when I first wrote pred_recdec I did the oopsie of making the error API just a single string with no raw data parts, and it turns out that doesn't just cause problems for using the API, but also how much stack space is used by the recursive descent function for some reason?

so I've been going in and changing how errors are made (in addition to adding machine-readable bits) to save stack space and the patterns I'm using to trick LLVM are weird, like "function + immediately invoked lambda" (????)

I just released my parsing framework for rust. It lets you write parsers that are as powerful as handwritten recursive descent, but in a much simpler, lower-boilerplate BNF format. I wrote a working C parser in it in a matter of days (instead of months).

https://wareya.wordpress.com/2026/02/01/beta-release-predicated-recursive-descent-for-rust/

@Krita @krita_artists

bug report (i can't register on the bugtracker)

when loading .ora files, krita 5.2.9 and 5.3 (prealpha) disagree with the OpenRaster spec and always load groups (stacks) in Pass Through mode, ignoring the "isolation" attribute. also, the default is supposed to be non-pass-through. i suspect that this bug is somehow caused by the fact that mypaint only supports (supported?) passthrough mode for groups, and ora is mypaint's main export format. but ora is a specced format.

opened an issue about a real C standard noncompliance bug in MSVC (it uses the wrong type for large int literals in C99 mode) and got a rejection where they repeat a falsehood about the standard 👍

https://developercommunity.visualstudio.com/t/Standards-compliant-C99-mode-uses-wrong/10789349