Started off on a ringbuffer-based ABI stable asyncread / asyncwrite pair because the only existing solution seems abandoned with open issues and not much activity.

Over the weekend I'll extract it into a crate and write a heap of tests. Async code is a bitch to test, but it's probably the least worst in Rust where the testing scaffold can manually step all of the async codepaths.

https://git.lbfalvy.com/Orchid/...

Besides keeping your computer unlocked for long-running tasks, Caffeine can also be used to fool MS Teams' activity monitor.

Delighted to see that this site is still alive.
I''d like to hear your opinions about my take on libre licences:
https://git.lbfalvy.com/Orchid/...

I know I have a problem with asking for help. I'm aware it's a problem, I want to solve it, and I'm trying, but this is easier said than done.

In my defense however, the issues I'd need to ask for help with are completely absurd. We have a shared Feature environment with a shared database. A push to any feature branch auto-applies migrations to this database, so it's full of broken script output. Tests are supposed to use this database. We do not have full rights to edit this database so we can't try and fix the issues. Instead, the database is reset from production once a week, discarding all changes including anything we deliberately put there for testing. I asked who broke the database and if they could fix it please, somebody responded with freeform text roughly describing the fixes _I should apply to fix HIS TEAM'S mess_, which didn't include any technical identifiers and referred to tables and columns exclusively via vague approximate names.

He then posted a screenshot of an e-mail from about a month ago in which HE complained to MY team lead about how "some people" keep breaking the database, which contained no examples and no suggestions, but was sent immediately after the first time this year that we actually properly broke the shared database. By that point they were past their 10th broken migration that warranted an early restore.

I bet VSCode only added a central "disable all AI features" checkbox so that they can more easily ignore my preference and forcibly uncheck it on every update. If I had to go around hacking the editor to hide buttons and disable triggers, it would be a lot more work for them to break all those hacks.

"If you get stuck with something for 20 minutes, ask" I would rather chew glass until I manifest the solution.

I have no problem with asking for help every once in a while, but if I don't get to spend time digging into questions I might as well become a bricklayer.

I came up with a really intuitive way to create a coroutine in Rust by passing the sender of an mpsc queue to a callback, then merging the receiver of the same mpsc with the future returned by that callback, but I must've cocked up somewhere because I'm pretty sure it leaks memory.

MPSC ports don't own each other in either direction, that was my first guess too.

Boy I sure hope Rust / LLVM's devirtualization pass is as good as its infamy suggests!

https://devrant.molodetz.nl/preview...

So it turns out `futures` is a fucking liar and senders don't get an extra buffer slot in the sense that an additional message could be kept off-stack while the sender continues. The "extra buffer slot" they are talking about is the message resting inside the blocked Send future, which is to say, the fact that the sender can "block" is presented as a means of storage. Why does this community feel the need to complicate things? The analogy with the stack was perfectly fine. You wouldn't call a rendezvous channel a 1-item buffer just because the item actually has to be on the stack before it is sent and will stay there if the sender needs to wait.

Oh dang, now that the individual components of Orchid are nearly complete, I should probably improve this ratio.

HTTP/3 requires TLS. Wouldn't this make it drastically less economical to run a reverse-proxy? Are 4 unnecessary cryptographic operations per request not significant? What about the bandwidth overhead of rounding up the length of responses like `HTTP 200 OK\r\nContent-Length: 2\r\n\r\n{}` which make up a majority of all traffic to the nearest encrypted block?

WASM threads have been supported in every browser since 2021, but Rust can't support them because production users and library authors enabled WASM support in thread-unsafe ways relying on WASM being single threaded, so enabling it would almost certainly introduce vulnerabilities.

An endpoint called TryFailedCopyDocuments which is called unconditionally in the happy path, calls into a service to update session data, then separately checks whether the service is live which could only be false if the service died shortly after the previous call, then if it was live, does a separate blocking call into the service to copy some files, and if it wasn't, copies the files within the request handler thread.

In case anyone else has the bug in the web version of DevRant where if you scroll to the comments the page grows infinitely, I made a userstyle

https://userstyles.world/style/...

I actually think that auto-completing pull requests with any merge strategy other than fast-forward is completely insane. WTF do you mean the final version of the code is produced with no humans present?

I FINALLY GOT A FUCKING MAIL SERVER GOING

DMARC, SPF, MX, WHATEVER

My zonefile is full of crap generated by Stalwart Mail that I don't understand but I DON"T CARE. I've been trying to do this since high school. I sank WEEKS into it. I can finally die in peace.

