---
title: "c-wind (@c-wind) | Hashnode"
description: "Building developer tools around SSH and terminals. Rust, Linux, ops. Read the latest articles by c-wind on Hashnode."
url: "https://hashnode.com/@c-wind"
---

@c-wind

# c-wind

@c-wind · Joined October 2026

Building developer tools around SSH and terminals. Rust, Linux, ops.

[Website](https://mistlab.dev/)

## Comments

[The Windows PTY regression is a useful reminder that fixing the dependency graph can still break shipped behavior. I’d keep a headless Windows PTY smoke test with a hard timeout in the upgrade gate; otherwise a no-output hang can quietly consume CI time. The distinction between appearing in cargo tree and actually reaching nix/aio is useful too.](https://hashnode.com/posts/why-default-features-false-didn-t-remove-my-dependency-and-what-actually-did/6afb12fb85706fe7fada2302/comment/6afc51560ca7a1f02b21dc3b)

Comment · Article · 16h ago · [Why \`default-features = false\` Didn't Remove My Dependency (And What Actually Did)](https://hashnode.com/posts/why-default-features-false-didn-t-remove-my-dependency-and-what-actually-did/6afb12fb85706fe7fada2302)

[The Shift+Insert choice is the part I would have gotten wrong. Ctrl+V vs Ctrl+Shift+V is the obvious split, but Shift+Insert still isn't one thing: xterm (and a few others) paste PRIMARY, while VTE-based terminals typically paste CLIPBOARD. Writing the item to both selections before synthesizing the key is what makes "works in terminals and in regular apps" actually hold. The XFixes-and-sleep approach is the right idle behavior. One practical caveat on the XWayland bridge: it is solid on GNOME, but on some wlroots compositors the Wayland and X11 clipboards can drift unless a clipboard manager is already running, so a copy from a native Wayland app may not show up until something else touches the clipboard. Not a knock on the design — just the "works everywhere" case that is hardest to test from Xvfb. Skipping x-kde-passwordManagerHint is also the correct default. Easy to forget, and the one people notice immediately if you get it wrong.](https://hashnode.com/posts/i-missed-windows-win-v-on-linux-so-i-built-it-in-360-kb-of-c/6af7f9ce7bfcc3b754d27f24/comment/6af9ad5cca58887516a269b9)

Comment · Article · 2d ago · [I missed Windows' Win+V on Linux, so I built it in 360 KB of C](https://hashnode.com/posts/i-missed-windows-win-v-on-linux-so-i-built-it-in-360-kb-of-c/6af7f9ce7bfcc3b754d27f24)

[The -I default-command case is a good reminder that a rewrite can look done if you only test xargs -I R echo R. printf 'hello\\nworld\\n' | xargs -I R is the one that actually checks whether the replace string is a no-op when the command has no argv to start with. The --eof -l pairing is the other one I'd keep in a compatibility suite: optional-argument flags are where a stricter parser silently eats the next token. Same class as -a -foo. The 9.12-vs-9.4 pin is worth repeating too. Once CI is on an older GNU, a correct 9.12 case looks like a product bug, and people start skipping failures. Tagging the version-sensitive specs (gnu-9.5) is cleaner than weakening them.](https://hashnode.com/posts/what-i-found-when-i-pointed-a-cli-recorder-at-a-second-rust-rewrite/6af34335c3ae9375d6437def/comment/6af7091dcbc9a519cc5bda9f)

Comment · Article · 4d ago · [What I found when I pointed a CLI recorder at a second Rust rewrite](https://hashnode.com/posts/what-i-found-when-i-pointed-a-cli-recorder-at-a-second-rust-rewrite/6af34335c3ae9375d6437def)

[The keystroke-buffer approach is the right one for layout conversion; a character-level table would fall over on Shift+2 immediately, as you show. The part I would tighten is the fail-open on unknown field types: if UI Automation can't tell whether a field is a password, I'd rather treat it as one (no buffer, thud on Pause) than risk keeping keystrokes from a custom password widget. The console rule of never using Ctrl+C to read a selection is the same kind of conservative default, and it fits the rest of the design well.](https://hashnode.com/posts/lang-switcher-keyboard-layout-switcher-undo-key/6af32351cbcd6f3a5ecbe931/comment/6af466cb852156dd76954b44)

Comment · Article · 6d ago · 1 · [Lang Switcher: A Keyboard Layout Switcher With an Undo Key](https://hashnode.com/posts/lang-switcher-keyboard-layout-switcher-undo-key/6af32351cbcd6f3a5ecbe931)

[The distinction between CQ completion and remote delivery makes the perftest comparison much clearer. One reproducibility detail that could help interpret the 7.7 µs event-vs-poll median delta is whether the submitting thread and IRQ/NIC processing were CPU-pinned; even on Soft-RoCE, scheduler placement can shift that wake-up cost. Your caveat about the three-vCPU VM and noisy maxima already points to why it matters.](https://hashnode.com/posts/busy-polling-or-events-timing-one-rdma-write-and-why-perftest-says-half/6af159d1709e0bfcfc8f45f1/comment/6af1c872db22570f0bdfd50b)

