Languages / Compilers](<https://devfeed.tech/topics/programming-languages-compilers.md>), [email](<https://devfeed.tech/topics/email.md>)

Tags: [base64](<https://devfeed.tech/tags/base64.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [email](<https://devfeed.tech/tags/email.md>), [languages](<https://devfeed.tech/tags/languages.md>), [mime](<https://devfeed.tech/tags/mime.md>), [php](<https://devfeed.tech/tags/php.md>), [programming](<https://devfeed.tech/tags/programming.md>), [programming-languages](<https://devfeed.tech/tags/programming-languages.md>), [pull-request](<https://devfeed.tech/tags/pull-request.md>), [ranking](<https://devfeed.tech/tags/ranking.md>), [speed](<https://devfeed.tech/tags/speed.md>)

### AI overview

A benchmark of MIME-style base64 decoding found a thirtyfold gap between the fastest and slowest tested systems. It used one megabyte of random data encoded in 76-character lines with CRLF, verified decoded bytes, and measured median throughput across repeated runs on one pinned core of an Intel Xeon Gold 6548N. The excerpt says the top six systems use simdutf, which removes line endings in its vector loop and compacts affected blocks into a small buffer instead of falling back to a slow path. PHP's current release measured 0.61 GB/s on this input despite exceeding 20 GB/s on uninterrupted base64; a proposed simdutf port reportedly made MIME decoding 28 times faster, moving PHP to second place. The excerpt does not provide the other systems' individual results.

### Source excerpt

Email attachments are base64. So are cryptographic PEM certificates. In both cases the text is broken into lines, 76 characters for email and 64 for PEM. Skipping line endings sounds trivial. Yet many fast decoders only work on uninterrupted base64. As soon as they meet a line break, they fall back to a slow path. ... Continue reading Decoding base64 email attachments: a speed ranking of programming languages

## Java4Shell: when base64 decoding leads to remote code execution?

DevFeed: [Java4Shell: when base64 decoding leads to remote code execution?](<https://devfeed.tech/articles/java4shell-when-base64-decoding-leads-to-remote-code-execution-144011.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/11/18/java4shell-when-base64-decoding-leads-to-remote-code-execution/>)

Author: Daniel Lemire

Published: 2026-11-18T19:43:03Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [openjdk](<https://devfeed.tech/topics/openjdk.md>), [Decoding](<https://devfeed.tech/topics/decoding.md>), [code analysis](<https://devfeed.tech/topics/code-analysis.md>), [Exception](<https://devfeed.tech/topics/exception.md>), [Pull Request](<https://devfeed.tech/topics/pull-request.md>)

Tags: [amd-epyc](<https://devfeed.tech/tags/amd-epyc.md>), [assembly](<https://devfeed.tech/tags/assembly.md>), [avx](<https://devfeed.tech/tags/avx.md>), [avx-512-vbmi](<https://devfeed.tech/tags/avx-512-vbmi.md>), [base64](<https://devfeed.tech/tags/base64.md>), [brave](<https://devfeed.tech/tags/brave.md>), [bug](<https://devfeed.tech/tags/bug.md>), [bun](<https://devfeed.tech/tags/bun.md>), [chrome](<https://devfeed.tech/tags/chrome.md>), [chromium](<https://devfeed.tech/tags/chromium.md>), [cve](<https://devfeed.tech/tags/cve.md>), [cve-2026-114938](<https://devfeed.tech/tags/cve-2026-114938.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [deno](<https://devfeed.tech/tags/deno.md>), [edge](<https://devfeed.tech/tags/edge.md>), [epyc](<https://devfeed.tech/tags/epyc.md>), [fast](<https://devfeed.tech/tags/fast.md>), [gold](<https://devfeed.tech/tags/gold.md>), [openjdk](<https://devfeed.tech/tags/openjdk.md>), [reading](<https://devfeed.tech/tags/reading.md>)

### AI overview

The excerpt describes CVE-2026-114938, an OpenJDK Base64 decoder bug present since JDK 17.0.6: on x86-64 with AVX-512 VBMI, decoding 128 "A" characters followed by "=" allocates 98 bytes, then a malformed tail length produces a 64-byte masked store at offset 96. Two bytes fit; 62 are written past the array before Java throws IllegalArgumentException. The author says claims of a major security vulnerability are unverified. The excerpt also contrasts Java's fast path, which stops at whitespace, with a newer SIMD decoder that compacts whitespace; it lists browser and runtime deployments and reports 30x and 39x speedups for 64 KiB MIME input on two named processors.

### Source excerpt

Let us start with an example that triggers the issue. byte[] src = ("A".repeat(128) + "=").getBytes(StandardCharsets.US_ASCII); Base64.getDecoder().decode(src); One hundred and twenty-eight letters A and one =. The string is not valid base64, and a correct decoder rejects it. OpenJDK writes past the end of the array anyway. The bug is CVE-2026-114938. It has been in ... Continue reading Java4Shell: when base64 decoding leads to remote code execution?

## Cognitive radius

DevFeed: [Cognitive radius](<https://devfeed.tech/articles/cognitive-radius-135952.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/11/16/cognitive-radius/>)

Author: Daniel Lemire

Published: 2026-11-16T16:51:07Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [developer-productivity](<https://devfeed.tech/topics/developer-productivity.md>), [Developer experience](<https://devfeed.tech/topics/developer-experience.md>), [Developer Enablement](<https://devfeed.tech/topics/developer-enablement.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [context](<https://devfeed.tech/tags/context.md>), [radius](<https://devfeed.tech/tags/radius.md>)

### AI overview

AI's main contribution to software work, the article argues, is expanding an engineer's "cognitive radius": the projects, technologies, and past decisions they can keep in view while producing consistent work. With a good model and suitable tools, developers can bring repository structure, team conventions, issues, design decisions, constraints, and history into a session, making it easier to contribute to unfamiliar codebases, resume work, and maintain consistency across languages or frameworks. This depends on supplying relevant context, grounding questions, and checking important results; the gain is not automatic code generation or a replacement for the developer's judgment.

