pieter

I was willing to gloss over the AI usage because I really like the Bun DX. I used Bun in Shell scripts and small prototypes before and decided to use it for something bigger.

But then in a week or so of use, tons of issues appeared. Use-after-free bugs when a client disconnects (both in Zig and the Rust port, funnily enough) which will kill Bun. Unimplemented feature which just return “0” always. Many backpressure bugs which means something like “return new Response(fs.createReadStream(bigFile))” will greedily read and OOM your box.

All of these have been reported, robobun has made fixes of various quality, but none of this has been addressed. Some of them have been reported for months.

I’d prefer Bun to work on stability rather than fancy features like image processing or threads.

Any details on what changed in the -002 models?

I’d really like libraries to put the language they’re for in the description.

Especially anything AI-related seems to just assume everybody uses Python.

What I would love is something that can give alternative tube sizes.

My bike uses 22x3 tubes, which are unusual and hard to come by. However, there have been reports of 20x4 bike tubes or 18x3 motorcycle tubes also fitting.

A website that recommends more widely available or cheaper “this will also fit” tubes would be amazing.

Anyone know how this compares to Framer?

Not being able to transfer BTC out of MtGox

I doubt domain length will really make any difference, gzip compression should take care of a longer domain name.

All these faster speeds won't do anything really for heavily populated areas. The most important thing is total available bandwidth (bytes/sec/hz), and it's unclear how better that will get. LTE itself doesn't improve a lot over 3G techniques, the thing that's making LTE faster in saturated conditions right now is that there's just more bandwidth compared to 3G.

Off shore wind power is about 25% of the wind power capacity of Denmark, the rest are 'normal' wind turbines. However, they'll contribute more to the total energy output. Off-shore wind turbines generate about 35-50% of their maximum capacity on average, compared to 20-30% of normal wind turbines.

Biggest thing I'm missing in the FF Aurora builds is a console in the Inspector / Debugger panes. Having to switch to the console pane all the time gets annoying quickly.

This still doesn't explain why the MRI implementation is accidentally threadsafe. Why doesn't the interpreter switch threads after reading the value from the hash but before storing the updated value?