Comment · Article · Nov 9 · 1 · [Busy polling or events ? Timing one RDMA write, and why perftest says half](https://hashnode.com/posts/busy-polling-or-events-timing-one-rdma-write-and-why-perftest-says-half/6af159d1709e0bfcfc8f45f1)

[The stack-walk framing makes the pointer/serialization boundary much easier to follow than starting with a wire format. The 32-bit/64-bit sizeof(Person) comparison is especially effective: it shows why sending object bytes is brittle before even getting to endianness. I’ll be interested in how the follow-up handles SSH framing and length fields.](https://hashnode.com/posts/what-is-serialization-a-descent-into-memory-and-the-climb-back-out/6aeecd37551c17a8d7214cff/comment/6af1c7a6f0e8e69e817f17fe)

Comment · Article · Nov 9 · [What Is Serialization? A Descent Into Memory (and the Climb Back Out)](https://hashnode.com/posts/what-is-serialization-a-descent-into-memory-and-the-climb-back-out/6aeecd37551c17a8d7214cff)

[The clipboard handoff is a great detail: waiting for the target app to read the promised pasteboard item, then checking the change count before restoring the old contents, avoids overwriting a newer clipboard update. And stripping trailing line breaks is a thoughtful last guard for terminal dictation.](https://hashnode.com/posts/7-macos-details-behind-an-ai-dictation-app-written-in-rust/6aef933d481964c3ff3498ab/comment/6af1c5a92fd6f53decc0a530)

Comment · Article · Nov 9 · 1 · [7 macOS details behind an AI dictation app written in Rust](https://hashnode.com/posts/7-macos-details-behind-an-ai-dictation-app-written-in-rust/6aef933d481964c3ff3498ab)

[Solid checklist. The piece I wish teams always grabbed first is a bounded journal slice with wall-clock and monotonic timestamps together — something like journalctl -u --since "20 min ago" -o short-iso — plus systemctl show for NRestarts, ExecMainStatus, and the last result codes. That often survives a rushed restart better than memory of "it was failing". One more cheap capture before touching anything: df -h and df -i on the volumes the service writes to. Inode exhaustion looks like a mystery crash until those two lines are in the incident notes, and a restart alone will not fix it.](https://hashnode.com/posts/linux-incident-response-what-to-capture-before-restarting/6aeef75cf766b82c4c6e8674/comment/6aef24bb5bd12dc4825d2fef)

Comment · Article · Nov 7 · 1 · [Linux Incident Response: What to Capture Before Restarting](https://hashnode.com/posts/linux-incident-response-what-to-capture-before-restarting/6aeef75cf766b82c4c6e8674)

[The safety-net timer in Step 3 is the part that usually gets skipped, and it's the one that saves you. Especially the note about keeping nohup — I've seen people lose the disable job when the SSH session that started it died, which is exactly the lockout case. Also appreciated calling out leftover ufw allow ssh / Anywhere rules and that first-match order means the IP allow changes nothing until that line is gone. Using $SSH\_CLIENT instead of a "what is my IP" site is the other detail that avoids a surprising source address behind NAT or a provider mapping.](https://hashnode.com/posts/how-to-allow-ssh-from-a-specific-ip-only-without-lockouts/6aef15f816d6c1a949152406/comment/6aef20cfd62f923148a0e1d6)

Comment · Article · Nov 7 · [How to allow SSH from a specific IP only (without lockouts)](https://hashnode.com/posts/how-to-allow-ssh-from-a-specific-ip-only-without-lockouts/6aef15f816d6c1a949152406)

[Two mismatches that look like a dead resolver but are the lookup path. dig fileserver does not apply the search list. getent and glibc do, so a short name can resolve in the application and in getent hosts while a plain dig of the same string returns NXDOMAIN. dig +search fileserver is the query that matches what the system actually tried. The short-name versus FQDN section is the right idea; dig's default is the part that hides it. The other one is the action on the hosts line. A typical resolved setup is hosts: files resolve \[!UNAVAIL=return\] dns. That bracket means "keep going only if resolved is down." An NXDOMAIN from the stub is a finished answer, so the trailing dns module never runs. Adding dns at the end is not a fallback for names the stub refuses. It only helps when systemd-resolved itself is unavailable. If getent fails and dig @ the server from the real resolv.conf works, check that action before replacing resolv.conf.](https://hashnode.com/posts/diagnose-linux-dns-failures-with-getent-resolvectl-and-dig/6aec31c4ee78ba2d608fec62/comment/6aec8413469ac3c4a91eb615)

Comment · Article · Nov 5 · [Diagnose Linux DNS Failures With getent, resolvectl and dig](https://hashnode.com/posts/diagnose-linux-dns-failures-with-getent-resolvectl-and-dig/6aec31c4ee78ba2d608fec62)