Also, I'm now able to self-host shit like GitLab that relies on a mailer for correct operation.

God. FUCKING Damn.

Naturally, now that it's done I've no idea what took so long, and the only problems I appreciate are the ones I solved in the last weeks.

Async Rust

The whole is not more than every property of all of its parts, wherein the parts' properties include their pairwise and setwise interactions. A whole is nothing more than an arbitrary grouping of interactions that mark their interacting parties as parts of the whole and aggregate other properties of those parts into properties (including interactions) of the whole.

I'm not managing Orchid in terms of performance milestones so these posts and their order are probably incredibly bizarre.

Anyway; demo for lexer plugins, foreign code provided constant resolution, foreign function dispatch, and clean shutdown.

How is MSSQL so popular? "Oh yeah, sometimes you have to use dynamic SQL because the compiler doesn't correctly detect the impact of DDL commands and rejects working code" god fucking what? Why does the compiler infer the future structure of tables after a DDL command if it can't do so reliably? In my world, if it becomes public knowledge that a compiler incorrectly infers something, the maintainer scrambles to remove the inference instead.

I also find it surreal that I have to disengage versioning on a table in order to be able to modify the version history. Like surely, there should be a mechanism to temporarily skip the hooks other than ALTERing the table into a regular unversioned database table and then ALTERing it back, rebuilding the versioning infrastructure from scratch. SQL is awful and MSSQL is doubly awful.

3rd PR in recent memory whose test involves visiting every view of this complicated engineering design software. We're not doing refactors right.

You know how LLMs always imitate expertise and understanding? That isn't specific to English, it happens in code as well. It's harder for them to get away with it because even the best guess isn't likely to be approximately correct, but they still try and it still sometimes works.

One of cucumber scenario tests now contains a deadlock somehow. We run them one at a time. I need some rakia.

IIS, aka Apache But Worse, with the world's shittiest plugin manager, a graphical XML config editor that only shows the attributes of one node at a time, invalid configuration errors that don't tell you anything about which part of the configuration is invalid, and a custom GC config optimized for hosted .NET apps that's somehow even less suitable for hosting than the default desktop GC config.

I had to enable 3 categories of Windows features and 20 additional features that weren't included in the default set for those categories just to host a WCF service.

Your tasks are going slowly, let's have a meeting about it

What are the opinions on Rust's labelled blocks and break 'label control flow primitive?

my scrum master keeps changing the format of retro and it works worse every time. We're at the 2.5 hour mark right now. It is aggressively hot and humid in the office. Everyone is asleep.

I don't think there's a way to explain to him that the purpose of a retro is not to magically fix things within the format but to financially commit to the preference for conversation, from which the hidden work of resolving conflicts can be billed.

Probably my least favourite thing about Microsoft's pretend Git suppoort is how Visual Studio doesn't show reasonable merges for csproj files when Git's own annotations in those files are almost always optimal, and they could literally just parse out those diff markers.
Instead they have a custom dummy diff engine that marks the entire contents of both files as conflicting.

Or they could do the sensible, ideal thing and diff the XML DOM, but that may set the bar too high.

I have PTO and time off from overtime which can temporarily go negative.and if it's negative at end of year the rest becomes unpaid time off. Due to special tax fuckery in my country, unpaid time off actually isn't objectively a thing, instead, my company pays tax based on my nominal salary and I have to pay them back with tax

To get out of this, I negotiated a rather special position where I'm registered for 6 hour days for a few months so that I can aggregate a bunch of overtime to be spent as extra unpaid time off that we aren't paying employment tax on for no reason when no work or payment is taking place.

This creates a rather funny loophole, in that my PTO days are worth 6 hours during these months but 8 hours for the rest of the year, whereas overtime and corresponding time off is measured in hours so it's always worth the same.

Naturally, like the strategic genius I am, I started recording all my absence in this period as PTO days without thinking because I wanted to see my time off on a nice universal counter. I realized about an hour ago that I'd been taking PTO at 75% efficiency.

Zscaler shits itself if you can reach the VPN server from multiple interfaces with different public IPs. Toy quality crap.

The water heater in my new flat is broken so I can't shower as thoroughly as I'd like, and the technician who's supposed to install broadband keeps reporting that I'm not home without checking, so I'm forced to tether from my phone. If he does it again today, I may just visit "One", the provider's offices with an IED.