[email protected]

@gengjiawen 网评不如Claude Code

@HaskRay 你咋不用 codex

百码归模，我看还是要归。有些同志认为，只能投喂精品代码，不能投喂屎山。这种看法，表明他们对数据飞轮的方针很不理解。一般说来，纯粹的乱码当然不让喂。但是，它不以乱码的面貌出现，而以"经过review的生产代码"的面貌出现，那就只好让它喂，这样才有利于模型对它进行鉴别和泛化。……

All rivers flow to the sea — and flow they shall. Some comrades believe we should only feed the model clean, elegant code, and never feed it legacy garbage. This view betrays a fundamental misunderstanding of the data flywheel doctrine...

Revised maskray.me/blog/2025-08-2… on LLVM's recent .prefalign change and filed a GNU Assembler feature request sourceware.org/bugzilla/show_…

maskray.me/blog/2026-02-2… Bit-field layout

🐱🐱

maskray.me/blog/2026-02-1… Call relocation types Some architectures use two ELF relocation types for a call instruction: (x86, m68k, s390(x), etc)

@djtodoro Have you considered maintaining this downstream? Big-endian here introduces either dynamic dispatch overhead or increased code size for the overwhelming majority of RISC-V usersendian. The code is relatively compact, the cost of carrying this out-of-tree should be quite small.

@HaskRay It would be nice if you can give a review on that one 🤣

github.com/llvm/llvm-proj… "[LLD][ELF][RISCV] Support big-endian RISC-V linking" This reminds me of Linus Torvalds's @mail.gmail.com/t/#mce138059dc56014643bbda330810183031ef5c06" target="_blank" rel="nofollow noopener">lore.kernel.org/lkml/CAHk-%3Dw… "New endianness problems are somebody ELSES problem"

@djtodoro Many people, including very high-profile ones, dislike this idea. The question is "does big-endian RISC-V earn its keep." Without a concrete user, shipping hardware, or at minimum a credible public roadmap, this imposes ongoing costs on maintainers for speculative benefit.

LLM-aided review may help at the margins — catching mechanical issues, summarizing patches — but the core bottleneck is human judgment.

For a project like LLVM, open contribution is not optional — so you're really choosing between quality and velocity. LLM-aided development dramatically increases contribution volume without increasing reviewer capacity.

Quality, Velocity, Open Contribution — pick two. If you try for all three, you get none — the maintainers burn out, the project becomes unsustainable. Lua and SQLite picked quality, and dropped both velocity and open contribution.

@rui314 harshanu.space/en/tech/ccc-vs… analyzes why the ccc-compiled code has poor runtime performance.

Claudeで自動でCコンパイラが書けたというブログを見て、Anthropicに$100課金してClaude Codeを使い始めてみたんだけど、めちゃくちゃすごいね 初見で僕の未発表のコードをよく理解してるし、僕が苦戦していた謎のクラッシュバグも治してくれたし、頼んだ機能や最適化をどんどん実装してくれるし、人間を超えている 自分でコードを書くのではなくて、何個かClaudeを並列動かして指示出し→結果チェックが仕事みたいに急になってしまった。プログラマをやっていたつもりだけどいきなり管理職になった気分

Asked Claude Code silly questions 'What's the code review style of XXX?'. Still not sure how well it actually performs reviews in someone's style, but the responses are fascinating. gist.github.com/MaskRay/9ebed6…

claudes-c-compiler禁用了RISC-V assembler relaxation，但没有发现真实原因是没有采用fragment-based设计。 16-bit x86没有弄对看上去是训练样本太少。 我在lld/ELF和blog里提了这个不准确的术语"absolute relocation"太多次好像误导了它。 它能处理asm goto还是挺惊人的。

I hope that someone can take inspiration from Solaris and macOS for ELF maskray.me/blog/2021-05-1… > In Mac OS X, the two-level namespace introduced in 10.1 (default unless you use ld -flat_namespace) is a similar model.

When a dynamic relocation references a defined symbol, get the hash from the GNU hash table. DT_GNU_HASH only stores 31 bits (the LSB is cleared for chain termination). The missing bit can be recovered openwall.com/lists/musl/202… updated my 2022-08-21-glibc-and-dt-gnu-hash

maskray.me/blog/2026-02-0… lld 22 ELF changes highlights: Distributed ThinLTO, Pointer Field Protection, RISC-V vendor relocations

Both my browser and I forgot the IP address for thakis's llvm build bot. OK, agents are good.