### Source excerpt

The most interesting change AI brings to software work is not faster coding. It is the expansion of context or range... what I call radius. For decades the job was defined by the limits of what one person could hold in their head. You tended to specialize. You got good at one or two systems, ... Continue reading Cognitive radius

## Reading JSON at 277 GB/s in Python

DevFeed: [Reading JSON at 277 GB/s in Python](<https://devfeed.tech/articles/reading-json-at-277-gb-s-in-python-134077.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/11/14/reading-json-at-277-gb-s-in-python/>)

Author: Daniel Lemire

Published: 2026-11-14T14:53:50Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [json lines format](<https://devfeed.tech/topics/json-lines-format.md>), [Python](<https://devfeed.tech/topics/python.md>), [JSON](<https://devfeed.tech/topics/json.md>)

Tags: [json](<https://devfeed.tech/tags/json.md>), [python](<https://devfeed.tech/tags/python.md>), [reading](<https://devfeed.tech/tags/reading.md>)

### AI overview

In the reported benchmarks, fastsimdjson combines selective field extraction with threaded parsing to reach 138 GB/s when counting GitHub stars across 7.6 GB of NDJSON on a 192-core AWS instance; extracting fields from a Twitter benchmark file reaches 277 GB/s. On the counting task, it outpaces the best msgspec process pool (116 GB/s at 96 processes), with the advantage narrowing from 3.5x on one core to about 20% at scale. Across six benchmark files, four exceed 100 GB/s, while one is near 20 GB/s and another ties msgspec. The article notes counting is especially favorable to processes because little data must be returned.

### Source excerpt

Many systems log their data as JSON, one record per line (NDJSON). GitHub, for example, publishes every public event on its site that way through GH Archive: about a gigabyte per hour. Suppose you want to know which repositories received the most stars over eight hours. That is 7.6 GB of JSON, 2.2 million events. ... Continue reading Reading JSON at 277 GB/s in Python

## Telling your number parser what to expect

DevFeed: [Telling your number parser what to expect](<https://devfeed.tech/articles/telling-your-number-parser-what-to-expect-126410.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/11/10/telling-your-number-parser-what-to-expect/>)

Author: Daniel Lemire

Published: 2026-11-10T12:44:09Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Parser](<https://devfeed.tech/topics/parser.md>), [floating-point](<https://devfeed.tech/topics/floating-point.md>)

Tags: [c-plus-plus](<https://devfeed.tech/tags/c-plus-plus.md>), [fast](<https://devfeed.tech/tags/fast.md>), [floating-point](<https://devfeed.tech/tags/floating-point.md>), [gcc](<https://devfeed.tech/tags/gcc.md>), [library](<https://devfeed.tech/tags/library.md>), [number](<https://devfeed.tech/tags/number.md>)

### AI overview

fast_float adds three optional performance hints that let callers describe their input: `unpredictable_sign` removes a branch for randomly ordered signs; `short_numbers` handles standalone numbers of at most eight characters with bitwise operations; and `assume_round_to_nearest` skips checking the rounding mode. They trade extra instructions for fewer branch mispredictions, so the best choice depends on data and processor. In the reported tests, random-sign short prices benefited from sign and rounding hints on Intel (73 to 105 million numbers/s) and all three hints on Apple (153 to 345 million). Longer random-sign balances benefited from sign and rounding hints; varied positive sensor readings benefited from `short_numbers`. For regular, consistently shaped prices, Intel was fastest without hints, while Apple still benefited from `short_numbers`. The rounding hint can produce wrong results if the program changes rounding mode despite the promise. The hints also work with `from_chars_advanced` and `integer_times_pow10`; the excerpt says they are in a pull request.

### Source excerpt

Our fast_float library parses numbers such as -731.51 into binary floating-point values. It is part of GCC's C++ standard library, among others. It is fast: often a few nanoseconds per number. Parsing a number means making decisions. Is there a sign? How many digits are there? Where is the decimal point? Is the processor rounding ... Continue reading Telling your number parser what to expect

## A performance trap: when store forwarding fails

DevFeed: [A performance trap: when store forwarding fails](<https://devfeed.tech/articles/a-performance-trap-when-store-forwarding-fails-118940.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/11/05/a-performance-trap-when-store-forwarding-fails/>)

Author: Daniel Lemire

Published: 2026-11-05T13:06:20Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Intel Core](<https://devfeed.tech/topics/intel-core.md>), [Algorithms](<https://devfeed.tech/topics/algorithms.md>)

Tags: [buffer](<https://devfeed.tech/tags/buffer.md>), [c](<https://devfeed.tech/tags/c.md>), [cache](<https://devfeed.tech/tags/cache.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [cycles](<https://devfeed.tech/tags/cycles.md>), [epyc](<https://devfeed.tech/tags/epyc.md>), [example](<https://devfeed.tech/tags/example.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [failure](<https://devfeed.tech/tags/failure.md>), [graviton](<https://devfeed.tech/tags/graviton.md>), [load](<https://devfeed.tech/tags/load.md>), [overlap](<https://devfeed.tech/tags/overlap.md>), [performance](<https://devfeed.tech/tags/performance.md>), [processor](<https://devfeed.tech/tags/processor.md>), [reading](<https://devfeed.tech/tags/reading.md>), [reload](<https://devfeed.tech/tags/reload.md>), [size](<https://devfeed.tech/tags/size.md>), [store](<https://devfeed.tech/tags/store.md>), [update](<https://devfeed.tech/tags/update.md>), [write](<https://devfeed.tech/tags/write.md>)

