Read schema 4 of the corpus - #32
Merged
Merged
Conversation
The corpus moved to schema 4 in the engine and this client was two schemas behind it, so the pin bump brings 1268 cases across a reader that knew about none of what the last two added. Three things arrive together because they arrive in the same files. Connections. A case may name the connection it runs on, and a setup statement may name its own, so that a case about a transaction can say what the other connection sees. A named connection is made the first time a case names one, by duplicating the connection the case opened with, because the two then share the write side and that is what a pool hands out. Opening the path twice would be two databases that happen to be the same file, which is a different question and not the one a case about a transaction asks. Graph values. A node is its table and the row it is, an edge is its table and the two rows it runs between, and a path is a sequence that alternates the two and holds an odd number of them. The engine's edge carries the position it holds among the edges out of its source, which the loader chose and no case writes down, so a value is normalised into the corpus's own shape before it is compared rather than compared against a number nobody picked. The export. A case may say what the result gives through Arrow, field by field and into the nested types, or say that Arrow refuses it. That is read off the C Data Interface with ctypes rather than through pyarrow, for two reasons. pyarrow is not installed to run the corpus, so a check that needed it would be a check that quietly does not run. And the C runner reads the interface directly too, so the two runners are looking at the same bytes and reporting them in the same words, which is the whole reason a case writes a format string down. The corpus run is a session fixture now. Every case gets a database of its own and there are over a thousand, so two tests reading one report were four gigabytes and three pytest generations of it, which filled a disk. One run, in a directory that goes when the run does. 1268 cases, 1262 passed, 0 failed, 6 unsupported. All six are a time written to the nanosecond, which is a digit finer than a Python datetime holds, and each one says so.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The corpus moved to schema 4 in the engine and this client was two schemas behind it. The pin moves to 6ee7d80 and the reader learns everything the last two schemas added, which is three things arriving together because they arrive in the same files.
Connections. A case may name the connection it runs on, and a setup statement may name its own, so that a case about a transaction can say what the other connection sees. A named connection is made the first time a case names one, by duplicating the connection the case opened with, because the two then share the write side and that is what a pool hands out. Opening the path twice would be two databases that happen to be the same file, which is a different question and not the one a case about a transaction asks.
Graph values. A node is its table and the row it is, an edge is its table and the two rows it runs between, and a path is a sequence that alternates the two and holds an odd number of them. The engine's edge carries the position it holds among the edges out of its source, which the loader chose and no case writes down, so a value is normalised into the corpus's own shape before it is compared rather than compared against a number nobody picked.
The export. A case may say what the result gives through Arrow, field by field and into the nested types, or say that Arrow refuses it. That is read off the C Data Interface with ctypes rather than through pyarrow, for two reasons. pyarrow is not installed to run the corpus, so a check that needed it would be a check that quietly does not run. And the C runner reads the interface directly too, so the two runners are looking at the same bytes and reporting them in the same words, which is the whole reason a case writes a format string down. A refusal reaches this client as a TypeError for a column Arrow has no type for and a ValueError for one that will not fit, and both are caught around the call that raises them and nowhere wider, so a TypeError out of the ctypes reader is still a bug rather than a refusal.
Every refusal wording in the reader was checked against
crates/zu-corpus/srcrather than guessed at, and four of them differed from what I first wrote. The reader was right and the tests were wrong.The corpus run is a session fixture now. Every case gets a database of its own and there are over a thousand, so two tests each running the whole corpus into a
tmp_pathwas four gigabytes a run and three pytest generations of it, which filled a disk here. One run, under a directory that goes when the run does.1268 cases, 1262 passed, 0 failed, 6 unsupported. All six are a time written to the nanosecond, which is a digit finer than a Python datetime holds, and each one says which value and why. 785 tests pass, ruff is clean.