Raph Levien

Apologies for swooping in, and for suggesting something that's a fairly major architectural change. I want to get my basic ideas for undo out there, and am happy to expand....

@justvanrossum Certainly. And apologies for not following up to Adrien's questions earlier; I've been slammed with other things (it'll hopefully be clear soon why), and writing code samples would take...

On closer look, I think [JSON Pointer](https://tools.ietf.org/html/rfc6901) is probably more appropriate than jsonpath, we don't need a rich query language, just a way to identify particular points in the tree....

Good to hear. This feature is primarily motivated by the needs of interactive GUI (where you might relayout a paragraph many times to deal with resize), but might possibly be...

Michael and Behdad both have the understanding of my original concern right, the ~~safe_split_mutate~~ [ETA: see below] predicate is a universal quantification, which is what we want. Actually for my...

> So you actually want stricter split point detection than what DirectWrite offers? Is that motivated by efficiency or convenience or theoretical purity? No, now that I'm rereading the issue...

Here's my analysis, which is really just an interpretation of what Sergey said. I think `safe_split_segs` is a useful predicate even in the German case. What happens is that the...

A bit of explanation: this is an attempt to make a clean, isolated test for some atomic coherence failures I've been observing. It's written in WGSL and uses WebGPU, but...

Another fascinating datapoint: if I fully enable the Vulkan memory model and change the load of the data word to `gl_SemanticsAcquire | gl_SemanticsMakeVisible`, then it passes. I believe that for...

I just pushed a new version of the test which uses atomics for both data and flag. The previous version was broken, as there was a race between nonatomic load...