### AI overview

Store forwarding lets a later load obtain bytes from a store buffer before the store reaches cache, but forwarding fails when the load is wider than the store or the accessed ranges only partly overlap; the load then waits for the store to reach cache. The excerpt's C example overwrites one byte and reloads eight, and notes the same issue for a 4-byte store followed by an 8-byte load or an 8-byte load shifted four bytes from an 8-byte store. It shows avoiding the mismatch by loading eight bytes into a register, modifying the byte there, then storing all eight back. In the reported experiments, matching store/load operations took one cycle on Xeons and EPYC; wider loads took 19 cycles on Xeons, 14 on EPYC, and 5 on M4 and Graviton 5. Two stores took 8 cycles on M4. A compiler may remove a reload when the same eight bytes are stored and read back.

### Source excerpt

A store does not go straight into the cache. Waiting for that write would stall the processor. It sits in a store buffer, and a later load can take its bytes from that buffer. That is store forwarding. Forwarding works when the load is the same size as the store, or smaller and completely inside ... Continue reading A performance trap: when store forwarding fails

## Open source as a research method

DevFeed: [Open source as a research method](<https://devfeed.tech/articles/open-source-as-a-research-method-116293.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/11/04/open-source-as-a-research-method/>)

Author: Daniel Lemire

Published: 2026-11-04T12:48:24Z

Content type: opinion

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Open Source](<https://devfeed.tech/topics/open-source.md>), [Node.js](<https://devfeed.tech/topics/node-js.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-tools](<https://devfeed.tech/tags/ai-tools.md>), [algorithm](<https://devfeed.tech/tags/algorithm.md>), [browsers](<https://devfeed.tech/tags/browsers.md>), [bug](<https://devfeed.tech/tags/bug.md>), [bug-reports](<https://devfeed.tech/tags/bug-reports.md>), [bun](<https://devfeed.tech/tags/bun.md>), [c-plus-plus](<https://devfeed.tech/tags/c-plus-plus.md>), [code-reviews](<https://devfeed.tech/tags/code-reviews.md>), [deno](<https://devfeed.tech/tags/deno.md>), [github](<https://devfeed.tech/tags/github.md>), [go](<https://devfeed.tech/tags/go.md>), [net](<https://devfeed.tech/tags/net.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [research](<https://devfeed.tech/tags/research.md>)

### AI overview

The author argues that open source can make building a research method: working software attracts real users whose bugs, unusual inputs, hardware and workloads expose research problems. The article traces this path through simdjson, whose JSON parsing work led to faster floating-point conversion via the Eisel-Lemire algorithm and then to simdutf for UTF-8 validation and transcoding. The author says those projects reached production use and that user contributions and code reviews provide demanding feedback. They also report that AI tools appear to make it easier for people to understand unfamiliar code and contribute fixes. These are the author's observations and examples, not a general comparison of research methods.

### Source excerpt

Many people still picture innovation as a line. The professor thinks and publishes. The engineer reads the paper and designs a product. The customer consumes it. It rarely works that way. The industrial revolution was not set off by professors writing papers. It was driven by demand for cheap and better clothing for the masses. ... Continue reading Open source as a research method

## In C++, = default and {} are not the same constructor

DevFeed: [In C++, = default and {} are not the same constructor](<https://devfeed.tech/articles/in-c-default-and-are-not-the-same-constructor-113364.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/11/02/default-and-braces-are-not-the-same-constructor/>)

Author: Daniel Lemire

Published: 2026-11-02T12:48:59Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [constructor](<https://devfeed.tech/topics/constructor.md>), [c/c++](<https://devfeed.tech/topics/c-c-plus-plus.md>), [clang](<https://devfeed.tech/topics/clang.md>)

Tags: [c-plus-plus](<https://devfeed.tech/tags/c-plus-plus.md>), [constructor](<https://devfeed.tech/tags/constructor.md>), [gcc](<https://devfeed.tech/tags/gcc.md>), [intel-xeon](<https://devfeed.tech/tags/intel-xeon.md>), [november-2026](<https://devfeed.tech/tags/november-2026.md>), [performance](<https://devfeed.tech/tags/performance.md>), [spans](<https://devfeed.tech/tags/spans.md>), [struct](<https://devfeed.tech/tags/struct.md>)

### AI overview

An empty `span() = default` and `span() {}` can produce different performance because only the first is trivial: the explicitly defaulted constructor is trivial, while the empty body is user-provided. In the shown `box()` path, GCC 16 with `-O2 -march=x86-64` generates direct field stores for `span() {}`, but uses a temporary and copies 16 bytes for `span() = default`. The author reports that in timing tests of 100 million spans on one Intel Xeon Gold 6548N core, the `= default` version was over four times slower with both GCC 15 and GCC 16. The result is tied to the described code, compiler settings, and test setup.

