Today work has focused on three things. The CI matters have been resolved. New matters concerned with the GNU/Hurd have been researched. Work on the Idris proof has continued.
Yesterday’s CI issues are solved. The patch failure was not reproducible. A re-run did the trick. The documentation command patch was submitted as a separate PR. The uClibc PR got feedback from a maintainer.
They preferred to close the PR. The target maintainer did not have immediate need for LFS. The proposed solution was to keep it as a draft PR. Maybe somebody in the future would need it. But the maintainer thinks otherwise. It’s closed.
The time_t tracking issue also received comments. Last a comment there provided a brief overview of the GNU/Hurd’s support. The maintainer mentioned the current bindings were probably not enough.
This was researched. The details can be found in the tracking issue. A brief summary follows. The GNU/Hurd has no support for 64-bit overrides. Its default time_t bitwidth is target-dependent. So time_t matches machine word size.
This means there is no need to change the Rust bindings. The current ones reproduce that exact behavior. Glibc also has symbols name redirections. This only applies to Linux targets. That means no work is needed for the GNU/Hurd.
The Idris proof has progressed nicely. The uninhabitedness proposition was completed. This indicates the non-negative proposition cannot have positive indexes. But the way there needed simplification. The initial idea changed.
Yesterday’s attempts were at a general proposition. The Idris type checker needs a full type in interface propositions. That requires proof-search in inductive types. The proof space is infinite for the general case.
The structure alone of the non-negative’s index is insufficient. There is no proof for the non-negatives that uses a predecessor view. That implies data at the type level is not matched. The type resulting from it is matched.
The data construction here refers to another type. This other type is implicit. It is also inductive. It is codata over the infinite number line. This means big trouble for totality. It also means the type checker will not halt.
But it halts. Only it halts with an error. That took a while to sink in. The uninhabitedness proposition has to be over the inductive hypothesis. That means the first negative. That means the predecessor of zero.
The inductive hypothesis here builds on a one-step induction. The inductive hypothesis of the non-negatives is zero. The one-step negative induction is thus impossible. No proof can be constructed for such a non-negative.
This meant a lack of generality in uninhabitedness. That means a need for contraposition lemmas. But then a new idea came to mind. The non-negative proposition does not capture all non-negatives.
The view over primitive integers was misunderstood. It is unlike Peano numbers. It is a sum. One starts with zero. That happens to be the inductive hypothesis. Then the predecessors are sums of negatives. The successors are sums.
The non-negative proposition was conceived wrongly. It assumed non-negatives only exist when built without predecessors. But a sum containing some negative term can also yield a non-negative.
Such a sum needs fewer or an equal number of negatives as positives. The analogy translates well to the coproduct view. A given non-negative can be proofed with multiple topologies. It can be solely the inductive hypothesis (zero.)
It can also add some number of successors. But it can also add some number of predecessors. The number of predecessors must be smaller than or equal to the successors. That is the essence of a non-negative.
So work started on a new proposition. This proposition only counts the predecessors and successors. This is fairly simple to proof. The tricky part is the compliemntary proposition.
It assumes an integer. Then it implies the above proposition. But it does so with an ordering proof obligation. The successor counter ought be greater than or equalto the predecessor counter.
None.
Work on the uClibc PR is officially done. The GNU/Hurd research seems solid. An answer is awaited. The plan is to answer to the command fix PR next week. That should happen on Thursday. The other days will be dedicated to ctest.
The Idris proof work will stop on Sunday. It is engaging. But the Idris community tutorial must be finished. Two more weeks will be given.