nikaia check main.nika showed it on default_limits(), and I wrote it off as noise from checking one file. It actually meant the build-time program no longer compiled. That cost one full package run with --bless, about 4 minutes, which wrote a .stderr for all 18 tests and left load.stdout empty. Getting the view checker to accept resolved cost the second most: ref mut params are refused, and so is storing views from a ref Node into a map.ref mut self that take the tree by value (Node, through drain()), as examples/1brc.nika does. I found this only after NK1158 and NK2302.collections::HashMap[ref String, T], and that it needs use std::collections..clone() of a view String gives owned text. Writing n.anchor?.clone() turned Node.anchor owned all through the grammar (NK1106 at lines 522 and 549).nikaia check on the file (fast, caught the view errors); nikaia run on a scratch copy with the defaults taken out, which showed the real rustc error; two small probes that narrowed the defect to a local bound from m.get; one scratch package run that verified the test output.nikaia test --bless on the real package, run while NK1127 was still unexplained, plus the cleanup of the .stderr files it wrote.every_nika_file test and pre-push-check, since nothing was committed.= f()) after your change means rustc failed on the lowered file.nikaia run on a scratch copy with the defaults taken out.--bless before nikaia check is clean, because a failed build blesses a .stderr for every test.Also add a line to the brief that the resolver keeps its state in a struct with ref mut self methods and takes the parsed tree by value.
I also created the local branch wip/y27-load, which the brief did not ask for, so the work is not lost. The owner may want to know that a branch exists.