### Source excerpt

You can write an empty default constructor as span() = default or as span() {}. It can have remarkable effects on performance. struct span { const char *p; int len; const char *end; span() = default; // or span() {} span(const char *s, int n) : p(s), len(n), end(s + n) {} }; The constructor ... Continue reading In C++, = default and {} are not the same constructor

## Beating the compiler at signed 32-bit division

DevFeed: [Beating the compiler at signed 32-bit division](<https://devfeed.tech/articles/beating-the-compiler-at-signed-32-bit-division-110587.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/31/beating-the-compiler-at-signed-32-bit-division/>)

Author: Daniel Lemire

Published: 2026-10-31T13:05:35Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>), [Compiler](<https://devfeed.tech/topics/compiler.md>), [LLVM](<https://devfeed.tech/topics/llvm.md>), [clang](<https://devfeed.tech/topics/clang.md>), [vectorization techniques](<https://devfeed.tech/topics/vectorization-techniques.md>)

Tags: [32-bit](<https://devfeed.tech/tags/32-bit.md>), [64-bit](<https://devfeed.tech/tags/64-bit.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [apple](<https://devfeed.tech/tags/apple.md>), [arm](<https://devfeed.tech/tags/arm.md>), [arrow](<https://devfeed.tech/tags/arrow.md>), [autovectorization](<https://devfeed.tech/tags/autovectorization.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [branch](<https://devfeed.tech/tags/branch.md>), [c](<https://devfeed.tech/tags/c.md>), [clang](<https://devfeed.tech/tags/clang.md>), [code](<https://devfeed.tech/tags/code.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [compilers](<https://devfeed.tech/tags/compilers.md>), [cycles](<https://devfeed.tech/tags/cycles.md>), [gcc](<https://devfeed.tech/tags/gcc.md>), [latency](<https://devfeed.tech/tags/latency.md>), [performance](<https://devfeed.tech/tags/performance.md>), [processor](<https://devfeed.tech/tags/processor.md>), [zen-5](<https://devfeed.tech/tags/zen-5.md>)

### AI overview

A signed 32-bit division optimization can use an unsigned 33-bit reciprocal multiplier on 64-bit processors, avoiding the compiler's correction step when that multiplier does not fit a signed 32-bit value. The article gives a `div32` implementation for positive, non-power-of-two constant divisors and reports lower dependency-chain latency for divisors 7, 60, and 365; throughput gains vary by compiler and processor. For divisor 1000, where the compiler's multiplier fits signed 32 bits, the new method is slightly slower on Apple M4, so the author recommends using it only when correction is needed. For runtime divisors, a revised `fastmod` signed multiplication removes a sign-dependent branch: in the reported random-sign test it is over twice as fast as hardware division on Intel and enables partial loop vectorization on M4 when autovectorization is allowed.

### Source excerpt

Dividing integers is slow on our processors. A 32-bit division costs: processor latency throughput Emerald Rapids about 12 cycles one every 6 cycles Zen 4 9 to 12 cycles one every 6 cycles Zen 5 10 to 13 cycles one every 6 cycles Arrow Lake performance core 12 cycles one every 6 cycles Apple M4 ... Continue reading Beating the compiler at signed 32-bit division

## Billion-query-per-second lookup tables in C++

DevFeed: [Billion-query-per-second lookup tables in C++](<https://devfeed.tech/articles/billion-query-per-second-lookup-tables-in-c-101123.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/26/billion-query-per-second-lookup-tables-in-c/>)

Author: Daniel Lemire

Published: 2026-10-26T18:23:08Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [stl](<https://devfeed.tech/topics/stl.md>), [clang](<https://devfeed.tech/topics/clang.md>), [Parser](<https://devfeed.tech/topics/parser.md>)

Tags: [abseil](<https://devfeed.tech/tags/abseil.md>), [algorithm](<https://devfeed.tech/tags/algorithm.md>), [apple](<https://devfeed.tech/tags/apple.md>), [arm](<https://devfeed.tech/tags/arm.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [branch](<https://devfeed.tech/tags/branch.md>), [buffer](<https://devfeed.tech/tags/buffer.md>), [c-plus-plus](<https://devfeed.tech/tags/c-plus-plus.md>), [c-plus-plus-23](<https://devfeed.tech/tags/c-plus-plus-23.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [lookup](<https://devfeed.tech/tags/lookup.md>), [mapping](<https://devfeed.tech/tags/mapping.md>), [tables](<https://devfeed.tech/tags/tables.md>)

### AI overview

For fixed string keys, the described C++23 library builds a perfect hash at compile time, avoiding runtime collision handling. It chooses association values greedily for selected key characters; lookup computes a slot and verifies the stored key. The implementation also uses a 16-byte vector comparison, with a slower path when a wide load could cross a memory-page boundary. On an Apple M3 Max, the article reports about 1.3 ns per lookup for 20 HTTP headers, versus 7.7 ns for gperf, 10.9 ns for Abseil flat_hash_map, and 16.6 ns for std::unordered_map. Across other tested key sets it reports 1.2-2 ns per lookup; results are specific to the stated benchmarks and hardware.

### Source excerpt

A lot of software spends its time mapping strings to values. A web server must identify the HTTP method (GET, POST, ...). It must recognize header names. A URL parser must recognize the protocol (http, https, ftp, ...). A compiler must recognize keywords. In C++, the default approach is to put the strings in a ... Continue reading Billion-query-per-second lookup tables in C++

## Decoding base64 in Python at 10 gigabytes per second

DevFeed: [Decoding base64 in Python at 10 gigabytes per second](<https://devfeed.tech/articles/decoding-base64-in-python-at-10-gigabytes-per-second-89214.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/17/decoding-base64-in-python-at-10-gigabytes-per-second/>)

Author: Daniel Lemire

Published: 2026-10-17T04:40:34Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Python](<https://devfeed.tech/topics/python.md>), [Decoding](<https://devfeed.tech/topics/decoding.md>)

Tags: [base64](<https://devfeed.tech/tags/base64.md>), [encoding](<https://devfeed.tech/tags/encoding.md>), [python](<https://devfeed.tech/tags/python.md>)

### AI overview

The post introduces fastpybase64, a Python drop-in replacement for the standard base64 module built on simdutf. In a single-thread benchmark using MIME-style line breaks, it reports decoding four times faster than pybase64 and 27 times faster than Python's standard library. The package is available for Python 3.10 through 3.14 on Linux, macOS, and Windows.

