the first test succeeds, but i have early typechecking in the parser that needs to be put into a second pass.

too late for that today, more tomorrow.

#devlog #LRDL

if you're getting an aneurysm trying to understand how those mtof cases work, that's because you're trying to read it as a function, but the "function head" is the `then` section, at the end of the block rather than the beginning.

in datalog/prolog, order would be correct:

mtof(m, ...) :- ge(m, 0), ...

but then regular rules are not in execution order.

i'm going to explain the relational model with diagrams in an extra post.

#devlog #LRDL

alright! looks like we're sufficiently complete now to have more refactoring fun on the user side.

this contrived example here defines disjunctive rules for our inline.

they inline as merge cases; types are inferred correctly.

#devlog #LRDL

sigh. i have one more thing to fix, which is sugar for recognizing expressions as conditions; e.g. when you put `B ...` in a case, or any other buffer B with an omitted boolean in last place, e.g. `eq a b`, this should internally expand to `B ... tmp:bool; cond tmp`

#devlog #LRDL

yes, hooray, it works again!

which means inlines should now theoretically work perfectly. i can make more music

#devlog #LRDL

only just managed to get the parser to work again. now the typechecking begins.

#devlog #LRDL

this turned into a bigger refactor. the parser needs to work without knowing anything about types, and that's presently not true. it's not impossible to change, but - requires a lot of furniture rearrangement.

#devlog #LRDL

the first test succeeds, but i have early typechecking in the parser that needs to be put into a second pass.

too late for that today, more tomorrow.

#devlog #LRDL

so i'm thinking, step 1 is adding "inline" buffers (a solution from soufflé), for which cases don't generate code, and which in usage get inlined directly into other cases.

this already covers the need i have for functional macros and i can put off supporting C exports to a later point.

#devlog #LRDL

alright. today's job: support for defining C functions in LRDL.

this should generalize far enough that we can implement `main` this way, and also write shared libraries with it.

but it requires opening the reactor pattern to external users. it significantly complicates interactions and update flow.

hmhmhm. difficult to extract the smallest most elegant idea that works.

#devlog #LRDL