Conversation
I must have accidentally added it at some point. Remove it, because it is inconsistent with the rest of the file.
Context.__eq__ was based on the string representation (str-rep) of the
context. This makes sense for the memoizer, which should consider two
contexts the same when they have the same str-rep, because it imitates H
and H only ever sees the str-rep of contexts.
However, this str-rep-based equality throws off other algorithms. See
the test case that I added to test_basic. There we have two contexts
with the same str-rep but different workspaces. Call them C1 and C2.
sched.resolve_action adds both of them to sched.pending_contexts, such
that:
sched.pending_contexts = deque([C2, C1])
After the Reply("$a1 $a2"), sched.choose_context determines:
choice = C1
Then it does sched.pending_contexts.remove(C1). However,
str(C1) == str(C2) → C1 == C2,
so what gets removed from sched.pending_contexts is C2; C1 remains. This
is wrong and leads to failure shortly after.
Fix this by making Context.__eq__ based on a context's workspace_link,
unlocked_locations and parent. I chose these three attributes, because
every context can be constructed from them.
This should work in the general case, but it breaks the memoizer. So
change the memoizer to explicitly stringify contexts before doing
anything with them.
Implemenation notes:
- Change Context.__eq__ to return NotImplemented when `other` is not a
Context, because that's what the Python docs recommend:
https://docs.python.org/3/reference/datamodel.html#the-standard-type-hierarchy
- Leave Context.__hash__ as it is, because hash(c1) == hash(c2) is
allowed to happen occasionally. Also, `hash` has to be fast, so it
makes sense to use the str-rep, which is already stored in the
context.
- The `str` calls inside Memoizer are repetitive, but I think it would
be overkill to implement an abstraction.
For exercising Patchwork with random inputs. Implementation notes: - I don't know if the code involving threading is correct. - Leave many TODOs, because doing them would good, but isn't crucial for the random test to be useful. - Some identifiers have a trailing underscore (hypothesis_) in order to avoid shadowing global variables. - One might think that starting a thread for every act() is slow, but on my machine, which is old, it increases the running time of the random test by only a few seconds (7 s → 8-12 s). - Someone in https://stackoverflow.com/questions/2829329/ advises to catch BaseException (instead of Exception) and re-raise in (Terminable)Thread.join. Don't do that, because a SystemExit in a child thread shouldn't terminate the parent, in my opinion. - Custom exception types are good. Cf. https://www.youtube.com/watch?v=wf-BqAjZb8M.
Member
|
Make a new PR that only contains the latest commit? Not sure what went wrong, but now that the other PR is merged into master, the diff should only show the new changes. |
Contributor
Author
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
2 participants
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.
For exercising Patchwork with random inputs.
Note that this branch is based on fix-context-eq, which is still a PR.