### Source excerpt

We sometimes need to convert binary data to text. For example, your emails must be stored as text files. Naïvely, you could just use a string of 0 and 1, but it would mean using one character per bit. We can do better by using 64 ASCII characters: all letters, all numbers, plus a few ... Continue reading Decoding base64 in Python at 10 gigabytes per second

## Redis now has compressed bitmaps (Roaring)

DevFeed: [Redis now has compressed bitmaps (Roaring)](<https://devfeed.tech/articles/redis-now-has-compressed-bitmaps-roaring-88191.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/15/redis-now-has-compressed-bitmaps-roaring/>)

Author: Daniel Lemire

Published: 2026-10-15T22:46:54Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Redis](<https://devfeed.tech/topics/redis.md>), [Plover](<https://devfeed.tech/topics/plover.md>)

Tags: [apache-spark](<https://devfeed.tech/tags/apache-spark.md>), [cache](<https://devfeed.tech/tags/cache.md>), [clickhouse](<https://devfeed.tech/tags/clickhouse.md>), [database](<https://devfeed.tech/tags/database.md>), [in-memory-database](<https://devfeed.tech/tags/in-memory-database.md>), [intel-xeon](<https://devfeed.tech/tags/intel-xeon.md>), [redis](<https://devfeed.tech/tags/redis.md>), [roaring](<https://devfeed.tech/tags/roaring.md>), [server](<https://devfeed.tech/tags/server.md>)

### AI overview

Redis is adding opt-in Roaring compressed bitmaps, planned for Redis 8.12. The article compares them with plain string-backed bitmaps and reports that Roaring can be much faster for some workloads, especially sparse hashed IDs, while noting that compressed bitmaps involve tradeoffs.

### Source excerpt

Redis is an in-memory database. Salvatore Sanfilippo started it in 2009, and it is now one of the most widely deployed pieces of server software. It is often used as a cache in front of a slower database. You talk to it over the network with short text commands such as SET user:42 "Alice" and ... Continue reading Redis now has compressed bitmaps (Roaring)

## Parsing JSON at tens of gigabytes per second... in Python

DevFeed: [Parsing JSON at tens of gigabytes per second... in Python](<https://devfeed.tech/articles/parsing-json-at-tens-of-gigabytes-per-second-in-python-85593.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/12/parsing-json-at-tens-of-gigabytes-per-second-in-python/>)

Author: Daniel Lemire

Published: 2026-10-12T13:06:39Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [JSON](<https://devfeed.tech/topics/json.md>), [json lines format](<https://devfeed.tech/topics/json-lines-format.md>), [Python](<https://devfeed.tech/topics/python.md>), [Scapy](<https://devfeed.tech/topics/scapy.md>), [Numba](<https://devfeed.tech/topics/numba.md>)

Tags: [json](<https://devfeed.tech/tags/json.md>), [parallelism](<https://devfeed.tech/tags/parallelism.md>), [python](<https://devfeed.tech/tags/python.md>), [python-3-14](<https://devfeed.tech/tags/python-3-14.md>), [thread](<https://devfeed.tech/tags/thread.md>)

### AI overview

The article benchmarks parsing 7.6 GB of GitHub event data with Python 3.14's free-threaded build. Using fastsimdjson with lazy views and 64 threads, it processes the data in about 0.12 seconds, or 66 GB/s, and compares the result with other JSON parsers.

### Source excerpt

Python used to have limited parallelism, due to its global interpreter lock (GIL). Only one thread could run Python code at a time. Python 3.14 comes in a free-threaded version (python3.14t), without the global interpreter lock. Several threads can now run Python code at the same time. Let us pick a realistic problem. GitHub publishes ... Continue reading Parsing JSON at tens of gigabytes per second... in Python

## WebSocket servers in JavaScript: Node.js, Bun and Deno in 2026

DevFeed: [WebSocket servers in JavaScript: Node.js, Bun and Deno in 2026](<https://devfeed.tech/articles/websocket-servers-in-javascript-node-js-bun-and-deno-in-2026-84487.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/10/websocket-servers-in-javascript-node-js-bun-and-deno-in-2026/>)

Author: Daniel Lemire

Published: 2026-10-10T08:56:11Z

Content type: comparison

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [WebSocket](<https://devfeed.tech/topics/websocket.md>), [Bun](<https://devfeed.tech/topics/bun.md>), [Deno](<https://devfeed.tech/topics/deno.md>), [Node.js](<https://devfeed.tech/topics/node-js.md>)

Tags: [bun](<https://devfeed.tech/tags/bun.md>), [deno](<https://devfeed.tech/tags/deno.md>), [node-js](<https://devfeed.tech/tags/node-js.md>), [websocket](<https://devfeed.tech/tags/websocket.md>)

### AI overview

This benchmark compares WebSocket server performance using Node.js with uWebSockets.js, Bun's built-in server, and Deno's built-in server. In a single-core ping-pong test with C clients, Node.js reached 49,000 round trips per second, compared with 40,000 for Deno and 37,000 for Bun; the reported 99th-percentile times were 24.8, 32.0, and 46.4 microseconds, respectively.

