Today work consisted of two things. A bit of a third thing also had to be done. The first thing was the ctest extension. The next thing was the Idris community tutorial. The latter continues from past weeks’ efforts.
Last week an issue had been found in the ctest algorithm. A recap follows. This algorithm is used to resolve reexports. It is part of the extension PR. It is needed to consider all paths an item can be referred to by.
This is then used in the skips. These are small utility functions. They are exposed to ctest consumers. They skip testing certain symbols. But those symbols could exist in paths other than their source path.
Last week the first revision had been finished. This implementation was being tested. The item container can end up with duplicate items. These come from “different” reexports. But these reexports refer to the same items.
This always holds for non-malformed input. This stage in ctest always gets correct input. So a deduplication proposition was proofed. This was finished today. Only the proposition had been elaborated last Tuesday.
This was then tested. Another bug was then found. This time it was unrelated to any ctest extension patch. It is a bug that currently lives in upstream. syn Visitor hooks run on all parsed items. This is sometimes a problem.
One such case are foreign blocks. These can exist within a function body. They can also exist within a module. But only the latter can be tested. This is because only the latter have a “global” path.
Nobody seems to have reported this issue. It is an odd usecase. It would pop up if some downstream consumer had function-level extern blocks. This has a simple fix. It is the same fix as used for imports. That was part of the extension.
Parsed reexports can not always use the syn Visitor hooks. This is because of the same reason as above. Function-level use statements should not be considered for resolution. The solution for other items is the same.
Idris work has started with the community tutorial. The first few chapters have been finished. Extra time has been spent on the problems. This is to satisfy stricter requirements than those in the statement.
Most of this has revolved around point-free style proofs. This is otherwise known as tacit code. Duplication of parameters seems impossible. Tacit seems to work only when no duplicate parameters are expected.
None.
The plan for the week follows as expected. The first three days will go to ctest. More tests have been run. They seem to prove correctness of the implementation. A write up to update the PR status is in the works.
The Idris work will continue. There is not much to comment here. Though there is one thing to keep in mind. One must avoid the temptation of proofs of correctness. They are one huge timesink. They will have their time and place.