#debugging — Public Fediverse posts

A closer look at XAML Incremental Hot Reload in .NET MAUI 11

https://tonyedwardspz.co.uk/blog/maui-11-xaml-hot-reload/

#dotnetmaui #xaml #visualstudio #dotnet #appdev #debugging #uidesign

AI makes you faster at writing code. But does it make you worse at debugging it?

An Anthropic study tested developers learning a new library:
• AI-assisted coders: 50%
• Coding by hand: 67%

The hardest hit? Debugging.

Turns out, they didn't trade speed for learning - they traded learning for nothing.

No trade-off. Just loss.

🔗 Watch the full video (transcript included) – https://www.infoq.com/presentations/ai-software-engineering-pipeline/

#AI #SoftwareEngineering #Debugging #AICoding #DeveloperProductivity #Anthropic #InfoQ

The iOS Crash We Could Not Reproduce

Key Takeaways
Intermittent production crashes often need controlled fault isolation when logs and local testing cannot identify…
#NewsBeep #News #Technology #CA #Canada #Concurrency #Debugging #Development #iOS #ioscrashsigbusdebuggingexperiment #Mobile #operatingsystems #Testing
https://www.newsbeep.com/ca/966510/

Making Progress, Using Development Edits

I’m an author. and one of the hardest things is to see what you’ve done wrong. If you have a spare human to edit you, it’s easier.

What About an LLM?

Well, yes. Advait Sarkar has a lovely article on that in the October 2024 Communications of the Association For Computing Machinery. called AI Should Challenge, Not Obey.

However, if I just ask for a dev edit, I’m really asking it to do two things:

• find out what a dev edit is, and also how to apply it, and
• have it score your work against that standard.

Of course, you don’t get to see what standard the LLM is using if you do that.

Recurse!

Instead of trying to do two things at once, drill down. Do one of them, inspect and debug it, and only then use it to do the second.

In the case I was interested in, I asked Claude what a dev edit was. It gave me a recipe on how to do it as a human, but it contained a lot of sucking up. When I ran it against a particular story, the response was so saccharin that I wanted to spit.

So I debugged it, and found it needed to be told things like “Start with what the part does well, because the author needs to know what to preserve. Be careful not to praise: being mentioned in this section is enough. Praise will make you sound insincere, and therefore less trustworthy, defeating the whole purpose of this review.”

And ‘Avoid vague feedback. “This section feels slow” is a poor start, but “This section covers ground the reader already has from the previous part — consider cutting the first three paragraphs and opening instead on the moment the character enters the room” is useful.’

When it stopped sucking up like Evodius[1] I ran it on some new work, and it went well. As of now, the prompt is in https://codeberg.org/davecb/Prompts/src//src/branch/main/developmental_edit.txt

Some caveats: if you try to use it twice on the same file, it will insist on finding more errors, even if they’re not errors (:-)) Keep your work in a single chat, feed it your changes and new drafts, and ensure its tiny brain has the context it needs to not start hallucinating.

Magic Trick

The trick is to make it write a prompt about how to do something, and then tell it to do that. In many cases, you can check it’s prompt against an article on how a human does that thing, and you can always debug and test it before you use it in produation.

–dave
[1. Evodius is the suck-up student in Augustine’s “On the Free Choice of the Will”, a dialogue on theology. I hated reading him blather on, in a philosophy course I took]