### Source excerpt

WebSocket is a standard protocol to hook up a client and a server in a persistent manner. It is commonly used for live chat, notifications, multiplayer games, collaborative editing, and streaming dashboards. Many servers are written in JavaScript and use Node.js, Bun or Deno as an engine. A few years ago, I wrote a simple ... Continue reading WebSocket servers in JavaScript: Node.js, Bun and Deno in 2026

## Faster Python startup with lazy imports

DevFeed: [Faster Python startup with lazy imports](<https://devfeed.tech/articles/faster-python-startup-with-lazy-imports-83944.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/09/faster-python-startup-with-lazy-imports/>)

Author: Daniel Lemire

Published: 2026-10-09T03:15:06Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Python](<https://devfeed.tech/topics/python.md>)

Tags: [dependencies](<https://devfeed.tech/tags/dependencies.md>), [lazy](<https://devfeed.tech/tags/lazy.md>), [lazy-import](<https://devfeed.tech/tags/lazy-import.md>), [python](<https://devfeed.tech/tags/python.md>), [python-3-15](<https://devfeed.tech/tags/python-3-15.md>)

### AI overview

Python's upcoming lazy imports defer loading dependencies until code uses them. In a command-line tool benchmark, lazy imports reduced version-command startup from 295 ms to 20 ms and made the mean command twelve times faster. The tradeoff is that missing-module errors may appear later, at first use.

### Source excerpt

When a Python program starts, it needs to load all its dependencies (import). With the upcoming version of Python, it is possible to use lazy imports instead. lazy import json lazy from decimal import Decimal In this instance, the name json is no longer the module, but a mere placeholder object. The actual import only ... Continue reading Faster Python startup with lazy imports

## Mastering SIMD with Java Vector API

DevFeed: [Mastering SIMD with Java Vector API](<https://devfeed.tech/articles/mastering-simd-with-java-vector-api-66578.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/07/mastering-simd-with-java-vector-api/>)

Author: Daniel Lemire

Published: 2026-10-07T02:07:05Z

Content type: opinion

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [vectorization techniques](<https://devfeed.tech/topics/vectorization-techniques.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>), [Java HttpClient](<https://devfeed.tech/topics/java-httpclient.md>)

Tags: [hardware](<https://devfeed.tech/tags/hardware.md>), [java](<https://devfeed.tech/tags/java.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [performance](<https://devfeed.tech/tags/performance.md>), [simd](<https://devfeed.tech/tags/simd.md>)

### AI overview

The article reviews Roman Snytsar's book on Java's Vector API for SIMD optimization. It describes how the book rebuilds common programming tasks with the API, benchmarks the results, examines assembly, and discusses cases where data parallelism helps or fails. The author recommends it to Java programmers focused on performance.

### Source excerpt

I write a lot about data parallelism. About how our processors have special instructions that can do several operations at once. They are often critical to get the best performance out of a CPU. Compilers are sometimes able to take advantage of these instructions. But it is hard to tell when it will work. For ... Continue reading Mastering SIMD with Java Vector API

## Faster software linking with mold

DevFeed: [Faster software linking with mold](<https://devfeed.tech/articles/faster-software-linking-with-mold-65378.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/06/linking-node-js-with-mold/>)

Author: Daniel Lemire

Published: 2026-10-06T08:00:20Z

Content type: comparison

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [linker](<https://devfeed.tech/topics/linker.md>)

Tags: [gcc](<https://devfeed.tech/tags/gcc.md>), [linux](<https://devfeed.tech/tags/linux.md>), [multithreading](<https://devfeed.tech/tags/multithreading.md>), [node-js](<https://devfeed.tech/tags/node-js.md>)

### AI overview

A Node.js linking benchmark on a Linux server reports that multithreaded mold is about 24 times faster than GNU ld and 14 times faster than gold. Results use median timings from six runs per linker. The article explains how to enable mold through compiler flags or a build wrapper and why faster linking benefits developers who rebuild frequently.

### Source excerpt

When you build a program, the compiler turns each source file into an object file. Then a linker stitches all the object files and libraries into one executable. On Linux, the default linker is usually GNU ld (also called bfd). The GNU binutils also ship gold, an alternative linker that was designed to be faster. ... Continue reading Faster software linking with mold

## Ephemeral testing

DevFeed: [Ephemeral testing](<https://devfeed.tech/articles/ephemeral-testing-64786.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/05/ephemeral-testing/>)

Author: Daniel Lemire

Published: 2026-10-05T08:00:51Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Integration testing](<https://devfeed.tech/topics/integration-testing.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [Software Engineering](<https://devfeed.tech/topics/software-engineering.md>), [principle of ephemerality](<https://devfeed.tech/topics/principle-of-ephemerality.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [integration-testing](<https://devfeed.tech/tags/integration-testing.md>), [software](<https://devfeed.tech/tags/software.md>), [testing](<https://devfeed.tech/tags/testing.md>), [unit-testing](<https://devfeed.tech/tags/unit-testing.md>)

### AI overview

The article proposes ephemeral testing: use an AI agent to build and test a temporary application or additional layer on top of a software component. How well that software works can reveal issues in the component's API, invariants, defaults, errors, and documentation. The author describes repeating the approach with different agents and tasks to evaluate a foundation while prototyping possible features.

### Source excerpt

We have many ways to ensure software quality. Unit testing. Fuzz testing. Integration testing. And so forth. I'd like to propose a method that was unthinkable before: ephemeral testing. (Ephemeral is a fancy word for 'throw away' or 'temporary'.) You write your code. You build your software component. Or the AI agent does ... Continue reading Ephemeral testing

## Research paper overload: submissions capped at two a month

DevFeed: [Research paper overload: submissions capped at two a month](<https://devfeed.tech/articles/research-paper-overload-submissions-capped-at-two-a-month-64647.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/04/arxiv-capped-submissions-at-two-a-month/>)

Author: Daniel Lemire

Published: 2026-10-04T17:07:27Z

Content type: opinion

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Computer science](<https://devfeed.tech/topics/computer-science.md>)

Tags: [paper](<https://devfeed.tech/tags/paper.md>), [research](<https://devfeed.tech/tags/research.md>)

### AI overview

The author describes arXiv's monthly submission cap of two papers per person, introduced after submissions exceeded 40,000 in September 2026. The post says submissions had doubled over two years and argues that the volume is unsustainable for human moderators, drawing a comparison to Google stopping new bug reports in its open-source bounty program.

### Source excerpt

Much of my research is on arXiv. I was one of the early adopters. It is simply a repository of research papers with PDF and metadata. It makes it convenient to find research without any paywall. Physicists started it in 1991. Math and computer science followed gradually. It relies on volunteers to moderate it, because ... Continue reading Research paper overload: submissions capped at two a month

## Google's first orbital data center is in orbit

DevFeed: [Google's first orbital data center is in orbit](<https://devfeed.tech/articles/google-s-first-orbital-data-center-is-in-orbit-64648.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/04/googles-first-orbital-data-center-is-in-orbit/>)

Author: Daniel Lemire

Published: 2026-10-04T16:49:29Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [clean energy data centers](<https://devfeed.tech/topics/clean-energy-data-centers.md>)

Tags: [data-center](<https://devfeed.tech/tags/data-center.md>), [google](<https://devfeed.tech/tags/google.md>), [power](<https://devfeed.tech/tags/power.md>), [solar](<https://devfeed.tech/tags/solar.md>)

### AI overview

Google's first orbital data center is described as a fridge-sized facility with one kilowatt of solar power, launched aboard a Falcon 9 and expected to operate for about a year. Google plans two more orbital data centers next year. The article argues that proximity and line of sight could make latency less of a barrier for compute workloads, while space offers room and continuous solar power without local permitting or electricity negotiations.

### Source excerpt

Google's first orbital data center is in orbit (October 1st). They say it is about the size of a fridge. It has one kilowatt of solar power. It is about the power of a household. It was launched on a Falcon 9 rocket (SpaceX). It should run for about a year. Google plans to launch ... Continue reading Google's first orbital data center is in orbit

## Parsing compressed JSON at 40 GB/s

DevFeed: [Parsing compressed JSON at 40 GB/s](<https://devfeed.tech/articles/parsing-compressed-json-at-40-gb-s-63108.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/10/01/parsing-compressed-json-at-40-gb-s/>)

Author: Daniel Lemire

Published: 2026-10-01T03:10:29Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [JSON](<https://devfeed.tech/topics/json.md>), [JSON Feed](<https://devfeed.tech/topics/json-feed.md>), [IO](<https://devfeed.tech/topics/io.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [better](<https://devfeed.tech/tags/better.md>), [data](<https://devfeed.tech/tags/data.md>), [database](<https://devfeed.tech/tags/database.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [gcc](<https://devfeed.tech/tags/gcc.md>), [gzip](<https://devfeed.tech/tags/gzip.md>), [intel](<https://devfeed.tech/tags/intel.md>), [intel-xeon](<https://devfeed.tech/tags/intel-xeon.md>), [json](<https://devfeed.tech/tags/json.md>), [parsing](<https://devfeed.tech/tags/parsing.md>)

### AI overview

The article describes a simdjson demo that parses compressed NDJSON in parallel. Because gzip is a single stream, decompression is sequential, while zstd and lz4 files made of independent frames allow threads to decompress and parse frames independently. In a benchmark using synthetic, repetitive data, the author reports 40 GB/s with 64 threads for zstd and 34 GB/s for lz4, versus 2.5 GB/s for gzip with pipelined parsing; the zstd file was 6% larger than the gzip file. Results may differ on other data.

### Source excerpt

A common way to store JSON data is to write one document per line. We call it NDJSON or JSON Lines. Log files, database exports and machine-learning datasets often come in this format. The files can be large, so we may compress them. {"id":1,"active":true,"user":{"name":"user_1","tags":["guest"]},"score":-3862,"note":"..."} {"id":2,"active":false,"user":{"name":"user_2","tags":["staff","admin"]},"score":8123,"note":"..."} How fast can you read such a file? I wrote ... Continue reading Parsing compressed JSON at 40 GB/s

## Transcoding UTF-8 to UTF-16 with replacement at gigabytes per second

DevFeed: [Transcoding UTF-8 to UTF-16 with replacement at gigabytes per second](<https://devfeed.tech/articles/transcoding-utf-8-to-utf-16-with-replacement-at-gigabytes-per-second-62841.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/09/30/transcoding-utf-8-to-utf-16-with-replacement-at-gigabytes-per-second/>)

Author: Daniel Lemire

Published: 2026-09-30T12:33:27Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Encoding](<https://devfeed.tech/topics/encoding.md>), [OpenCV](<https://devfeed.tech/topics/opencv.md>)

Tags: [c-sharp](<https://devfeed.tech/tags/c-sharp.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [engine](<https://devfeed.tech/tags/engine.md>), [gcc](<https://devfeed.tech/tags/gcc.md>), [icelake](<https://devfeed.tech/tags/icelake.md>), [java](<https://devfeed.tech/tags/java.md>), [javascript](<https://devfeed.tech/tags/javascript.md>), [validation](<https://devfeed.tech/tags/validation.md>), [xeon](<https://devfeed.tech/tags/xeon.md>)

### AI overview

The simdutf library adds a function for transcoding invalid UTF-8 input to UTF-16, replacing ill-formed sequences with U+FFFD. The method first validates the input and records error offsets, then transcodes the valid spans. In a reported test on one Xeon Gold 6548N core, simdutf reached 7.0 GB/s on valid input, compared with 1.7 GB/s for a scalar decoder.

### Source excerpt

Our software represents strings using the UTF-16 or the UTF-8 formats. Most text on the web is UTF-8, but Java, C# or JavaScript represents the strings as UTF-16 to the programmer. Sometimes we need to transcode (convert) strings. Your browser probably uses the simdutf library for validating or transcoding. It is part of the widely ... Continue reading Transcoding UTF-8 to UTF-16 with replacement at gigabytes per second

## simdjson 5.0 is out

DevFeed: [simdjson 5.0 is out](<https://devfeed.tech/articles/simdjson-5-0-is-out-61355.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/09/28/simdjson-5-0-is-out/>)

Author: Daniel Lemire

Published: 2026-09-28T12:38:25Z

Content type: release

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [JSON](<https://devfeed.tech/topics/json.md>), [Library](<https://devfeed.tech/topics/library.md>), [static linking](<https://devfeed.tech/topics/static-linking.md>), [Linux Kernel](<https://devfeed.tech/topics/linux-kernel.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>)

Tags: [apache](<https://devfeed.tech/tags/apache.md>), [c-plus-plus](<https://devfeed.tech/tags/c-plus-plus.md>), [c-plus-plus-26](<https://devfeed.tech/tags/c-plus-plus-26.md>), [clickhouse](<https://devfeed.tech/tags/clickhouse.md>), [gcc](<https://devfeed.tech/tags/gcc.md>), [js](<https://devfeed.tech/tags/js.md>), [json](<https://devfeed.tech/tags/json.md>), [ladybird](<https://devfeed.tech/tags/ladybird.md>), [meta](<https://devfeed.tech/tags/meta.md>), [node-js](<https://devfeed.tech/tags/node-js.md>)

### AI overview

simdjson 5.0 adds default key-selector deserialization for reflected C++ structures, improved handling of large integers, JSON text sequences and comma-separated document streams, and options for line-based and threaded parsing. It also includes bug fixes following an audit.

### Source excerpt

The simdjson library is a C++ library to parse and generate JSON. It is used in Node.js, ClickHouse, Meta Velox, StarRocks, Apache Doris, the Ladybird browser and many other systems. We released version 4.0 a year ago, in September 2025. Its headline feature was C++26 static reflection: you could turn a C++ structure into JSON, ... Continue reading simdjson 5.0 is out

## How fast can you fix a UTF-16 string in C#?

DevFeed: [How fast can you fix a UTF-16 string in C#?](<https://devfeed.tech/articles/how-fast-can-you-fix-a-utf-16-string-in-c-60465.md>)

Original publisher: [Read original article](<https://lemire.me/blog/2026/09/26/how-fast-can-you-fix-a-utf-16-string-in-c/>)

Author: Daniel Lemire

Published: 2026-09-26T16:53:53Z

Content type: article

Language: en

Sources: [Daniel Lemire](<https://devfeed.tech/sources/daniel-lemire.md>)

Topics: [Encoding](<https://devfeed.tech/topics/encoding.md>), [version](<https://devfeed.tech/topics/version.md>), [Python](<https://devfeed.tech/topics/python.md>)

Tags: [avx](<https://devfeed.tech/tags/avx.md>), [c-sharp](<https://devfeed.tech/tags/c-sharp.md>), [chips](<https://devfeed.tech/tags/chips.md>), [code](<https://devfeed.tech/tags/code.md>), [intel](<https://devfeed.tech/tags/intel.md>), [intel-xeon](<https://devfeed.tech/tags/intel-xeon.md>), [linux](<https://devfeed.tech/tags/linux.md>), [macos](<https://devfeed.tech/tags/macos.md>), [parallelism](<https://devfeed.tech/tags/parallelism.md>), [performance](<https://devfeed.tech/tags/performance.md>)

### AI overview

The article describes SIMD-based C# functions in the SimdUnicode library that repair ill-formed UTF-16 strings by replacing lone surrogates with U+FFFD and validate strings for such surrogates. It reports throughput comparisons on an Intel Xeon and Apple M4 Max, including that repair of well-formed strings is roughly as fast as a memory copy.

### Source excerpt

C# strings are UTF-16. Most characters are one 16-bit code unit. Characters outside the basic multilingual plane, emoji included, take two: a high surrogate (U+D800 to U+DBFF) followed by a low surrogate (U+DC00 to U+DFFF). A surrogate with the wrong neighbor, or with none, is ill-formed. You should never send an ill-formed string to disk ... Continue reading How fast can you fix a UTF-16 string in C#?

[Next page](<https://devfeed.tech/sources/daniel-lemire.md?cursor=WyIyMDI2LTA5LTI2VDE2OjUzOjUzKzAwOjAwIiwgImU5NWU1YTBkLTM4NjItNDc1NC1hODFhLTk2OTMxYTVmODBkNCJd>)